An AI agent for CPG field sales should do eight distinct jobs across the route to market: recognize the shelf, guard photo integrity, match and expand outlets, answer manager questions, suggest what to sell, train the rep, read orders and invoices, and audit distributor warehouses. That checklist — not a demo-day chatbot — is the honest yardstick. Nearly half of retail and CPG companies are already using or assessing agentic AI, and most vendors demonstrate one or two of the eight. Score all of them before you shortlist anything, because the capabilities that sound least glamorous are usually the ones that decide whether your data is real.

What Should an AI Agent for CPG Actually Do?

Start from the field day, not from the technology. A rep wakes up, gets a route, drives to a store, checks the shelf, pitches products, takes the order, files the visit, and moves on. A manager wants answers, not dashboards. A distributor’s stock needs verifying. Every one of those moments is a place where an agent either acts or doesn’t — and that is the whole evaluation.

The market pressure is real. Bain’s analysis of AI agents argues that the buying process is collapsing into agent-led research, and NVIDIA’s 2026 retail and CPG survey reports 47% of organizations using or assessing agentic AI. But adoption pressure is exactly why the checklist matters: urgency makes buyers lazy, and lazy buyers buy demos instead of platforms.

A caution on wording before the list: “AI-powered” is not a capability. It is a adjective that costs nothing to print, and it appears on the deck of every vendor in the category. The checklist below only counts a capability when it can be demonstrated on your data, in your market conditions, end to end. Anything that can’t be demoed is a roadmap item, and roadmap items get zero points.

The 8-Capability Checklist for Any AI Agent for CPG

Score each capability 0–2 (absent / partial / demoable with your own data). Anything below 10 out of 16 means the “agent” is a feature, not a platform.

#	Capability	What it does	Where it acts
1	Perfect store image recognition	Counts SKUs and facings, scores shelf share and contract compliance, identifies POSM and special displays from one photo	At the shelf, every visit
2	Replay / fake-photo detection	Flags reused or screen-captured photos so faked visits can’t pollute execution data	Every photo step
3	Store matching & expansion	Matches unlisted outlets by similarity score; recommends new stores from map + market data	Territory planning
4	AI chat report & KPI suggestion	Answers plain-language questions with live execution data; diagnoses weak KPIs and proposes actions	Manager cockpit
5	AI sell-in suggestion	Generates a store-specific pitch from store profile and local best-sellers	At the counter
6	AI training & exam	Role-play simulation pods with auto-scoring; adaptive learning paths with measured pass rates	Back office / onboarding
7	Order & invoice recognition	Reads courier slips and invoices, extracts amounts and dates, validates against addresses and windows	Back office / expense control
8	AI warehouse audit	Recognizes product types, quantities, and share in distributor warehouses; auto-generates audit reports	Distributor sites

Capabilities 1–2: The shelf loop and its lie detector

Perfect store image recognition is the capability everyone claims, so test it hard: hand the vendor a photo of your own shelf and watch it count facings, flag a POSM display, and score compliance. Push into the corners too — floor stacks, end caps, column wraps — because the irregular displays are where recognition models actually differ. Then ask the follow-up question almost nobody asks — what happens when the rep says the AI is wrong? A mature AI agent for CPG has an appeal workflow: the rep corrects the misread, the correction is logged, the score updates. That loop does double duty: it keeps reps engaged with the system instead of resentful of it, and it generates exactly the labeled corrections that make the recognition better in your market over time.

And underneath both sits replay detection. It is the least glamorous capability on the list and the most important, because a platform that cannot tell a live shelf photo from a screenshot is a platform whose entire dataset you cannot trust. Every KPI your leadership team reads — visit compliance, shelf share, display execution — inherits whatever the photo layer tolerates. Ask the vendor to attempt a faked submission in the demo and watch what happens.

Capabilities 3–4: Coverage and the manager’s questions

Store matching and big-data expansion decide how complete your outlet universe is. When a rep can’t find a store on the map — and in traditional-trade markets, that’s a daily event — the agent should recommend similar outlets with a similarity score and address, so the visit still gets logged against the right business. On the greenfield side, map data plus market data can point your developers at coverage gaps before competitors find them. On the reporting side, ask the vendor to answer a real question live — “which twenty stores declined this week and by how much?” — and time the answer. A chat report that runs on the live outlet master beats any scheduled dashboard, because it works the way managers think; the best ones go further and suggest what to fix first.

Capabilities 5–8: Sell-in, skills, paperwork, and the warehouse

AI sell-in suggestion is where agents earn their keep commercially: the right basket, pitched with the right talk-track, at the moment the rep is face to face with the store owner. The pitch part matters more than buyers expect — a suggestion without words is a chart, and reps don’t sell from charts. Training and exam capabilities close the skills loop: simulation pods where reps rehearse objection handling and get scored, with adaptive learning paths for the gaps. Order and invoice recognition quietly removes hours of back-office typing every week — courier slips read and validated against addresses and date windows, invoices extracted for reimbursement. And warehouse audit extends the agent’s eyes into the distributor, where a surprising share of “phantom availability” lives: the ERP says stocked, the warehouse shelf says otherwise.

Which Capabilities Matter First — by Role?

If you run sales: start with 5, 1, and 4. Sell-in suggestions move revenue this quarter; shelf recognition protects it; conversational reporting tells you where to push next. Route and training capabilities compound later.

If you run IT or data: start with 2, 8, and 7. Replay detection determines whether the data estate you’re building is worth anything; warehouse audit and document recognition shrink the integration surface with distributors and finance. The shelf loop will get all the demo attention, but the integrity and back-office capabilities are where CIOs earn their keep.

If you run sales operations: start with 3 and 6. Coverage completeness and rep ramp time are the levers you actually own, and both are measurable within a quarter. A territory that gains ten percent real outlets and a rep cohort that ramps in weeks instead of months will fund the rest of the checklist on their own.

How to Run the Checklist With Your Own Data

1. List the eight capabilities as scorecard rows. Circulate it to the vendor before the demo so nobody wastes time on theater.
2. Demand your data in their demo. Your shelf photos, your order slips, your store list. Canned demo images prove nothing.
3. Score 0–2 per capability, live. Partial credit is where most vendors land on 6–8; full marks on 1–5 is the differentiator.
4. Probe the failure paths. Wrong recognition, faked photo, ambiguous invoice — what happens next? Appeal workflows and replay detection are the tell.
5. Total it and walk away below 10. A platform scoring 14+ across all eight is a platform; a platform scoring 6 is a chatbot with a roadmap slide.

How eBest Maps to the Checklist

All eight are shipped capabilities on eBest’s route-to-market suite, not roadmap promises: perfect store image recognition with facings, POSM, and special displays; replay detection on every photo step; store matching and big-data outlet expansion; AI Chat Report with AI KPI Suggestion for managers; AI Selling Story at the counter; the scenario simulation pod plus AI Training & Exam; order and invoice AI recognition; and AI warehouse audit. They run on one outlet master across SFA, TPM, and DMS, with client deployments across beverage, food, and spirits — Coca-Cola and Carlsberg among the live references. We publish the checklist because we’d rather be scored on all eight than demoed on two.

Frequently Asked Questions

What is an AI agent for CPG field sales?

An AI agent for CPG field sales is software that autonomously executes parts of the route-to-market workflow — planning visits, recommending order baskets, verifying shelf execution from photos, answering manager questions, auditing distributor stock — instead of only recording what reps did. The defining trait is action: the agent takes the next step and logs the result without being prompted at each one.

How many capabilities should an AI agent for CPG have?

Eight covers the full route to market: shelf recognition, photo-integrity detection, outlet matching and expansion, conversational reporting with KPI suggestions, sell-in suggestions, training and exam, order and invoice recognition, and warehouse audit. Vendors typically demo two or three; the back-office and integrity capabilities are where platforms separate from point solutions.

Can an AI agent for CPG detect faked store visits?

The good ones can. Replay detection analyzes each submitted photo for signs it was reused or screen-captured, flags the submission, and lets the rep confirm or retake. Without it, visit-compliance data quietly rots, because a rep under time pressure will always be tempted by yesterday’s photo. Make it a scored checklist item, not an afterthought.

Does an AI agent for CPG work offline?

It must. Rural routes and dense urban markets alike lose connectivity daily, so visits, orders, and photos should queue offline and sync on reconnection. Ask specifically how image recognition behaves without a signal, since the shelf loop is where most of the agent’s value concentrates and where most demos quietly assume perfect 4G.

How do we evaluate an AI agent for CPG vendor quickly?

Use the 8-capability checklist as a scorecard, demand demos on your own data, score each capability 0–2 live, probe the failure paths — wrong recognition, faked photo, ambiguous invoice — and walk away from totals under 10 out of 16. A structured hour like this tells you more than a month of sales calls.

The eight capabilities are not a wish list — they are the operating system of a modern route to market, and the market is moving fast enough that scoring vendors casually is the expensive option. Run the checklist, demand your data in the demo, and let the totals speak. The platform that scores on all eight is the one that will still be acting for you a year from now.

In the contest of agentic AI vs traditional SFA, the honest starting point is this: traditional SFA records what your field team did; agentic AI decides and acts on what should happen next. Traditional SFA remains the system of record for visits, orders, and compliance — and it is genuinely good at that. Agentic AI adds a layer that plans routes, recommends order baskets, reads the shelf from a photo, and escalates exceptions on its own. For most CPG teams the right answer is layered, not replacement — but the layer only works on top of solid records, and 47% of retail and CPG organizations are already using or assessing it.

What Traditional SFA Still Does Well — and Where It Stops

Give the incumbent its due, because the upgrade conversation goes nowhere without it. Traditional SFA digitized the visit: geo-tagged check-ins, order capture with pricing and schemes, collections, shelf photos, attendance. It created accountability where there was none, and the audit trail it produces is the foundation everything else stands on. That record-keeping excellence is also why the agentic AI vs traditional SFA debate confuses buyers — the newer layer depends on the older one being done right.

Where it stops is the follow-through. A traditional SFA tells you that store 4472 ordered 60% of its usual volume last week. It does not tell the rep what to do about it while the rep is still standing in the aisle. The report lands at month end; the moment to act has passed. Because field selling is a same-day business, the gap between recording and acting is where revenue leaks — missed scheme tiers, empty facings, dormant outlets nobody nudged.

What can an AI agent do that traditional SFA can’t?

Act without being asked. In an SFA with agent capabilities, the system notices that store 4472 is sliding, cross-checks its order history and the local best-sellers, and pushes the rep a suggested basket with a pitch to match — before the visit starts. It reads the shelf photo the rep takes and scores compliance on the spot, then flags the facing gap to the supervisor the same day. That is the practical difference in agentic AI vs traditional SFA: one waits for humans to close the loop, the other closes it.

The search market noticed. Ahrefs data shows “agentic ai” climbing to roughly 111,000 monthly US searches while task-focused AI queries decline — buyers have moved from asking what AI is to asking what it can run.

Where Agentic AI Rollouts Go Wrong

Before the comparison table, the honest part: agents fail in predictable ways, and none of the failure modes are mysterious.

Dirty data first. An agent acting on a stale outlet master — closed stores still on the route, new outlets missing — produces confident nonsense. Teams that skip the data cleanup and jump straight to the AI layer usually spend a quarter discovering what their SFA administrator already knew. Fix the records, then let the agents loose.

Surveillance positioning. The moment reps perceive the agent as a tracker rather than a helper, adoption dies and workarounds bloom — spoofed check-ins, reused photos, bare-minimum data entry. This is precisely why photo-integrity features exist: they protect honest reps from the environment dishonest ones create. Position the agent as the thing that takes work off the rep’s evening, and the same features read as protection instead of policing.

Over-automation. An agent that rewrites routes every morning without explaining itself will lose the trust of a veteran rep in one bad week. The workable pattern is suggestion first, automation later, with the rep’s accept-and-reject behavior feeding the configuration. Autonomy is earned by the system, not declared by the vendor.

No owner. Agentic layers blur the line between sales operations and IT. Teams without a single named owner end up with configuration nobody maintains and KPIs nobody watches. The successful rollouts we see all have one person whose scoreboard includes agent acceptance rate — and who cares about it.

Agentic AI vs Traditional SFA: Side-by-Side

Dimension	Traditional SFA	SFA with Agentic AI
Core role	System of record	System of record + system of action
Visit planning	Fixed beat lists, manually adjusted	Agent builds and re-balances routes to hit coverage goals
At the counter	Rep decides from memory and price list	Agent suggests the order basket and the scheme to unlock next tier
Shelf	Photo stored for later review	Image recognition scores facings and compliance immediately; exceptions escalate
Manager insight	Static dashboards, month-end review	Conversational reports — ask, get the answer, drill down live
Data entry	Rep fills forms	AI reads orders, invoices, and photos; rep confirms
Failure mode	Data goes stale, follow-up slips	Agents act on live data; gaps surface same day

Read the table left to right and the pattern is clear: nothing on the right side works without the left side. Agents acting on garbage visit data just produce faster garbage. The agentic AI vs traditional SFA decision, then, is really a sequencing decision — records first, agents second, one system underneath.

Will Agentic AI Replace Field Sales Reps?

No — and the reason is commercial, not sentimental. The rep’s irreplaceable work is human: negotiating with the store owner, reading the room, keeping the relationship warm when a competitor’s merchandiser is crowding the shelf. What AI can take over is the around-the-work: planning the day, drafting the pitch, filing the report, checking the scheme math. EY’s analysis of agentic commerce makes the same point for the back office — humans own judgment and exceptions; agents own repetition and scale.

So the practical framing for agentic AI vs traditional SFA isn’t “which one survives.” It’s which one your reps want. A rep whose agent hands them a ready route, a store-specific pitch, and an instant compliance read gets hours of selling time back — time spent face to face with the store owner, which is exactly where the deal is won. Teams that position agents as co-pilots see adoption; teams that position them as surveillance see workarounds.

There’s also a hiring angle that rarely makes the vendor decks. Young reps ramp faster when the agent narrates the why — this store orders every other Tuesday, that owner responds to range suggestions, this scheme unlocks at forty cases. The institutional knowledge that used to take two years of riding along with a senior rep gets encoded in the suggestions themselves. That matters in markets where field turnover is high and the training bench is thin.

How to Pressure-Test the Shift in 30 Days

Before you re-platform anything, run a bounded test on the data you already have.

1. Pick one territory and one loop. Choose a territory with clean visit history and a single high-value loop — order suggestion or shelf compliance.
2. Baseline for two weeks. Measure productive visits, lines per order, and facing compliance under the current SFA. No changes.
3. Switch the agent layer on for two weeks. Same reps, same stores, agent suggestions active. Reps can accept or reject every suggestion; log the acceptance rate.
4. Compare and interview. Compare the KPIs, then ask reps what the agent got wrong. Rejected suggestions are your configuration backlog, not a failure verdict.
5. Decide with numbers. If acceptance rates and the KPI deltas hold, expand. If not, you learned it in 30 days on one territory, not in a 12-month contract.

How eBest Layers Agents on SFA

eBest runs SFA and DMS on one outlet master, which is what makes the agent layer coherent: AI Chat Report answers a sales director’s plain-language questions with live store-execution data; AI KPI Suggestion diagnoses weak territories and proposes corrective actions; AI Selling Story drafts the store-specific pitch from order history and local best-sellers; perfect store image recognition turns shelf photos into compliance scores, with replay detection guarding the integrity of every photo. Beverage and food brands running this stack — the client results include live deployments with Coca-Cola and William Grant & Sons — treat the records layer and the agent layer as one system, because that’s how the loop closes. That is the quiet lesson of every agentic AI vs traditional SFA debate: the teams that win started by fixing the outlet master.

The comparison, in one line: traditional SFA made the field visible. Agentic AI makes it responsive. You need both, in that order — and the teams that get the order right are the ones quietly pulling ahead.

Frequently Asked Questions

What is the difference between agentic AI and traditional SFA?

Traditional SFA records field activity — visits, orders, photos, collections — and reports on it. Agentic AI acts on that record: it plans routes, recommends order baskets, scores shelf compliance from photos, and escalates exceptions without waiting for a manager. The short version: SFA is a system of record; agentic AI adds a system of action on top. Both are needed, and the order matters — records first, agents second.

Is agentic AI the same as generative AI in SFA?

Not quite. Generative AI produces content — text, summaries, answers — when prompted. Agentic AI uses models like that as components but adds autonomy: goals, multi-step execution, and tool use. In an SFA context, generative AI might draft a sales message; an agentic system decides which store needs the message, times it, sends it, and logs whether the order followed.

Will agentic AI replace field sales reps in CPG?

No. The rep’s human work — negotiating, relationship-building, reading the store owner — stays human, and it is where deals are won. What agents replace is the around-the-work: route planning, report filing, pitch drafting, scheme math. The realistic outcome is reps with more selling time, not fewer reps.

What should CPG teams keep in traditional SFA after adding agentic AI?

The records. Visit compliance, order history, payment tracking, and audit trails remain the foundation — agents are only as good as the data they act on. Keep a single outlet master, clean visit data, and disciplined photo capture, because every agent capability multiplies the quality (or the garbage) already in the system.

How do we start with agentic AI if we already have an SFA?

Run a bounded pilot rather than a re-platform. Pick one territory and one loop — order suggestion or shelf compliance — baseline two weeks, enable the agent layer for two more, then compare KPIs and interview reps about rejected suggestions. Decide with those numbers. Most teams that pilot this way expand within a quarter because the acceptance data makes the case for them. And keep the pilot narrow on purpose: one loop, one territory, a clear owner. Wide pilots produce averages; narrow pilots produce decisions.

Traditional SFA made the field visible; agentic AI makes it responsive — and the 47% of CPG organizations already testing agents (NVIDIA survey, EY analysis) suggests the window for a calm, evidence-based pilot is now. Start with one territory, keep the records honest, and let the acceptance data argue for the expansion.

An agentic AI platform for CPG is software where AI agents plan, decide, and act across the route to market — building the visit route, generating the store-specific pitch, reading the shelf from a photo, and triggering follow-up — instead of waiting for a human to interpret a report. The fastest way to evaluate one is to apply five criteria: does it act at the shelf, does it run on one data model, does it name its capabilities, does it protect data integrity, and can it actually go live in your markets. Nearly half of retail and CPG companies are already using or assessing agentic AI, so the question on the table is no longer whether to buy — it’s how to buy well.

Why the Agentic AI Platform Question Hit CPG in 2026

NVIDIA’s third annual State of AI in Retail and CPG survey found that 47% of organizations are using or assessing agentic AI, and nine in ten plan to increase AI budgets this year. Search demand moved the same way: “agentic ai” now draws roughly 111,000 US searches a month, up 37% year over year, while the older “AI for a task” query cluster is shrinking.

So every vendor with a dashboard has reprinted the word “agentic” on their deck. That’s the trap. A chatbot bolted onto a legacy SFA can summarize yesterday’s visits, and it will happily call itself an agent. The gap between that and a true agentic AI platform is where implementations stall — the assistant talks, the shelf stays untouched, and the rep does the real work anyway.

The cost of choosing wrong compounds quietly. A failed assistant costs a license; a failed “agent” costs a year of field data that nobody trusts, a sales team trained to ignore the app, and a board that concludes AI doesn’t work in route to market. That second outcome is far more common than the vendors admit, and it is almost always a selection failure, not a technology failure — the buyer scored slides instead of scoring capabilities.

Here’s the thing: you can spot the difference in a thirty-minute demo if you know what to look for. Five criteria do most of the work.

What an Agentic AI Platform Actually Is (and Isn’t)

Not an agentic AI platform: a chatbot that answers questions about your sales data. A report that recommends an action and then waits. A voice bot that books calls. Those are assistants. Useful, but assistants.

Is an agentic AI platform: software where agents carry a goal — “make sure the display agreement is honored in this territory” — and move through the steps themselves: check the shelf photo, score compliance, raise the exception, notify the supervisor, log the outcome. Gartner has framed the industry shift as exactly this, from assistive to agentic systems.

The distinction matters more in CPG than in most industries, because field selling, distribution, and promotion are one continuous motion. An agent that only handles conversation leaves the shelf — the place revenue is actually won — untouched.

How is an agentic AI platform different from “AI-powered” SFA?

Ask a vendor what their AI does when nobody is watching. On an AI-powered SFA, nothing: the insight sits in a dashboard until a human reads it. On an agentic AI platform, the agent acts — it recommends the order basket at the counter, verifies the shelf from the photo the rep just took, and escalates the store that missed its scheme tier. Automation runs the steps you wrote; an agent reaches the goal you set. That one-line test is worth pinning above every demo.

The 5 Criteria: How to Evaluate an Agentic AI Platform

Score every candidate against these five. If a platform fails the first one, the rest barely matter.

Criterion 1: The agent acts at the shelf, not just in the conversation

This is the CPG-specific test. Ask the vendor to show the full loop: rep takes a shelf photo → image recognition counts facings and scores compliance → the result posts to the visit record → an exception triggers when a facing drops. In a real agentic AI platform, that loop runs without anyone exporting CSVs. Ask to see it live, on a shelf photo you supply — not their canned demo image.

Criterion 2: One data model across SFA, DMS, TPM, and DSD

Agents are only as good as the data they can act on. If the visit app, the distributor system, and the promotion tracker are separate products stitched by integrations, your agent will act on stale fragments. Ask directly: is the outlet master shared? Does the order an agent recommends flow into the distributor’s stock check and the promotion’s spend record without a sync job? On eBest’s stack, for example, SFA and DMS run on one outlet master precisely so capability loops like sell-in suggestions can see distributor stock.

Criterion 3: Named, auditable capabilities — not “AI-powered”

Vague claims are a red flag. An honest agentic AI platform names its agents and what each one does: AI sell-in suggestion generating a store-specific pitch, perfect store image recognition scoring facings and POSM, an AI chat report that answers manager questions in plain language, warehouse audit reading distributor stock photos. Naming matters because it makes the capability testable — you can demand a demo of each named feature and score what you see. Count how many discrete, named capabilities the platform actually ships. Fewer than five across the full route to market means you’re buying a feature, not a platform.

Criterion 4: Data integrity and human-in-the-loop controls

Nobody’s favorite slide, and the one that decides whether your data is worth anything in eighteen months. Two questions: First, can the system detect a faked visit — a rep submitting yesterday’s photo from a different store? Replay detection exists for this; if a vendor has never heard of the problem, their execution data is fiction waiting to happen. Second, what happens when the AI is wrong? Look for an appeal workflow — the rep corrects a misread shelf count, the correction is logged, the model result updates. A platform that forces blind trust in its AI will leak both trust and data quality.

Criterion 5: Deployment reality in your markets

Finally, ask how it lands. Offline coverage for rural routes? Local languages for voice and chat? Time to go live measured in weeks, not quarters? Which integrations ship as standard — ERP, distributor systems? The agentic AI platform that demos beautifully but needs a six-month implementation in your top market isn’t agentic where it counts.

Two follow-ups separate the realists from the brochure-writers. Ask how historical data migrates: a serious vendor loads your existing outlet master and visit history into the evaluation environment, because agents need history to be useful on day one. And ask which market went live most recently, in which country, in how many weeks — then ask for a reference call. The pattern across the industry is blunt: platforms that run on one data model go live in weeks; bolted-together stacks spend those weeks building integrations instead.

Agentic AI Platform Evaluation: Scorecard to Bring to the Demo

Criterion	What great looks like	Question that exposes the gap
Acts at the shelf	Photo → recognition → compliance score → exception, live	“Run the loop on my shelf photo, now.”
One data model	Shared outlet master across SFA/DMS/TPM	“Does the agent’s order hit the distributor’s stock check directly?”
Named capabilities	5+ discrete, named, demoable features	“List every named AI capability you ship.”
Data integrity	Replay detection + rep appeal workflow	“Show what happens when a rep submits a reused photo.”
Deployment reality	Offline mode, local languages, weeks-to-live	“Which market went live last quarter, in how many weeks?”

Bring the scorecard. Make each vendor fill it in front of you. The exercise usually ends the debate faster than any RFP.

One discipline tip: score the demos the same day, while the details are fresh. Memory flattens differences, and by the third demo every platform blurs into the same buzzwords. The evaluation notes your team writes the afternoon of the demo are the ones the decision should rest on — not the follow-up decks the vendors send afterward.

How eBest Approaches Agentic AI for CPG

eBest built its agent layer on top of a full route-to-market suite — SFA, DMS, TPM, DSD — so the agents act on one data model by default. The named capabilities map to the criteria above: AI Selling Story generates the store-specific pitch during the visit; perfect store image recognition scores facings, POSM, and special displays with a rep appeal workflow built in; replay detection guards visit authenticity on every photo step; AI Chat Report lets a sales director ask “which stores declined this week?” and get the answer, not a dashboard tour; AI warehouse audit extends the loop into distributor stock. Client results across beverage and food customers — including live deployments with Coca-Cola and Carlsberg — show the pattern: the value shows up when the agent closes the loop, not when it chats.

Our own bias, stated openly: we’d rather name eight capabilities and let you test each one than promise an all-knowing agent. That’s what the criteria are for.

Frequently Asked Questions

What is an agentic AI platform in CPG?

An agentic AI platform in CPG is software where AI agents autonomously execute route-to-market work — planning routes, recommending orders, verifying shelves, auditing distributor stock — against a shared data model, rather than only answering questions or producing reports. The defining test is action: when the agent finds a problem, it takes the next step itself and logs the result.

How is an agentic AI platform different from a chatbot?

A chatbot responds; an agent acts. Ask the vendor for one workflow where their AI completed a multi-step task — checked data, made a decision, triggered an action, and recorded the outcome — with no human prompting each step. If every example ends with “and then the manager decides what to do,” you’re looking at a chatbot with a strategy deck.

How much does an agentic AI platform cost for a CPG company?

Most enterprise agentic AI platforms for CPG price per user per month, with modules priced separately, on annual contracts — the SFA-plus-agents tier typically lands in the same range as established field-sales suites. The honest answer: pricing varies too much by market and module count to quote blindly, so ask each vendor for a per-user, per-module breakdown and make the agent layer a visible line item, not a bundled mystery.

Can an agentic AI platform work offline in rural markets?

It has to. In the markets where CPG growth is happening — rural India, Southeast Asia, parts of Africa and South America — connectivity drops daily. The platform should queue visits, orders, and photos offline and sync automatically on reconnection. Ask specifically how image recognition behaves offline, since the shelf loop is where the value concentrates.

How long does it take to implement an agentic AI platform?

For a mid-size CPG rollout on a modern SaaS platform, plan on weeks per market for the core loop — outlet master, routes, order capture, shelf recognition — with the agent layer configured, not coded. If a vendor quotes six months before your first rep sees the app, the implementation burden, not the AI, is the product. Ask for the last three go-lives and their timelines.

Fifty years of route-to-market software taught the industry to record the field. Agentic AI is the first wave that can work it — but only if you buy the version that acts. Walk into the next demo with the five criteria, run the shelf loop on your own photo, and count what the platform does when nobody prompts it. The platform that keeps working is the one to shortlist.

Merchandising agreements are the quiet contracts that decide whose cooler sits beside the checkout, which brand owns the eye-level facings, and whose poster hangs in the window. In a beer business running through tens of thousands of small stores, these deals were traditionally signed on paper, checked from memory, and enforced on trust — which often meant they were not enforced at all. So when Asahi set out to modernize its route to market in China, the agreement itself became the thing to fix: mobile signing, standardized execution criteria, and automatic conflict checks built into the daily workflow of its field sales team.

What is a merchandising agreement?

A merchandising agreement is a contract between a brand and a retail outlet in which the store commits to specific in-store execution — cooler placement, shelf facings, poster displays, floor stacks — and the brand funds that commitment. In the non-alcoholic beverage trade the annual version is often called a cooperative merchandising agreement, or CMA, and the mechanics are the same across categories: the brand pays for guaranteed visibility, the retailer delivers it, and both sides depend on proof.

Because visibility is physical, every clause is a testable standard. Is the cooler plugged in and running? Does the brand’s share of the shelf hold? Is the poster intact and in the agreed position? The non-alcohol beverage industry runs this as a yearly negotiation cycle, with resets concentrated in the first quarter — Gartner’s 2026 analysis of computer vision replacing manual retail counting describes how that verification burden is shifting from clipboards to phones.

The four shapes an agreement takes

In practice, brands do not run one agreement type; they run a portfolio. A required item — say, a branded cooler — every qualifying store must take. Optional items, like shelf facings or poster displays, the store chooses. Pick-N structures let a store choose any three displays from a menu of six. Bundles package several items at one price, with different terms for peak season versus off-season. Each shape carries its own pricing windows, its own renewal rules, and its own ways of colliding with the others — which is exactly where paper processes start to fail.

Why do merchandising agreements break down at the store level?

The breakdown is rarely dramatic. It is accumulation, and it looks the same in every market that still signs on paper.

The paperwork itself is the first leak. A field rep visits a store, negotiates a display package, fills in a form, collects a signature, and mails it back. The store cannot see what it already has signed; the rep cannot see it either, because last year’s contracts live in a filing cabinet or a spreadsheet nobody maintains. The result is overlap: paid months that stack on top of each other, agreements that quietly extend past their intended end, and renewals that start before the old term has expired — each one a small overpayment or a dispute waiting to happen.

The second leak is conflict. Some agreements cannot coexist: a store moving from a premium display tier to a standard tier, or switching channel types mid-contract, should not hold both at once. On paper, nothing stops it, because the person signing in the store has no view of what the system already holds.

The third leak is standards drift. The same display clause means different things in different regions when the criteria live in people’s heads instead of in a system. And when execution is not measured against a written standard, follow-up becomes opinion: the brand believes it bought facings it never received, the store believes it delivered, and there is no record either side can point to.

None of this is carelessness. It is what happens when the agreement — a structured, rule-heavy object — is handled as unstructured paper.

Case in point: digitizing merchandising agreements at Asahi

Asahi runs one of the most recognized beer portfolios in the world, and in China its field team signs merchandising agreements with a long tail of small-format stores — grocery stores, corner shops, traditional trade outlets. The specifics of the program are Asahi’s to tell, so this account deliberately stays at the level of mechanics rather than commercial terms. What is worth examining is the shape of the fix, because it generalizes to any brand managing display deals at scale.

The signing journey moved to the phone

Instead of paper forms traveling between store and office, the rep opens the store’s record in the mobile app and starts a signing request there. The system proposes the start of the paid period automatically, based on configured parameters rather than the rep’s arithmetic. The rep selects the display items the store will take, the store owner signs on the device, and the agreement enters the system of record at the moment of signature — not days later when the paperwork lands.

Execution standards became configuration, not folklore

Every display item carries its acceptance criteria inside the platform, visible to the rep before signing and to the auditor after. What counts as a compliant cooler — placement in the main traffic flow, powered on, kept clean, holding the agreed product range — is written once as a standard and applied the same way in every region. Numerical and percentage-based criteria are supported side by side, so a rule can be a simple pass/fail or a measured share. The rep can also pull up the standard for any item on the spot, which ends the “I thought it meant something else” conversation before it starts.

The system catches conflicts before the signature

This is the part worth underlining. When a rep initiates a re-signing for an agreement that was previously terminated or expired, the platform checks the old record and rolls the start of the paid period forward past its end — no manual dates, no silent overlaps. When a store switches from one agreement tier to a mutually exclusive one, the same check applies across the exclusion. The conflict that used to surface months later as a billing dispute now surfaces as a message on the screen at signing time. Because the rule is enforced by the system, it is enforced identically everywhere, which is the only way a national standard survives contact with a national field force.

How do you prove an agreed execution actually happened?

A signed agreement is a promise; execution is a photo. The natural close of the loop is capture during the routine store visit, and this is where the AI layer earns its place.

With AI image recognition built into the visit workflow, the rep photographs the cooler or the shelf, and the system reads what is there — facings, product presence, display condition — and scores it against the criteria the agreement defined. There is an appeal path when the AI misreads a scene, so the rep stays in control of the result rather than the algorithm. And because photo evidence can be faked with a screenshot or an old picture, replay detection runs on every photo step: a recaptured image gets flagged before it pollutes the compliance report. Honest reps are protected, and the dishonest path becomes expensive.

That said, verification photos are only useful if someone can actually interrogate them at scale. Managers who ask “which stores fell out of compliance on cooler placement this month?” get the answer by asking the platform in plain language, through AI Chat Report, rather than commissioning another spreadsheet. The picture of execution stops being a quarterly retrospective and becomes a running score. This is the broader shift McKinsey describes across distribution operations: the value is not collecting more field data, it is closing the loop between what was promised and what was done.

From paper to proof in four steps

1. Model the agreement portfolio. Required, optional, pick-N, and bundle structures go into the platform as configuration, each with its own pricing windows and renewal rules.
2. Write the standards. Every display item gets explicit, region-independent acceptance criteria the rep and the auditor both see.
3. Turn on conflict validation. Overlapping paid months and mutually exclusive tiers are checked automatically at signing time, not in year-end reconciliation.
4. Close the loop with photo verification. Routine visits capture AI-scored execution evidence, with replay detection guarding the integrity of every photo.

What merchandising agreement software should actually do

Plenty of tools touch this problem; few manage the agreement itself.

Capability	Paper + spreadsheet	Generic checklist app	Agreement-aware SFA
Signing at the store	Paper form, mailed back	Not supported	Mobile signature at the moment of the deal
Agreement types & seasons	Manual memory	Flat task lists	Required / optional / pick-N / bundle as configuration
Overlap & exclusion checks	None	None	Automatic, applied before the signature
Execution standard	Ambiguous prose	Generic checklist	Structured criteria, same in every region
Proof of execution	Disputed	Photos without context	AI-scored against the signed standard
Traceability	Filing cabinet	Partial	Every record carries store, date, and version history

The pattern that works treats merchandising agreements as data — structured, versioned, and enforced by rules — rather than as documents.

How eBest handles merchandising agreements end to end

The Asahi program described here runs on eBest’s sales force automation platform, with the agreement engine sitting inside the same system reps use for visits, orders, and evidence capture — so signing, standards, conflict checks, and photo verification are one workflow instead of four tools. Where the funding side of display deals needs managing, the same chain connects to trade promotion management, and the store-level records reconcile naturally with the distributor management layer upstream.

The outcomes are qualitative but consistent, and they mirror what the mechanics predict: signing happens in the store instead of after it, the same standard applies everywhere, paid periods stop overlapping, and every agreement carries its own auditable trail. For brands weighing a similar move, the client success stories page is the honest place to see how the pieces fit together in other networks.

Frequently Asked Questions

What is a merchandising agreement in the beverage industry?

It is a contract in which a retailer commits to specific in-store execution — cooler placement, shelf facings, posters, floor displays — and the brand funds that commitment. In non-alcoholic beverages the annual negotiated version is often called a cooperative merchandising agreement. The agreement is fundamentally a set of testable standards, which is why digital versions pair each clause with measurable acceptance criteria and photo-based verification.

How do brands enforce merchandising agreements at the store level?

Enforcement has two halves. The first is at signing: the agreement should be captured digitally with structured items, prices, paid periods, and automatic checks for overlapping terms or mutually exclusive deals. The second is after signing: reps photograph the cooler and shelf during routine visits, AI image recognition scores the photos against the agreed standards, and exceptions route to managers while the evidence trail — store, date, location — remains attached to every record.

What causes conflicts between merchandising agreements?

Most conflicts come from paid-month overlap and mutual exclusion. A store re-signs an agreement type it previously terminated, and the new paid period starts before the old one ends; or a store holds two tiers at once that the program rules say cannot coexist. These are arithmetic and rule problems, which is why they belong in software: the system knows each store’s agreement history and rolls start dates forward or blocks the signing before the conflict is created.

Can merchandising agreements be signed on a mobile phone?

Yes, and that is the point of digitizing them. The rep opens the store record, selects the display items, and the store owner signs on the device; the agreement enters the system of record immediately, with the paid period calculated from configured rules. Mobile signing also makes the standards visible at the moment of commitment — the rep can show the acceptance criteria for any item on the spot, which removes ambiguity from the deal before it is signed.

What should I look for in merchandising agreement software?

Look for four things. Portfolio modeling: required, optional, pick-N, and bundle structures with seasonal pricing as configuration, not custom code. Conflict validation: automatic checks for overlapping paid months and mutually exclusive tiers applied at signing. Execution standards: structured acceptance criteria shared by reps and auditors alike. And proof of execution: AI image recognition scored against the signed standard, with replay detection on the photos and an appeal path when the AI gets it wrong.

The agreement is where in-store execution is won or lost, long before anyone opens a cooler door. Brands that treat merchandising agreements as structured data — signed on the phone, standardized in configuration, validated by rules, and proven by photos — stop paying for visibility they never received. If display deals are still living in filing cabinets, that is the first workflow worth digitizing.

An AI warehouse audit is a photo-based check in which computer vision recognizes the products, quantities, and product share sitting inside a distributor’s warehouse, compares them against what the brand’s records claim, and generates the audit report automatically. It exists because distributor stock records and physical reality rarely match, and because manual stock audits are slow, expensive, and easy to work around. For CPG brands, it turns inventory truth from a quarterly audit event into a five-minute routine on a phone.

Why do distributor stock records drift from reality?

Every CPG brand that sells through distributors lives with the same quiet gap. The system says one number; the warehouse shelf says another. Because a distributor management system records what was *invoiced*, not what is physically sitting on the racks, the two drift apart from day one.

The causes are mostly ordinary, which is exactly why they are hard to stop. Goods get damaged in handling and stay in the system as sellable stock. Cartons expire in a corner of the warehouse and nobody rotates them out. Returns come back from retailers and never get booked. Slow-moving SKUs pile up behind fast movers and quietly become dead stock. And then there is the deliberate side: when a claim, a rebate, or a sales incentive hangs on how much stock is “in the channel,” there is a standing reason to over-report it. Third-party stock audit firms in markets like India have publicly described cases where a large share of recorded stock turned out to be expired or damaged but was still shown as active inventory — findings that only surfaced because someone finally counted.

The hidden cost of phantom inventory

Phantom inventory does real damage downstream, and none of it shows up on a single invoice. Replenishment plans built on inflated stock records under-order, so the brand loses sales it thought it already had. Trade promotion reviews use the same records, so promo ROI gets judged on numbers that were never true. Finance carries credit exposure against stock that does not exist. McKinsey’s work on AI in distribution operations notes that leading distributors now use AI to cut inventory levels by 20 to 30 percent while improving fill rates — but every one of those gains rests on knowing what stock is actually there. You cannot optimize inventory you cannot trust, so the counting problem comes first.

What Is an AI warehouse audit?

An AI warehouse audit applies computer vision to this counting problem. Instead of sending auditors with clipboards and barcode scanners, the brand asks for photos. A field rep, a distributor’s own staff, or a scheduled audit workflow captures images of the warehouse racks and stock areas. The AI then recognizes which products are visible, estimates quantities and facings, computes each product’s share of the recorded stock, and generates an audit report that reconciles against the brand’s own records.

In practice, the output answers the questions a sales director actually asks: Does the distributor hold the stock the system says they hold? Which SKUs are overstocked, which are missing, and which have quietly expired? Is stock being staged only for audit day and dispersed afterward? Because the audit rides on the same platform that captures daily field activity, it can run continuously on a sample of visits rather than as a disruptive annual event — which is precisely the shift Gartner describes in its 2026 analysis of AI and computer vision replacing manual inventory counting: the value is not counting faster, it is removing the counting work altogether.

What the AI sees in a warehouse photo

The recognition layer is doing several jobs at once. It identifies products by brand, pack, and variant — distinguishing, say, a 500 ml PET bottle from a 1.5 l one on the same rack. It counts facings and cartons to estimate quantity. It flags damaged or expired-looking stock for follow-up. And critically for CPG, it distinguishes your products from competitors’ products sitting in the same distributor warehouse, because distribution share inside the warehouse is a leading indicator of what will appear on store shelves next month.

How Does an AI warehouse audit work in the field?

The workflow is deliberately simple, because the people capturing the evidence are field reps and distributor staff, not auditors. A working rollout has four steps.

Step 1: Set the audit scope before anyone picks up a phone

Decide which distributors, which SKUs, and which claim types the audit covers. Most brands start narrow — top distributors, the SKUs tied to an active promotion or a dispute — then widen the scope once the reconciliation patterns are understood. The scope lives in the platform, so the app knows exactly what a rep is expected to capture when they open the audit task.

Step 2: Capture the photos — and prove they are real

The rep walks the racks and takes photos inside the distributor’s warehouse, including in low-connectivity sites where uploads queue and sync later. Here is the part most checklist apps skip: photo evidence can be faked with a screenshot or an old picture, so every audit photo passes replay detection. If the system recognizes the image as a replayed or recaptured photo rather than a live shot of the rack, it flags the submission before it pollutes the report. Honest reps are protected, and the dishonest path gets expensive.

Steps 3 and 4: Reconcile against the DMS and act

The AI processes the photos, produces per-SKU counts and share, and the platform reconciles them against the distributor’s recorded stock in the DMS. Variances surface immediately — missing stock, unrecorded returns, excess slow movers, expired goods still on the books. From there the workflow branches into action: the distributor disputes a count through an appeal and the photo gets re-reviewed, or the brand corrects its records, adjusts a claim, or triggers replenishment based on what is actually on the rack. The report is generated automatically, so the audit trail — who photographed what, when, and where — comes for free.

AI warehouse audit vs manual stock count: What actually changes?

The honest comparison is not AI versus humans; it is what each method can realistically cover at distributor scale.

Dimension	Manual audit firm	Rules-based scanning (barcode/checklist)	AI warehouse audit
Time per site	Days to weeks	Hours	Minutes per photo pass
Coverage	Sample of distributors, quarterly at most	Sites with scanners and trained staff	Any site with a phone; runs on routine visits
Fraud resistance	Dependent on auditor presence	Weak — a scanned substitute passes	Flags replayed or recaptured photos automatically
Data captured	Totals per SKU	Totals per SKU	Products, quantities, share, competitor presence, damage flags
Cost profile	Per-audit engagement fees	Hardware plus training	Included in the field platform
Audit trail	PDF report after the fact	Partial	Photo, time, and location attached to every count

For brands running third-party audits today, the pattern that works is hybrid: continuous AI audits on the wide distributor base, with manual verification reserved for the outliers the AI surfaces.

How eBest runs AI warehouse audits inside the DMS

eBest treats the AI warehouse audit as one scene in a full route-to-market chain rather than a standalone app, which is what makes the numbers reconcile. The audit capability recognizes product types, quantities, and product share on warehouse shelves, then auto-generates the audit report for brand activity management — and because it sits inside the same platform as the distributor management system, every count lands directly against the records it is meant to verify.

Two adjacent capabilities make the evidence trustworthy. First, the same AI image recognition engine that reads store shelves — with a rep-side appeal workflow when it misreads a photo — also reads warehouse stock, so one recognition standard covers both sides of the trip. Second, replay detection is built into every photo step, so the audit photo of the rack is provably a live capture. Teams that run DSD routes can layer audits onto the direct store delivery workflow, since the van that replenishes stores is also the natural photo point for the warehouse it loads from. Beverage and FMCG networks eBest works with — including brands operating Coca-Cola and Carlsberg distribution-scale systems — run thousands of daily visit photos through this pipeline, which is what turns audit from an event into a habit. Results from brands running the broader platform are qualitative but consistent: cleaner reconciliation, faster claim validation, and fewer arguments about what the channel actually holds. To see how audits fit into a wider execution stack, start with client success stories.

Frequently Asked Questions

What is an AI warehouse audit in FMCG?

It is a photo-based inventory verification in which a phone camera captures the distributor’s warehouse racks and computer vision recognizes the products, quantities, and share of stock visible. The system reconciles those counts against the brand’s records and generates an audit report automatically. Unlike a manual stock audit, it can run on routine field visits instead of requiring a special audit event, so the brand gets continuous visibility into whether distributor stock matches the books.

How accurate is the image recognition?

Accuracy is highest when photos follow a consistent capture pattern — racks face-on, adequate lighting, one bay per frame — and the AI flags low-confidence reads instead of guessing. Equally important, the workflow is not blind trust: when a rep believes the AI misread a photo, they file an appeal, the image is re-reviewed, and the corrected result updates the report. That human-in-the-loop correction loop is what keeps the numbers converging on reality over time instead of drifting.

Does it work offline in low-connectivity warehouses?

Yes. Distributor warehouses are frequently concrete basements and steel racks where signal is poor, so capture runs offline-first: photos and audit data queue on the phone and sync when connectivity returns. The recognition happens server-side after sync, so the rep’s workflow stays the same whether they are in a metro DC or a rural stockist. This matters because audits that only work on good Wi-Fi quietly stop being performed in exactly the warehouses that need them most.

How is a warehouse audit different from a shelf audit in stores?

A shelf audit checks what is *on display* at the retail end — facings, planogram compliance, out-of-stocks, POSM. A warehouse audit checks what is *in stock* at the distribution end, one step earlier in the chain. The two answer different questions: shelf audits explain why a store looks the way it does; warehouse audits explain whether the distributor could have supplied it at all. Brands running both get a connected view — a persistent shelf gap with adequate warehouse stock points to an execution problem, while a gap that starts in the warehouse points to supply.

How long does a rollout take?

Because the audit rides on the existing field-sales app rather than a separate tool, the practical rollout is scope-setting and training, not a new deployment. A typical start: define the SKU list and distributor sample, brief reps on capture patterns, run audits on routine visits for four to six weeks, and review reconciliation variances before widening scope. Brands already operating the platform for daily visit reporting tend to treat the warehouse audit as a configuration step on top of workflows their teams use every day.

The stock you cannot see is the stock you keep paying for — in write-offs, in claims, and in forecasts built on fiction. An AI warehouse audit closes that gap with something every field rep already carries, and it keeps the evidence honest in a way clipboard audits never could. If distributor reconciliation is on your agenda this year, the counting problem is the right place to start.

Van sales software runs the one distribution model where the vehicle is the warehouse: stock is loaded in the morning, sold and invoiced at the retailer’s doorstep, and reconciled the same evening. Because every unit of truth — what left, what sold, what came back, who paid — is created on the road, the software has to control the whole route day, not just record it. This article walks that route day end to end and lists the controls that keep direct store delivery (DSD) routes honest.

Why Van Sales Is the Hardest Route to Market to Control

Pre-sales has a paper trail by design: the order is booked today and fulfilled from a warehouse tomorrow, so the stock movement lands in a system somewhere between the two events. Van sales has no such buffer. The moment a van leaves the depot, the brand’s inventory is sitting in a vehicle on a rural road, and the only people who know what is inside it are the driver-salesman and, eventually, the paper log.

That single structural fact explains most van sales failures. Stock checks happen at the depot, not at the doorstep. Cash and credit change hands at the same moment the sale happens, against a credit limit nobody can see from the driver’s seat. Returns and near-expiry stock ride back silently at the end of the day. By the time anything is keyed into a spreadsheet, the correction window has closed — the same retailer has already been skipped twice, and the missing cartons have become a monthly argument instead of a same-day fix. Van sales software exists to close exactly that gap, and the operations that treat it as a recording tool rather than a control system keep paying for the gap anyway.

Gartner’s 2026 data and analytics predictions describe the general version of this problem: organisations that keep their evidence locked in manual, specialist workflows are choosing slower decisions by default (Gartner D&A predictions, March 2026). In a van sales operation the lag is not a productivity nuisance; it is the difference between catching a route leak on Tuesday and discovering it at month-end.

What van sales software actually replaces

A proper van sales app replaces the paper route book and the end-of-day tally — the ritual where the rep counts leftover stock, subtracts it from the load-out sheet, and hopes the cash matches. Instead of that reconstruction, every event is captured where it happens: the load-out when the van is picked, the sale and invoice at each stop, the return with its reason code, the settlement when the driver is back. The reconciliation stops being an evening detective story and becomes a report that already agrees with itself.

How Does Van Sales Software Work Across the Route Day?

The mechanics follow the vehicle, so it helps to walk the day in order. Done well, each handoff feeds the next one, and nothing depends on the rep’s memory after a twenty-stop day.

Load-out: the van becomes a stock location

The day starts with a picking list against the planned route, and van sales software treats the van as a moving warehouse with its own stock ledger. Every case scanned or confirmed onto the vehicle is deducted from depot stock and credited to that van. Because the load-out is a transaction rather than a note, the evening reconciliation has a starting number nobody can argue with. Skip this step, and the rep spends the whole day selling from an estimate.

Sell, invoice and collect at the doorstep

At each stop the rep works from live van stock, not a printed price list. The sale is priced with any active scheme applied automatically, the invoice is generated on the spot, and the payment method — cash, mobile money, or credit against the retailer’s account — is recorded in the same action. Before extending credit, the app shows the retailer’s outstanding balance, so the decision to deliver on terms is made with the account in view rather than on trust. This is where van sales software earns its keep: the till, the ledger and the price book all ride along, so the distributor’s cash cycle shrinks along with the paperwork.

Returns and end-of-day reconciliation

Back at the depot, the driver declares unsold stock, damages and expiries with a reason code, and the system closes the loop: load-out minus sales minus returns should equal what is physically back on the van, and the cash collected should match the day’s invoices. Discrepancies are flagged the same evening, while the route is still fresh. That is the entire promise of DSD route accounting — every unit of stock has a documented life from the depot shelf to the retailer’s counter and back.

Van Sales vs Pre-Sales: Which Model Fits Which Product Line?

Most FMCG distributors run both models side by side, and the right question is not which is better in the abstract but which fits the product’s velocity and the retailer’s buying pattern.

Dimension	Pre-sales (book today, deliver tomorrow)	Van sales (sell, deliver and invoice in one visit)
Best for	Larger baskets, slower-moving SKUs, planned purchases	Perishables and high-turnover lines: dairy, bakery, beverages, snacks
Stock visibility	Warehouse-held until delivery	Van-held all day; live stock is the rep’s only truth
Cash cycle	Invoiced at delivery, often on terms	Frequently settled on the spot
Retailer experience	Predictable, but misses impulse and top-up demand	Responsive to what the shelf actually needs today
Control risk	Order-to-delivery gap can be audited centrally	Stock, cash and credit all live on the road
Software implication	SFA order capture plus delivery scheduling	Full route-day control: load-out, doorstep sale, returns, settlement

Because many routes are hybrids, the practical requirement is that one van sales software platform handles both motions from the same customer, stock and scheme master data — otherwise the van’s numbers and the warehouse’s numbers drift apart by design.

Which Van Sales Software Features Keep the Route Honest?

Feature lists all start to sound alike, so it is more useful to ask which controls specifically prevent the ways a route leaks. Six earn their place:

• Real-time van stock. The van’s ledger updates with every sale and return, so a rep three stops later cannot sell cases that left the vehicle an hour ago.
• Offline-first operation. Routes cross basements, rural roads and dead zones; the sale, invoice and return must all work with no signal and sync cleanly when it returns, with no duplicate entries.
• Credit visibility at the point of decision. The outstanding balance appears before credit is extended, and limits are enforced by the system rather than by the rep’s memory.
• Returns with reason codes. Damaged, expired and refused stock each carries a code and a disposition, so a route with an unusual damage pattern surfaces in a dashboard instead of in a rumour.
• Proof the visit happened. GPS-stamped check-ins paired with a photo of the actual shelf — and, importantly, a check that the photo was taken at the store rather than re-shot from an old image. Replay detection is the control that keeps the whole dataset trustworthy, because a falsified visit poisons every number built on it.
• One stock ledger from van to warehouse. The van is a stock location in the same ledger the depot runs on, not a parallel spreadsheet that gets pasted in at night. Inventory accuracy is a solved problem when the data is captured continuously — large operators using continuous machine checks have cut phantom-stock incidents dramatically compared with periodic manual counts (PYMNTS on AI inventory accuracy, June 2026) — but it only holds if every moving location posts to the same ledger.

How to Choose Van Sales Software for a Distribution Network

Selection projects go wrong when they evaluate screens instead of route days. A four-step sequence keeps the evaluation grounded.

Step 1: Map the route day before the demo

Write down how a van actually runs today: how the load-out is picked, what the rep does at a cash retailer versus a credit retailer, how returns come back, when reconciliation happens. Score every candidate against that real sequence rather than against a feature matrix, because the gaps that matter are always in the handoffs.

Step 2: Demand one stock ledger from van to warehouse

Ask specifically whether the van is a stock location in the same ledger as the depot, or a separate module that syncs. The first answer survives a mid-day stock query; the second produces two versions of the truth exactly when someone needs one number.

Step 3: Test offline in your worst network area

Run the pilot on the route with the worst connectivity, not the best. If the app can complete a full selling day offline — sales, invoices, returns, credit checks — and reconcile without duplicates, it will survive everywhere else. A demo on office Wi-Fi proves nothing about a rural beat.

Step 4: Pilot two vans before a fleet rollout

Instrument the pilot on measurable questions: does the evening reconciliation balance on the first pass, how long settlement takes, whether returns carry usable reason codes. Two vans for a few weeks will tell you more than a twelve-slide evaluation, and the same two vans become the reference route for training the rest of the fleet.

How eBest Runs Van Sales Inside One Route-to-Market Platform

eBest treats van sales software as part of the route-to-market platform rather than a bolt-on delivery app, which is what makes the controls above enforceable. Van sales and DSD execution runs on the same stock ledger, customer master and scheme engine as the rest of the network, so the van’s day reconciles against the depot’s books automatically. Around that core, several AI capabilities do specific, nameable work: AI Order Recognition reads courier slips and order documents and validates them against the shift address and date, while AI Invoice Recognition extracts the tax ID, date and amount from a photographed invoice so doorstep settlements stop relying on retyped numbers. AI Warehouse Audit extends the same photo-based recognition to stock itself — product types, quantities and share recognised from warehouse images, generating an audit report that keeps distributor and van-side stock claims honest.

The sales side gets support as well. AI Selling Story recommends what to pitch at each store based on its profile and local best-sellers, and hands the rep a ready talk-track for the doorstep conversation. Meanwhile replay detection runs on every visit photo, flagging a re-shot image before it can masquerade as today’s visit — the quiet control that makes the route’s data worth trusting. This all sits inside the same SFA platform field teams already use for visits, with distributor management keeping primary and secondary sales in one view. Brands running this stack include Coca-Cola, whose UAE field organisation went live on eBest SFA across six sub-teams from day one, alongside Carlsberg, Nestlé, Unilever, P&G, Mars and Danone — the customer results library shows how those rollouts came together.

Frequently Asked Questions

How does van sales work in FMCG distribution?

The van is loaded with stock at the depot in the morning, the driver-salesman follows a planned route, and at each retailer the sale, delivery and invoice all happen in a single visit — often with cash collected on the spot. At day’s end, unsold stock, damages and collections are reconciled against the load-out. Van sales suits perishable, high-turnover categories where retailers need frequent small top-ups rather than large pre-booked orders.

What is the difference between van sales and pre-sales?

In pre-sales, a rep books an order today and a separate delivery fulfils it later from the warehouse; the stock stays centralised until that delivery. In van sales, the vehicle carries the stock, so selling, delivering, invoicing and collection all happen at once. Pre-sales fits larger baskets and planned purchases; van sales fits impulse-driven top-ups and cash-and-carry trade. Most distributors run both, which is why good van sales software keeps both motions on shared master data.

How do you control stock on a van?

Treat the van as a stock location in the same ledger as the warehouse. Record the load-out as a transaction, deduct every sale and return in real time, and reconcile load-out minus sales minus returns against the physical stock when the van returns. Real-time deduction and same-evening reconciliation are the two controls that turn van stock from an estimate into an auditable number.

What features should a van sales app have?

Good van sales software covers the non-negotiables first: real-time van stock, offline-first operation with clean sync, on-the-spot invoicing with scheme pricing, credit limits visible before delivery, returns captured with reason codes, GPS-stamped visit proof, and one stock ledger shared with the depot. Anything else — route planning, analytics, promotion tracking — sits on top of those foundations.

Can van sales software work without internet connectivity?

Yes, and it must. Routes routinely cross areas with weak signal, so the app has to complete sales, invoices, returns and credit checks fully offline and sync automatically when connectivity returns. The test is not whether the screens work offline but whether the evening reconciliation balances without duplicates or lost entries after a full offline day.

A van sales operation does not fail because reps stop working; it fails because the route day leaves no evidence behind. Van sales software that captures the load-out, the doorstep sale, the return and the settlement as they happen turns that evidence into a route you can actually manage — and the AI checks on top of it decide whether the numbers deserve trust. Pick two vans, map their day honestly, and run the pilot where the signal is worst. Talk to the eBest route-to-market team about putting those controls on your own fleet.

Conversational analytics lets a CPG sales leader ask their field data a question in plain language — why did shelf compliance drop in that region, which distributors are sitting on ageing stock, which promotion produced orders but no repeat purchase — and get an answer in seconds, without building a report or waiting in an analyst’s queue. In a route-to-market business, that shifts analytics from a monthly ritual to a habit that happens between store visits.

Why CPG Sales Data Hides Behind Dashboards

A CPG field organisation produces an unreasonable amount of data. Every visit logs a check-in, a shelf photo, an order, a competitor note, a display confirmation. Distributors report stock that changes daily. Trade promotions generate spend records that need reconciling against what actually sold. The data is not the problem. The problem is that almost none of it is askable.

Traditional business intelligence answers the questions somebody anticipated. A dashboard shows visit completion by region, sales by SKU, coverage by route — because someone decided last quarter that those were the numbers worth building. The moment a real question appears — a territory manager sees a decline and wants to know which stores, which SKUs, which reps are driving it — the dashboard goes quiet. The question goes into a ticket queue, an analyst pulls a extract, and the answer arrives next week, attached to a decision that was already made.

That lag has a cost nobody books. Gartner’s analysts put it bluntly in their 2026 data and analytics predictions: the boundary between the person asking and the system answering is dissolving, and organisations that keep insights locked behind specialist workflows are choosing slower decisions by default (Gartner D&A predictions, March 2026). Meanwhile the BI industry has converged on the same conclusion from the product side — conversational interfaces have become the primary way non-technical users get answers from governed data, with natural language query now a standard expectation rather than a novelty (ThoughtSpot 2026 BI trends).

For a CPG sales organisation, the interesting question is not whether this shift happens. It is what a conversational analytics layer has to be connected to before its answers mean anything.

How Does Conversational Analytics Work for a CPG Field Team?

The mechanics are simpler than the marketing suggests, and the details that matter are unglamorous. A working loop has three motions: the question becomes a query, the answer arrives with its provenance attached, and the answer turns into something a rep can actually do.

From question to query, without SQL

Under the hood, the system translates the question into a structured query against the platform’s own data — what engineers call text-to-SQL. The manager types or speaks “show me outlets where the display was agreed but not delivered this cycle”, and the system resolves outlets, display agreements and delivery status into the actual tables where that evidence lives. Done well, the manager never learns a schema, a filter convention or a chart builder. The barrier drops from “learn this BI tool” to “describe what you need”, which is exactly the capability gap that has kept field analytics in the hands of a few analysts.

From answer to action, not just to a chart

An answer that ends in a chart still leaves the hardest step to the reader. This is the difference between generic BI chatbots and a conversational analytics capability that lives inside the route-to-market platform: the diagnosis can hand off to the execution layer. When the answer shows that a KPI is slipping — coverage down in one district, lines per store flat in another — the system can generate the improvement suggestion attached to that diagnosis, so the territory manager leaves the conversation with a next action rather than a screenshot. And when the action is a sales conversation, the platform can go one step further and shape the pitch the rep delivers at that specific outlet.

From data to trust, because provenance is visible

The third motion is the one buyers forget to test. An answer is only useful if the manager can see what it was built from — which visits, which shelf photos, which order records, over which date range. Analytics grounded in the visit itself, where every shelf claim traces back to a photo the rep actually took, answers the sceptic’s question before it is asked.

Conversational Analytics vs Traditional Dashboards: Which Answers What?

The honest answer is both, for different jobs. Dashboards remain the right tool for the recurring, standardised view — the weekly review, the leaderboard, the month-end pack. Where conversational analytics changes the game is the long tail: the unanticipated questions that make up most of a manager’s actual thinking, and that no dashboard roadmap will ever pre-build.

Dimension	Spreadsheets	Traditional dashboards	Conversational analytics inside the RTM platform
Who can get an answer	The person who built the file	Whoever the dashboard was built for	Anyone who can describe the question
Unanticipated questions	Possible, slow, manual	Only if pre-built	First-class use case
Time to answer	Hours to days	Instant, but fixed questions	Seconds, any question
Follow-up questions	Restart the analysis	New dashboard request	Natural continuation of the same thread
Data provenance	Wherever the copy-paste stopped	The configured data source	Traces to the visit: photos, orders, check-ins
Reaches distributor teams	Rarely	If licensed separately	Yes, when they use the same platform
Turns into rep action	No	No	Diagnosis can attach a suggestion, then a pitch

The last row is where the category either earns its place or does not. An answer that stops at insight is a faster report. An answer that continues into the rep’s next store visit is an operating habit.

Can Sales Managers Trust Answers They Did Not Build?

Healthy scepticism is the right starting position, because a natural language interface can produce a confident, fluent, wrong answer just as easily as a right one. Trust in conversational analytics is not a feeling; it is a set of properties you can demand and verify.

The first property is a governed data model. The system needs one agreed definition of “coverage”, “shelf share” or “active outlet”, so the same question returns the same answer for every manager who asks it. Without that semantic layer, two people asking the identical question can get two different truths, and the tool loses credibility in a week.

The second is provenance on every answer. In a field context this is a genuine advantage, because the evidence chain is unusually concrete: the shelf claim comes from a photo the system already validated, the order from a confirmed transaction, the visit from a check-in with a timestamp. When a manager can drill from the answer to the store, to the visit, to the image, the conversation shifts from “do I believe the AI” to “what do we do about it”.

The third is access control that travels with the question. A regional manager should get answers scoped to their region, and finance-scoped questions should stay out of the field app entirely. Conversational convenience without permissioning is a data leak with a friendly interface.

How to Deploy Conversational Analytics in a CPG Sales Organisation

Deployment succeeds or fails on the first sixty days, and the pattern that works is deliberately narrow. Start smaller than feels ambitious, prove trust on questions people actually ask, then let usage pull the expansion.

Step 1: Pick the questions that already repeat

Collect the recurring questions from weekly reviews and manager chats for two weeks. Deploy against that list first — coverage, stock ageing, promotion follow-through. A pilot measured on questions people genuinely ask beats a pilot measured on feature completeness, because adoption follows usefulness rather than the reverse.

Step 2: Ground the answers in execution data, not summaries

Connect the conversational layer to the platform where the evidence lives — the visit workflow, the shelf photos, the orders, the distributor stock feeds. Answers built on second-hand extracts inherit every error in the extract; answers built on the visit itself can be traced to a store in two taps.

Step 3: Set the semantic definitions once, centrally

Agree what “active outlet”, “coverage” and “compliance” mean, encode those definitions in the data model, and resist the temptation to let each region customise. This is the least glamorous step in the project and the one that most often determines whether managers trust the answers by day sixty.

Step 4: Wire the answer to the action it implies

Close the loop before scaling. When the diagnosis says coverage is slipping, the suggestion should attach to the diagnosis and reach the rep’s plan. When the insight is a sell-in opportunity, the pitch should reach the rep before the visit. Conversational analytics that ends at the chart is a dashboard with extra steps.

How eBest Puts Conversational Analytics Inside the Route-to-Market Platform

eBest treats the question box as part of the field platform rather than a bolt-on analytics product, which is what makes the answers specific. AI Chat Report is a natural-language, text-to-SQL capability over the store-execution data the platform already collects — a sales director asks which skill gap is costing execution in which region, or which outlets stopped ordering a SKU after the promotion ended, and gets the answer in seconds instead of a quarterly review. AI KPI Suggestion sits one step behind the answer: when the diagnosis shows a KPI lagging, it generates the improvement action attached to that finding, so the conversation ends with a plan rather than a screenshot.

The loop then reaches the shelf itself. Because AI Image Recognition validates what is on the shelf from the rep’s photo — facings, shelf share, display compliance — an answer about execution traces to evidence the system has already checked, not to a rep’s self-report. And when the data says a specific outlet is ready for a wider range, AI Selling Story turns that insight into the pitch the rep delivers at the door. Ask, diagnose, act, verify — one motion instead of four tools.

This sits inside the same platform that runs the visit. The eBest SFA platform produces the visit and execution data the questions are asked against, trade promotion management supplies the promotion context behind every follow-through question, and for brands selling through partners, distributor management system data brings distributor stock into the same conversation. Deployments include Coca-Cola, whose UAE field organisation went live on eBest SFA across six sub-teams from day one, alongside Carlsberg, Nestlé, Unilever, P&G, Mars and Danone. The eBest customer results library shows how those rollouts came together.

Frequently Asked Questions

What is conversational analytics?

Conversational analytics is a way of interacting with business data through natural language questions instead of pre-built reports. The system interprets the question, generates the underlying query, returns the answer with context, and supports follow-up questions in the same thread. In a CPG field context, the questions are about execution — visits, shelf conditions, orders, promotions, distributor stock — and the answers draw on the data those activities produce.

How is conversational analytics different from a dashboard?

A dashboard answers the questions somebody anticipated when they built it, instantly and consistently. Conversational analytics answers the questions nobody anticipated — which is most of what managers actually need to know. The two are complements: dashboards for the recurring weekly view, conversation for the long tail of specific, situational questions, each asked once or twice and never worth building a report for.

Does conversational analytics require SQL or data skills?

No. The point of the natural language layer is that describing the question replaces building the query. A territory manager asks in plain language and the system handles the translation into a structured query. What the organisation does need is a governed data model behind the interface, so that “coverage” or “active outlet” means the same thing for every person who asks.

Can conversational analytics give wrong answers?

Yes, and any honest vendor will say so. The defences are structural: a governed semantic layer with one agreed definition per metric, answers that carry their provenance so a manager can trace from the number to the store, the visit and the photo, and permissioning that scopes what each role can ask. With those in place, the error surfaces as a traceable disagreement rather than a confident fiction.

What should a CPG company look for in conversational analytics software?

Three things beyond the chat interface. First, grounding in the execution data itself — visit records, shelf photos, orders, distributor stock — not in second-hand extracts. Second, a closed loop from answer to action, so a diagnosis can attach an improvement suggestion and reach the rep’s plan. Third, shared platform access for distributor teams, because in most emerging markets the people standing in front of the shopkeeper do not appear on your payroll.

Most CPG sales organisations do not lack data; they lack a way to ask it a question at the moment the question matters. That is the whole promise of conversational analytics, and the reason it belongs inside the route-to-market platform rather than beside it. The practical next step is small: pick the five questions your weekly review keeps circling, point a conversational layer at the data that answers them, and measure how much faster the decision follows the question. Talk to the eBest route-to-market team about running that pilot on your own field data.

SFA software for CPG companies is a mobile-first field-sales platform built for route-to-market, not account management. Where a generic CRM tracks deals, CPG SFA software directs store coverage, captures perfect-store execution in the field, and routes reps through thousands of outlets a week. The strongest deployments add AI sell-in suggestion, AI route optimization, and perfect store image recognition so a rep walks into each visit knowing exactly what to pitch and where to go next.

Why CPG Companies Need SFA Software for CPG Companies

Most consumer-goods brands lose more sales to broken execution than to weak strategy. A planogram gets agreed at headquarters, yet the shelf in a tier-3 city tells a different story. That gap is the whole reason SFA software for CPG companies exists. It turns the field force from a group of visitors into a measured, coachable system.

Consider the weekly reality of a beverage or snacks brand. A single key-account manager may own 40 hypermarkets; a van-sales driver may cover 60 traditional trade outlets a day. They don’t need a CRM that logs a “lead.” They need a tool that tells them which stores to visit, what to check when they arrive, and what to sell based on what that specific store actually buys. Off-the-shelf sales tools miss this, because they were designed for B2B deal cycles, not for repetitive, high-frequency store coverage.

That is the core argument for purpose-built SFA software for CPG companies: it treats outlet coverage as the primary unit of work. Visits are scheduled by route and priority, not by whoever picked up the phone. Execution data — share of shelf, POSM presence, stock status — is captured at the shelf, not reconstructed later in a spreadsheet. And because the data lands in near real time, a regional manager can spot a failing outlet before the month closes, not after.

The payoff is operational, not cosmetic. When execution is visible, coaching becomes specific. A rep who skips the frozen-aisle secondary display shows up in the data; a territory with slipping perfect-store scores gets attention while it is still fixable. NielsenIQ’s retail-execution research consistently ties on-shelf availability to lost sales at the moment of purchase, which is exactly the failure CPG SFA software is built to prevent (NielsenIQ insights).

What Makes SFA Software for CPG Companies Different from a CRM?

This is the question that quietly decides most buying mistakes. A CRM answers “who are my prospects and where is the deal?” CPG SFA answers “which stores did we cover, what did we find, and what did we sell?” The objects are different: outlet instead of account, visit instead of opportunity, execution instead of pipeline.

Because the unit of work is the store visit, the data model has to support things a CRM never touches — van stock, secondary placements, planogram compliance, and distributor reconciliation. If your tool can’t model a traditional-trade outlet that has no fixed buyer and no email address, it isn’t SFA software for CPG companies. It’s a contact manager wearing a field-sales costume.

How Does SFA Software for CPG Companies Differ from a Generic CRM?

The difference shows up the first time your team works in a store with patchy signal. Generic platforms assume a connected laptop and a calm Wi-Fi session. CPG field work assumes a crowded aisle, a dead zone, and a rep who has thirty seconds before the store owner loses patience. That single assumption — connectivity — splits the two product categories apart.

Take architecture first. True SFA software for CPG companies is offline-first, not merely offline-capable. The rep opens the app on a metro train, reviews the day’s route, captures a shelf photo, and syncs the moment a signal returns. A generic CRM that “works offline” usually means a cached view that breaks the moment the rep tries something the cache didn’t anticipate. In emerging markets — India, Indonesia, Kenya, the Middle East — that gap is the difference between adoption and quiet abandonment.

Then there’s the route-to-market lens. CPG selling is repetition at scale: the same 200 outlets, visited on a rhythm, with execution checked every time. A generic CRM optimizes for the rare, high-value negotiation. CPG SFA optimizes for the daily grind of coverage, so the system pushes the next best store, not the next big deal. Gartner’s sales-technology coverage repeatedly frames field-sales execution as a distinct discipline from pipeline CRM, which is why the two are evaluated on different criteria (Gartner sales technology).

Offline-First Architecture for Emerging Markets

Connectivity is not a nice-to-have in CPG field work; it is the first thing that breaks. Offline-first means the rep’s full working set — outlet list, visit history, product catalog, and capture forms — lives on the device. Sync is background and conflict-aware, so two reps editing the same outlet don’t clobber each other. Brands running rural or peri-urban coverage cannot treat this as optional.

Route to Market, Not Account Management

The second differentiator is what the system optimizes. CPG SFA software sequences visits by coverage need, promotional calendar, and outlet priority. It asks “which stores are due, and in what order, to protect distribution?” A CRM asks “which opportunity is closest to closing?” Those are different questions with different daily behaviors. Choose the tool that asks the CPG question.

Perfect Store Execution Built In

Third, execution has to be captured where it happens. Photograph the shelf, and the system should score share of shelf and planogram compliance on the spot. This is perfect store image recognition — a named capability, not a vague “AI module.” When the rep sees the gap immediately, the correction happens during the visit, not in next month’s review.

What AI Capabilities Should SFA Software for CPG Companies Include?

AI is where most vendors now over-promise and under-deliver. The fix is to name the capability and the workflow it touches. For SFA software for CPG companies, four capabilities actually move execution, and each maps to a moment in the rep’s day.

First, AI sell-in suggestion works at the start of a visit. The system reads the store’s profile and local best-sellers, then recommends which products to pitch and generates a tailored selling talk-track. The rep sells smarter in the first ninety seconds, instead of guessing.

Second, AI route optimization sequences the day. It weighs outlet priority, distance, visit frequency, and live conditions to build a route that protects coverage without burning the rep’s hours on the road. McKinsey’s consumer-goods work on route-to-market points to exactly this kind of optimization as a lever for both cost and reach (McKinsey CPG insights).

Third, perfect store image recognition turns a shelf photo into structured execution data — facings, share of shelf, POSM presence, and special displays — with a human-in-the-loop appeal when the AI misses. Fourth, AI Chat Report lets a manager ask a plain-language question and get a store-execution answer back, no BI training required. Stack those four and the field force becomes measurably sharper without adding headcount.

How to Choose and Deploy SFA Software for CPG Companies

Choosing well is less about feature lists and more about matching the tool to your route-to-market reality. The framework below keeps the decision grounded, then turns it into a deployment sequence most brands can run in a single quarter.

Start by mapping the actual work. How many outlets, across which channels, served by which model — direct, distributor, or van? A tool that fits a 5,000-outlet direct force will choke a distributor-led market, and vice versa. Next, shortlist only vendors built for CPG field execution, and reject any demo that leans on B2B-CRM language. Then pilot offline-first in one region before you sign a scaling commitment. Finally, switch on the AI capabilities in order — sell-in suggestion and route optimization first, because reps feel them on day one, then perfect-store recognition and conversational reporting once the data is flowing.

The deployment itself follows five steps. Treat them as a sequence, not a menu.

1. Map your route-to-market reality. List outlets, channels, and the selling model (direct, distributor, van) so the platform matches how you actually go to market.
2. Shortlist CPG-native vendors. Remove any tool that frames the field force as “leads” — keep only those built for store coverage and execution.
3. Run an offline-first pilot in one region. Prove the app works in dead zones before you bet the entire force on it.
4. Turn on AI capabilities in order. Start with AI sell-in suggestion and AI route optimization, then add perfect store image recognition and AI Chat Report.
5. Scale with distributor and promotion integration. Connect DMS and trade promotion management so execution and incentives stay in sync.

How eBest Solves This

eBest builds SFA software for CPG companies as part of one unified route-to-market stack — SFA, DMS, TPM, DSD, and B2B on a single platform, rather than bolted-on modules. That matters because execution problems rarely stay inside one system: a missed display is a promotion problem, a stockout is a distributor problem, and a weak pitch is a training problem.

The real proof is in live rollouts. Coca-Cola’s UAE SFA deployment went live across six sub-teams with full day-one adoption, a story documented in our Coca-Cola SFA rollout. Global brands such as Nestlé and Unilever run eBest across complex, distributor-heavy markets where offline-first coverage is non-negotiable. Under the hood, the same AI capabilities named above — AI sell-in suggestion, AI route optimization, and perfect store image recognition — sit inside the SFA layer, while a distributor management system and trade promotion management software keep the rest of the chain connected. The result is fewer silos and a field force that executes the plan the company actually wrote.

Frequently Asked Questions

Q1: What is the best SFA software for CPG companies?

The best SFA software for CPG companies is the one built around store coverage and execution, not deal pipelines. Look for offline-first mobile capture, route-to-market sequencing, perfect-store execution, and distributor integration out of the box. AI sell-in suggestion and AI route optimization are strong differentiators because reps feel them immediately. Avoid any platform that frames your outlets as “leads” — that signals a CRM wearing a field-sales costume. Validate with a real pilot in a low-connectivity region before you commit, since that is where most generic tools quietly fail.

Q2: How long does it take to deploy SFA software for CPG companies?

Most brands can run a focused deployment inside a single quarter. The sequence is straightforward: map your route-to-market, shortlist CPG-native vendors, pilot offline-first in one region, switch on AI capabilities in order, then scale with distributor and promotion integration. The pilot is the critical gate — prove the app works in dead zones first. A unified stack like eBest’s shortens the timeline because SFA, DMS, and TPM already share data models, so you are not integrating three separate systems by hand.

Q3: Which AI capabilities does eBest include in its CPG SFA software?

eBest’s SFA layer includes several named, workflow-specific AI capabilities rather than one vague “AI module.” AI sell-in suggestion recommends products to pitch and generates a tailored talk-track per store. AI route optimization sequences the day around coverage and priority. Perfect store image recognition turns a shelf photo into structured share-of-shelf and planogram data, with a human appeal when the AI misses. AI Chat Report lets managers ask plain-language questions and get store-execution answers back. Each capability maps to a real moment in the rep’s day.

Q4: Is SFA software for CPG companies useful for small or traditional-trade markets?

Yes, and often it matters more there than in modern trade. Traditional-trade outlets — small groceries, baqalas, kiosks — have no fixed buyer and no email, so execution visibility depends entirely on what the rep captures at the shelf. Offline-first architecture is essential, because connectivity is unreliable. AI route optimization helps a van-sales driver cover more outlets per day, and perfect store image recognition flags missing displays without a supervisor physically visiting. The brands that win traditional trade are usually the ones that made field execution measurable first.

Q5: How is SFA software for CPG companies different from a generic CRM?

The core difference is the unit of work. A CRM optimizes for the deal: who is the prospect, where is the opportunity, when will it close. SFA software for CPG companies optimizes for the store visit: which outlets are due, what to check on arrival, and what to sell based on that store’s history. The data model reflects this — outlets, visits, van stock, and planogram compliance instead of accounts, pipelines, and leads. If your tool cannot model an outlet with no email address, it is a CRM, not CPG SFA.

Ready to see what purpose-built SFA software for CPG companies looks like in the field? Explore the eBest SFA platform or talk to our route-to-market team about a pilot in your toughest region.

AI sales training uses a simulated buyer — and, in consumer goods, a simulated store situation — to let a field rep rehearse the exact conversation they are about to have, then scores the attempt and returns coaching within minutes. For a route-to-market team it is less an HR initiative than a coverage fix: the months a new rep spends learning on live stores get compressed, because practice no longer has to wait for a manager to ride along.

Why Field Sales Training Breaks at CPG Scale

Here is the thing most enablement programmes get wrong. They import a training model built for people who sit at desks, and then wonder why it evaporates in an aisle. An inside sales team can rehearse a call, record it, and have a manager dissect it thirty seconds later. A CPG field rep makes the decision standing up, in a store, in front of a person who has thirty seconds and a shelf that needs facing. There is no recording, no debrief, and usually no one watching at all.

That asymmetry is why field sales training fails in a specific and predictable way. The knowledge transfers. The judgement does not.

So the cost lands where it is hardest to see. According to SPOTIO’s 2026 State of Field Sales survey, 44% of field sales teams take three or more months to ramp a new hire to full productivity, and on high-turnover teams only 47% of new reps reach full productivity within two months, against 70% on low-turnover teams (SPOTIO field sales research). The same data shows field reps spend only 43% of their time actually selling — the rest goes to travel, admin and preparation.

Read those two numbers together and the problem sharpens. A rep who is not yet productive is still consuming a territory. Coverage is being paid for and not delivered, and in markets where a single rep carries hundreds of outlets, an under-ramped rep is not a personal performance issue — it is a distribution gap.

Meanwhile, the macro case for fixing this has become unusually well documented. In a May 2026 study of 227 chief sales officers, Gartner found that organisations providing sellers with AI-enabled next best actions are 2.6 times more likely to achieve commercial growth, and — more pointedly for this discussion — that organisations prioritising upskilling sellers on AI are 2.4 times more likely to achieve strong revenue growth (Gartner sales research, May 2026). That second multiplier is the one worth pinning above a capability manager’s desk. The tool is not the investment. The person using the tool is.

The Ramp Gap Is a Coverage Problem, Not an HR Problem

Most brands file sales onboarding under HR, and then measure it with completion rates. That framing is why the problem survives. A completion rate tells you that a rep sat through the module; it says nothing about whether that rep can walk into a store, read the situation, and recover when the store owner pushes back on a promotion.

Reframe it as coverage and the arithmetic changes. If a territory carries 300 outlets and a rep reaches full productivity in month four instead of month two, the brand has absorbed two months of partial coverage across 300 outlets. No training completion dashboard will ever show that number, because the loss is an absence rather than an event.

Gartner’s analysts make the same argument from the enablement side: value has moved away from static content and periodic sessions, toward guidance that reaches the seller in the workflow where the work actually happens (Gartner enablement research, April 2026). For a field team, that workflow is the visit — not a learning portal somewhere else.

How Does AI Sales Training Work for a Field Rep?

The mechanics are less exotic than the category name suggests, and they matter more than the model. A working AI sales training loop has four motions: seed the scenario, rehearse the conversation, score it against execution criteria, and repeat until the behaviour holds. What separates a field-ready implementation from a demo is where the scenario comes from.

Pick the scenario from the outlet, not from a persona

Generic role-play products ask you to configure a buyer persona — a sceptical economic buyer, a procurement lead, a VP of Finance. That works when your seller is on a call with a named buyer whose role you know in advance. It falls apart in traditional trade, where the “persona” is a store owner whose objection is not about budget cycles but about whether the promotion is worth the shelf space and whether the last delivery arrived on time.

When the scenario is sourced from the outlet instead, the rep practises against their own territory. So the practice object becomes the shelf photo from last visit, the outlet’s own order history, the SKU that has been quietly declining, and the display commitment that was agreed and never delivered. That is not a simulation of selling. It is a rehearsal of Tuesday.

Rehearse the exact pitch, then get scored on it

This is where a training capability stops being standalone. Once the system already generates a store-specific recommendation and pitch, the rep can rehearse that pitch before walking in, and the AI plays the store owner rather than a generic gatekeeper. For a beverage or dairy route, the objections that matter are practical ones: not enough space, stock still on hand, the price of the promotion, the display that blocks the aisle.

Scoring is the part buyers underweight. A scorecard built on talk-to-listen ratio is nearly useless for a rep who never makes a call. What a field capability manager needs is scoring against things the business already tracks — did the rep open with the display opportunity, did they handle the space objection, did they close with a specific order. Then coaching points to a named behaviour rather than a general instruction to improve.

Repeat on the rep’s own time, not on the manager’s calendar

The structural constraint in field coaching has never been willingness. It is arithmetic. A manager with fifteen reps and thirty minutes each spends more than seven hours a week coaching, and that is the first thing sacrificed when the quarter tightens — which is exactly when coaching matters most.

On-demand practice removes the scheduling dependency. Because sessions are repeatable and scored consistently, a new rep can run the same store scenario five times over a weekend and watch the score move. Mindtickle’s 2026 enablement research points the same direction, finding that top-selling reps complete roughly twice as many practice repetitions as the rest of the team. Repetition, not intensity, is the variable that moves.

AI Role-Play Sales Training vs Traditional Coaching: What Actually Transfers?

There is a fair criticism of this category, and it deserves an answer rather than a deflection: reps can get good at performing for a bot without getting better at selling to a person. That risk is real whenever the simulated buyer is invented. It shrinks sharply when the scenario is drawn from the rep’s own outlet data and the scoring rules are the same ones the business uses to judge execution.

Dimension	Ride-along coaching	Desk-based role-play tools	AI sales training inside the RTM platform
Scenario source	Whatever store the day produces	Configured buyer personas	The rep’s own outlet: shelf history, order data, agreed displays
Availability	Manager calendar only	On demand	On demand, on the phone already in the rep’s pocket
Repetitions per rep	A handful per cycle	Unlimited	Unlimited, between visits
Scoring basis	Manager memory, subjective	Generic rubric: talk ratio, filler words, pacing	Execution criteria the business already tracks
Relevance to traditional trade	Depends on the route that day	Low — modelled on call-based selling	High — store owner objections, space, displays, price
Reaches distributor salesmen	Rarely	No	Yes, when they are users of the same platform
Consistency across regions	Varies by manager	Consistent	Consistent, in local language
Follow-through to a live visit	Verbal	None	The rehearsed pitch is the pitch that gets delivered

The last row is the one that decides renewal. Training that ends when the session ends is a content library with a score attached. Training that feeds directly into the conversation the rep is about to have is part of the operating system.

Can AI Sales Training Reach Distributor Salesmen You Do Not Employ?

In most emerging markets the answer determines whether a training programme has any reach at all. A brand may directly employ a few hundred field reps and sell through distributors who employ thousands of salesmen. Those salesmen are the ones standing in front of the shopkeeper, and they are outside the reach of any HR-owned training platform.

This is where an AI sales training capability has to be judged on where it lives. If it exists inside the platform that distributors already use to book orders and log visits, coverage follows the channel rather than the payroll. A distributor salesman opens the same app, sees the same product story, and practises the same objection handling as a brand-employed rep. Nothing extra to licence, nothing to integrate.

Practically, three requirements make or break that reach. First, it has to work on a mid-range Android phone, because that is the device in the market. Second, it has to function offline or near-offline, since rural routes do not wait for signal. Third, it has to operate in the local language, because objection handling practised in a second language does not survive a real store.

Get those three right, and training stops being an HR programme for the headcount you control and becomes a capability layer across the whole route to market. Get them wrong, and the distributor’s team learns by trial and error in front of your customers — which is how inconsistent pricing, missed displays and quiet SKU declines begin.

How eBest Puts AI Sales Training Inside the Route-to-Market Platform

eBest treats rep capability as part of the field platform rather than a separate learning system, which is what makes the scenarios specific instead of generic. The scenario simulation pod covers real sales situations — a pharmacy manager raising an objection, a product pitch to a key account, a display negotiation in a convenience store — and the rep role-plays the conversation while the system scores the attempt and pushes targeted improvement points. AI Training & Exam adds a knowledge base with adaptive learning, so each rep’s path is shaped by what they actually get wrong rather than by a fixed syllabus.

Two capabilities make the practice loop grounded. AI Selling Story recommends what to pitch for a specific outlet based on its profile and local best-sellers, then generates the pitch itself — which means the scenario the rep rehearses is the pitch they are about to deliver, not a textbook script. AI Image Recognition turns the shelf into the scenario object: facings, shelf share and display compliance from a photo, so the conversation being practised starts from what is genuinely on the shelf. Then AI Chat Report lets a sales director ask, in plain language, which skill gap is costing execution in which region, instead of waiting for a quarterly capability review.

This sits inside the same platform that runs the visit. The eBest SFA platform carries the visit workflow the rep practises for, trade promotion management supplies the promotion the rep has to re-sell at the shelf, and for brands working through partners, distributor management system data is what brings distributor salesmen into the same practice loop. Deployments include Coca-Cola, whose UAE field organisation went live on eBest SFA across six sub-teams from day one, alongside Carlsberg, Nestlé, Unilever, P&G, Mars and Danone. You can see how those rollouts came together in the eBest customer results library.

Frequently Asked Questions

What is AI sales training?

AI sales training is the use of AI to simulate a selling conversation, let a rep rehearse it repeatedly, and score the result automatically against a defined standard. In a CPG field context the simulated party is a store owner or buyer rather than a phone prospect, and the scenario is drawn from the outlet’s own data — shelf history, order pattern, agreed displays. The practical outcome is that a rep can practise the conversation they are about to have, on demand, without a manager present.

How is AI sales training different from traditional role play?

Traditional role play needs a partner: a manager, a peer, or a classroom. It happens on a schedule, it is scored from memory, and it cannot scale past the number of people available to run it. AI sales training removes the partner dependency, so a rep practises as often as they want, at any hour, and every attempt is scored against the same rubric. The trade-off is scenario quality: practice is only as useful as the situation it puts the rep in.

Does AI sales training work for field reps who never make phone calls?

It does, provided the scenario is field-shaped. Most AI role-play products are built around call-based selling — cold calls, discovery calls, pricing conversations with procurement — so a rep whose job is a thirty-second conversation in a store finds the personas irrelevant. Field-ready AI sales training has to rehearse the objections that actually occur on a route: no shelf space, stock still on hand, promotion pricing, delivery timing and blocked aisles.

How does AI sales training shorten sales onboarding ramp time?

It compresses the part of onboarding that used to require a live store and a supervisor. A new rep can rehearse the standard visit conversation, the objection handling and the display ask before entering a single outlet, and receive a score each time. Because the scoring is consistent, a manager can see which specific behaviour is lagging — the opening, the objection, or the close — rather than waiting for a slow ramp to reveal itself in a quarterly number. The rep arrives at the first visits having already made the mistakes somewhere harmless.

What should a CPG brand demand from a sales training software for CPG companies?

Four things. Scenarios sourced from real outlet data rather than invented personas, so practice transfers. Scoring tied to execution criteria the business already measures, not generic speech metrics. Coverage of the distributor’s salesmen as well as brand-employed reps, or the programme stops at your payroll. And a mobile, offline-capable, local-language experience, because that is the environment the field force actually works in. If a platform cannot answer all four, it is a content library rather than a capability layer.

Every coverage number a sales director watches is downstream of a conversation that either happened well or did not happen. AI sales training matters in consumer goods because it moves the first attempt at that conversation somewhere cheap — before the rep is standing in front of someone with thirty seconds and a shelf that needs facing. The practical next step is not a company-wide rollout; it is one region, one territory, one quarter, measured against ramp time and coverage. Talk to the eBest route-to-market team about running that pilot.

AI shelf monitoring is the practice of using computer vision to read a shelf photo and turn it into on-shelf availability data that a field rep can act on during the same store visit. It exists to close one specific gap: the distance between what a distributor’s system says is in stock and what a shopper can actually reach. Product sitting in the back room, shelf empty, nobody alerted — that is the failure this capability removes.

Why Out-of-Stocks Keep Costing Sales You Already Earned

Most consumer-goods brands do not lose revenue because demand vanished. They lose it because the product was in the building but not on the shelf. Industry research puts the share of SKUs shoppers want that are unavailable at the moment of a visit somewhere between 8% and 12%, and the damage is not limited to one missed basket. A shopper who cannot find their brand buys a competitor’s product, and a meaningful share of them never come back.

The uncomfortable part is where the problem lives. Roughly 60% of out-of-stock incidents trace back to in-store execution rather than a supply chain failure. The truck arrived. The delivery was scanned. The case went into the back room and stayed there. Meanwhile nobody on the brand side had a signal that anything was wrong, because the retailer’s system still showed the SKU as in stock. This is phantom inventory — the perpetual record says eight units, the shelf says zero, and the replenishment engine stays quiet because it trusts the record (NielsenIQ insights).

Promotions make it worse. Promoted items go out of stock at roughly double the normal rate, which means the weeks a brand spends the most money are often the weeks its availability is weakest. That combination is expensive in a way that rarely surfaces in a single report: trade spend is booked, the display is agreed, and the shopper still walks out empty-handed.

So the practical question for a commercial team is not “what is our in-stock rate?” It is “how often could a shopper actually buy this SKU from the shelf?” Those are different numbers, and AI shelf monitoring exists because the second one is the one that converts (Gartner sales research).

McKinsey’s consumer-goods work keeps arriving at the same conclusion from a different direction: the return comes from acting on execution data, not from collecting more of it (McKinsey CPG insights). That is the standard a shelf-monitoring deployment should be judged against.

The Gap Between Warehouse In-Stock and Shelf Availability

Availability leaks at every layer of the chain, and each layer is usually reported in isolation. A distributor can report 98% in-stock while the store sits at 96%, because a delivery was skipped. The store can sit at 96% while the shelf is genuinely buyable only 92% of the time, because product is in the back room instead of faced. Stack those leaks, and a SKU that looks healthy in a warehouse dashboard is available to the shopper far less often than anyone assumed. That final number is the one that decides whether demand converts — and it stays invisible unless something actually looks at the shelf, at shelf level, often.

How Does AI Shelf Monitoring Work in a Store Visit?

The workflow is deliberately short, because it has to fit inside a visit that is already time-boxed. In practice, AI shelf monitoring runs as four connected motions: capture, recognition, correction, and verification. What separates a working deployment from a conference demo is that none of them are deferred to the next visit or to a monthly report.

Step 1: Capture the shelf at the moment of decision

The rep photographs the shelf with the phone already in hand. No extra device, no fixed camera, no retailer sign-off required. Capture has to be near-frictionless, because a workflow that adds thirty seconds per shelf simply will not survive contact with a 60-outlet week. This step is also where data integrity is protected. Replay detection runs on every image, so a photo reused from a previous visit gets flagged before it ever becomes a number in a report.

Step 2: Recognition, scoring, and the honest flag

Once the image lands, recognition does the counting that used to consume the visit: which SKUs are present, how many facings each holds, where the gaps are, and whether the agreed display material is actually up. That is broader than a stock-out check. A capable engine reads the main shelf, point-of-sale materials such as shelf strips and light boxes, and special displays such as floor stacks and end caps, then expresses the result as quantitative shelf share and compliance rather than a subjective note.

The detail that matters most is what happens when the AI is wrong. Shelves are messy, lighting is poor, and packaging changes. A system that presents its output as final quietly erodes trust the first time it miscounts a high-visibility SKU. A system that lets the rep correct the result, and routes that correction into an automated re-check, keeps the field force on side. That is why the appeal path is a feature rather than an afterthought.

Step 3: Close the loop before the rep leaves the store

Detection alone changes nothing, because someone still has to fix the shelf. So the output of recognition has to land as an action inside the same visit — restock from the back room, correct the display, raise the order, or log the exception with photo evidence. This is the step most shelf-intelligence tools leave open, since their pipeline ends at a dashboard rather than at a task. When the loop closes in-store, availability improves the same day instead of the same month. As Gartner frames the shift, value has moved from generating insight to acting on it inside governed workflows (Gartner supply chain research).

Step 4: Verify the fix and learn across visits

A corrected shelf is not a solved shelf. The pattern across visits is what tells a sales director whether an outlet has a chronic replenishment problem, whether a promotion never made it to the floor, or whether a rep is systematically avoiding the back room. Because every visit produces a comparable measurement, AI shelf monitoring builds a trend line instead of a snapshot. In practice, that trend is what turns an availability number into a coaching conversation — and it is why the capability belongs inside the platform that runs the visit, not in a separate audit tool that ships a spreadsheet once a month.

What Should CPG Teams Demand From Shelf Monitoring AI?

Not every tool that calls itself AI shelf monitoring measures the same thing, and the differences decide whether you get continuous visibility or another monthly audit. Compare the three delivery models honestly before you shortlist.

Capability	Manual store audit	Fixed shelf cameras or sensors	AI shelf monitoring on a field device
Coverage	Whatever the visit plan reaches	Only instrumented stores	Every visit, every outlet in the beat
Hardware cost	None	Cameras, sensors, installation, maintenance	None — runs on the rep’s existing phone
Retailer cooperation	Access required	Installation required	Not required
Frequency	Weekly to monthly	Continuous inside the store	Every visit, plus a continuous trend line
Capture integrity	Rep judgement, hard to verify	Sensor-based	Replay detection on every photo step
Handling AI mistakes	Not applicable	Vendor-defined	Rep can correct, and the correction triggers a re-check
Closing the loop	Sometimes, later	Alerts the store team	Action issued during the visit
Analytics access	Export and pivot	Vendor portal only	Ask the data a question in plain language

Run that comparison against your own territories before shortlisting. In many emerging and mid-market geographies the camera column is effectively unavailable, because no retailer will install shelf sensors across thousands of traditional-trade outlets. That does not mean those outlets go unmeasured. It means the sensor is the phone in the rep’s pocket, and the platform has to be good enough to work from a single photograph taken in a crowded aisle.

Three requirements follow. First, the platform must be offline-first, because the shelf does not wait for a signal and a dead zone should not cost you a measurement. Second, correction has to be built in, because uncorrectable AI output stops being trusted within weeks. Third, shelf data has to be queryable in natural language, because an availability problem that needs a BI analyst to explain it is an availability problem that stays unfixed.

How eBest Puts AI Shelf Monitoring to Work

eBest builds AI shelf monitoring into the same platform that runs the visit, which is what makes the correction step possible rather than aspirational. The eBest SFA platform captures the shelf inside the standard visit flow, and AI Image Recognition handles main-shelf, POSM, and special-display checks — facings, shelf share, and compliance counted from the photo. If a count is wrong, the rep raises an appeal and the AI re-checks, so the field force retains ownership of its own data.

Two more capabilities matter for the loop. Replay detection runs on every photo step, including check-in and check-out, so availability figures rest on images that were genuinely taken in the store today. Then trade promotion management connects what the shelf shows to the promotion that was supposed to be running, which means a missing display is caught while the promotion still has time to work. On the analytics side, AI Chat Report lets a regional manager ask a plain-language question about shelf availability and get an answer without exporting a dataset, while AI KPI Suggestion converts a persistent availability gap into a named improvement action instead of a red cell nobody owns.

That combination runs in production with CPG customers including Coca-Cola, whose UAE field organization went live on eBest SFA across six sub-teams from day one, alongside Nestlé, Unilever, P&G, Carlsberg, Mars, and Danone. For brands selling through distributors, distributor management system data keeps secondary sales aligned with shelf reality, so replenishment is not driven by a phantom number. You can see how these deployments come together in the eBest customer results library.

Frequently Asked Questions

What exactly is AI shelf monitoring?

AI shelf monitoring is the use of computer vision to analyze an image of a retail shelf and produce structured availability data — which SKUs are present, how many facings they hold, and where the gaps are. Because the analysis runs on a field rep’s phone, it works in outlets where installing cameras or smart shelves is not realistic. Practically, it replaces the manual counting step of a store audit, so the same information arrives faster, more consistently, and with photo evidence attached to it.

How is AI shelf monitoring different from on-shelf availability?

On-shelf availability is the metric; AI shelf monitoring is the method used to measure and improve it. On-shelf availability expresses the percentage of time a product is physically on the shelf and buyable, with 95% or higher the usual target for key SKUs, and the gap against that target is where lost sales hide. AI shelf monitoring is the mechanism that generates that figure at store and SKU level during normal visits, instead of inferring it from point-of-sale data or waiting for a periodic audit.

Do we need shelf cameras or sensors to run this?

No, and that distinction is often the deciding factor. Camera-and-sensor deployments deliver continuous in-store measurement, yet they require retailer agreement, capital expenditure, and installation in every store you want covered. A phone-based approach treats the visit itself as the measurement event, so coverage follows the field force rather than the hardware budget. Most brands end up hybrid: sensors where a retailer co-invests, and AI shelf monitoring everywhere else.

How accurate is the recognition, and what happens when it is wrong?

Accuracy is high enough for commercial use on mainstream packaging, but no vision model is perfect — reflective film, damaged packs, and new artwork all create edge cases. The design answer matters more than a benchmark number. The rep should be able to correct a count, and that correction should trigger an automated re-check rather than being silently overwritten. A platform with no correction path tends to see adoption collapse the first time it miscounts a SKU the store manager is watching.

Does AI shelf monitoring replace field reps?

It replaces the part of the visit reps dislike and do badly under time pressure: manual counting and report writing. The judgement work stays human, because negotiating a display, persuading a store owner to free up a facing, and deciding which gap is worth escalating are not counting problems. The practical effect is that the same rep covers the same outlets with better information and more time in front of the person who controls the shelf — which is why availability improves fastest where the technology is treated as a co-pilot rather than a replacement.

Availability is won or lost in the thirty seconds a rep spends in front of a shelf, and most brands have never held a reliable record of what happens in those thirty seconds. AI shelf monitoring changes that by making the shelf readable, correctable, and actionable inside the visit rather than after it. If out-of-stocks are quietly taxing your trade spend, the cheapest experiment is a single region run on shelf AI before you commit across markets. Talk to the eBest route-to-market team about a pilot in your toughest territory.