Next Best Action Agent — Class 5 of the Asseco Platform AI Agents Framework. AI-driven recommendations for FMCG field sales. Three prioritized actions per store visit, tailored to each location. Used at scale across 55,000+ field users in 62 countries.
Next Best Action Agent by Asseco Platform: AI recommends what to do right now, in this store. Three sub-agents in Class 5: In-store Recommendation Agent (Retail Image Recognition), Guided Selling Agent (Sales and Retail Execution, RAO feature, POI Best-in-Class 2025), Order Recommendation Agent (Third Party Field Execution).
AI Agents Framework → Class 5
Next Best Action Agent: AI recommends what to do right now, in this store.
Every store is different. Every visit has a different context. Next Best Action Agent analyzes the data from each specific store — purchase history, shelf state, contract compliance, store profile — and ranks the three most impactful actions for the rep to take, right here, right now.
Every store is different. AI knows what to do in this specific one.
A field sales rep walks into a store with dozens of possible actions: rotate products, propose additional listings, adjust the display, verify contract compliance. The problem: every store is different. Different purchase history, different shelf, different owner priorities. Universal guidelines don’t work.
Next Best Action Agent is an AI-driven engine that analyzes the data from each specific store — purchase history, shelf state, contract terms, store profile — and delivers three concrete, prioritized actions for the rep during the visit. The three most impactful things to do, now, in this particular store.
AI correlates data that human perception cannot hold at once.
The rep opens the app before entering the store. The agent has already collected and analyzed the data: recent orders and purchase trends from Trade Data Hub, current shelf state from Shelf Recognition Agent, active contract terms with the outlet or chain, demographic profile of the neighborhood, store type (neighborhood mini-market, convenience, HoReCa outlet), and outcomes from similar stores in the region. On this basis, the agent generates a ranked list of actions — from most to least impactful for this specific store.
Collect
The agent pulls data from every source that matters for this store: purchase history from Trade Data Hub, shelf state from Shelf Recognition Agent, contract terms, store profile, patterns from similar outlets. Before the rep arrives at the door.
Invisible backend — runs continuously
Rank
Business rules and ML models correlate the data and rank possible actions by expected impact — from most to least valuable for this specific store at this specific moment. Each recommendation carries its own justification: why this, why now.
Data-driven prioritization, not intuition
Deliver
The rep opens Mobile Touch and sees the three most important actions for this store, each with a concrete justification. No checklist to tick off. A ranked plan to execute. The rep doesn’t wonder what to do — walks in with a plan and starts the conversation with specifics.
Top 3 actions, delivered in the app the rep already uses
What changes when AI pre-ranks the work.
Universal guidelines meet the reality of one store. Next Best Action Agent replaces the generic checklist with a priority plan — tailored to this outlet, this moment, this data.
What reps work with today
- Plan the visit from generic guidelines — “check the shelf, propose new products” — no reference to the specifics of this outlet
- Prioritize by intuition — what feels important, not what will actually have the biggest effect in this store
- No context from previous visits — every visit starts from zero, no knowledge of what worked in similar places
- Visit time absorbed by on-the-spot analysis instead of conversation with the owner and executing tasks
- Key actions missed — reps don’t see the opportunity because they don’t have the full picture of the data
- Head office gets reports of what was done, but without context — were those the most important actions for that store?
What reps get with AI
- AI analyzes data before the visit — orders, shelf, contract, store profile — and surfaces three concrete actions with the highest potential
- Every recommendation carries its own justification — why this action, why now, what the data says
- Context from similar outlets — AI knows what worked in stores with the same profile and suggests proven approaches
- The rep walks in with a ready plan — the conversation with the owner starts from specifics, not from diagnosis
- Head office sees not only what was done, but whether those were the most important actions — and monitors AI recommendation effectiveness over time
- Every skipped recommendation has a reason code — managers learn why reps deviated, and the agent improves
Five data sources. One ranked decision.
Each source answers a different question about this specific store. The agent correlates all five and ranks the actions accordingly.
Purchase history
Transaction data: order frequency, volume, rotation, seasonality. The agent detects shifts in patterns — a drop in a category where you were leading, growth in competitors’ share, a new product line appearing in the basket.
Example recommendation: “Beer category purchases dropped 40% over the last two months. Propose a return to the previous level with sales arguments based on data from similar outlets.”
Shelf state
Data from Shelf Recognition Agent — product presence, number of facings, planogram compliance, position relative to competitors, shelf visibility. The agent compares the current state against contract terms and optimal exposure for this store type.
Example recommendation: “Your premium product has one facing on the bottom shelf; competition has three at eye level. Propose an exposure correction aligned with the contract — or renegotiate terms if the current agreement doesn’t give you leverage.”
Contract terms
The agent verifies the execution of contracts signed with chains and outlets. Checks whether agreed actions have been implemented: product listing, shelf position, POS materials, price levels. If the contract is not being executed, the agent flags this as a visit priority.
Example recommendation: “Contract specifies seasonal product exposure through end of month. Product is not on the shelf. Priority: conversation with the owner and display setup.”
Store profile & demographics
The agent considers outlet type (neighborhood mini-market, station convenience, large convenience), sales floor area, location, and neighborhood demographics (age, income, population density). Recommendations match the potential of the place.
Example recommendation: In a neighborhood store with family clientele — emphasis on family-size packaging. In an outlet near a train station — snacks and small formats.
Patterns from similar outlets
The agent aggregates data from outlets with similar profiles — type, size, location, demographics — and identifies proven actions that delivered results in those places. If introducing a given product in stores similar to the one you’re visiting increased order volume by 30%, the agent ranks that recommendation higher. It doesn’t guess. It knows what works.
Example recommendation: “In 15 outlets with this profile in the region, introducing product X increased orders 3x over the baseline period. Proposed: introduce X in this store, backed by regional benchmark data.”
The rep doesn’t walk in with a checklist to tick off. The rep walks in with a plan.
A ranked set of the three most important actions for this specific store — generated by an agent that processes data a human cannot hold at once. The visit stops being a routine inspection. It becomes a strategic intervention built on facts: what to change, why it pays off, what the data says about the potential.
That’s the difference between a visit run off a checklist and a conversation built on data.
Next Best Action Agent fits your stack — whatever shape it’s in.
Next Best Action Agent is the delivery layer of the framework — the moment the AI recommendation meets the rep. What feeds the agent is flexible. You can take the full Asseco Platform framework (Classes 1–4 deliver the data, scoring, and context). You can bring your own data pipelines and scoring logic, and let us deliver only the rep-facing layer. Or you can mix.
Full framework
Asseco Platform runs Classes 1–4 under the hood. Trade Data Hub unifies your data, Recognition Agents structure observation, Scoring & Prioritization ranks outlets and contracts, Context Agents assemble the visit briefing. Next Best Action Agent delivers the top 3 actions to Mobile Touch.
Simplest setup. Single vendor. Lowest integration overhead.
Bring your own
You already have your data lake, customer scoring models, and CRM-based visit context? Next Best Action Agent plugs in via API. Your inputs, our rep-facing delivery layer. You keep investment in your existing stack, add the recommendation layer on top.
For teams with mature internal analytics. Faster to deploy than replacing the stack.
Hybrid
Mix and match. Our Shelf Recognition Agent for shelf state data — but your internal scoring logic. Or our Trade Data Hub for distributor data — but your CRM-based visit context. Whatever split fits your current stack.
For teams mid-transition. Keep what works, adopt what’s missing.
Whatever feeds the agent, the output is the same: the top 3 actions in Mobile Touch, with justification, at the moment the rep needs them.
Two scenarios where Next Best Action Agent drives the outcome.
Every recommendation the agent delivers to the rep is the last step of a chain. Below we show the full chain: which classes contribute the data, and where Next Best Action Agent formulates the action the rep actually sees. That’s how a recommendation earns its place in the rep’s app — backed by the work of the framework, not by intuition.
Two more scenarios — visit prioritization and distribution expansion — are primarily driven by Class 3 (Scoring & Prioritization). We keep them briefly below as related scenarios, and treat them in full depth on the Class 3 spoke (coming soon).
Use Case 1 · Product mix optimization
Every centimeter of shelf space is currency — spend it on products that rotate.
Challenge: FMCG producers in traditional trade serve thousands of small and mid-sized outlets where sales floor space is limited — every shelf meter is a compromise. The outlet can’t carry the full portfolio. It picks products it believes will rotate, and skips the rest. The problem: those choices are often suboptimal. The dealer keeps products with declining rotation and doesn’t introduce new items that are already working in similar stores. The rep sees it during the visit but has no objective data to convince the owner. The dealer replies “That won’t sell here” — and there’s no counter-argument.
Why traditional methods fail: The rep knows what the producer wants on the shelf. The dealer knows what they believe rotates in their store. Neither has an objective view of what actually works in outlets with a similar profile — size, location, demographics. The result: product mixes are suboptimal, the dealer carries low-rotation items, the producer loses potential, and the conversation about assortment changes is based on beliefs, not on data.
How Asseco Platform solves it:
- The agent analyzes transaction data from this outlet: which products rotate fast, which sit on the shelf, how the purchase structure shifts over time
- It compares that data with results from outlets with a similar profile (surface area, type, location, customer profile) and identifies products that work there but haven’t been tested here
- Next Best Action generates a pre-visit recommendation: “Product X shows low rotation — 2 sales in the last 6 weeks. In similar stores, product Y rotates 3x faster. Propose a swap — data from 15 outlets in the region confirms the potential.”
- The rep enters the conversation with specifics: which products to change, why, what the data says about places like this store
- The outlet gets not the rep’s opinion but a benchmark grounded in facts — data is harder to dismiss than a suggestion
Effect: Product mixes in small outlets stop being accidental. The producer knows which products optimize rotation in each store type. The owner receives a concrete change proposal based on data from similar places — and doesn’t have to guess what will work. The rep leads the conversation as an advisor, not a portfolio salesperson.
“I no longer ask if the store wants a new product. I show them the data: what rotates in outlets like theirs, what they earn per shelf meter. The conversation starts with facts, not with persuasion.”
— Representative feedback pattern from Next Best Action deploymentsChain in this scenario: Class 1 (Trade Data Hub unifies transaction data from this outlet and similar stores) → Class 3 (Store Potential Scoring benchmarks rotation across similar-profile outlets) → Class 5 (Next Best Action Agent formulates the specific recommendation the rep sees in Mobile Touch).
Use Case 3 · Proactive exposure correction
Contract says one thing, the shelf shows another — AI catches the gap before it escalates.
Challenge: FMCG producers negotiate contracts with chains that specify exposure terms: number of facings, shelf position, POS material visibility, planogram compliance. The problem: execution of those terms is variable. The chain signs the contract, but for many reasons (product rotation, competitor pressure, merchandising errors) the actual in-store exposure deviates from the agreement. The rep notices during a visit — but if the store is visited once a month, the gap can persist for weeks before correction. Result: the producer pays for terms that aren’t delivered, and the chain has no signals that would help return to compliance.
Why traditional methods fail: Shelf Recognition data shows shelf state, CRM data shows what’s in the contract — but these two sources often don’t talk to each other in real time. The rep can spot discrepancies if they happen to remember the contract terms for that outlet and have time to inspect the shelf carefully. Result: exposure problems are caught with a lag, and interventions are not prioritized by where the deviation is largest.
How Asseco Platform solves it:
- The agent automatically compares data from Shelf Recognition Agent (facing count, shelf position, POS material presence) with contract terms stored in the system
- If exposure deviates from the agreement, Next Best Action Agent generates an alert and places the correction as a visit priority at that outlet
- During the visit, the rep sees the recommendation: “Contract specifies 4 facings of product X on the middle shelf. Shelf Recognition shows 1 facing on the bottom shelf. Priority: exposure correction aligned with the contract.”
- The system delivers concrete data for the conversation with the manager: what’s in the contract, what’s on the shelf, what the state looked like in previous visits
- Head office monitors compliance and sees which outlets need intervention before the deviation becomes a relationship issue
Effect: Deviations from contracts are caught automatically and corrected proactively — not after weeks, but during the visit. The owner receives objective data showing the exposure doesn’t match the agreement, and the rep has arguments grounded in facts rather than impressions. The producer pays for terms that are actually delivered, not merely declared. Compliance stops being a matter of luck and memory — it becomes a monitorable process.
“The system tells me directly: here the contract isn’t being executed, here the shelf looks different than it should. I don’t have to remember the terms of every agreement. AI remembers them and signals where the problem is.”
— Representative feedback pattern from Next Best Action deploymentsChain in this scenario: Class 2 (Shelf Recognition Agent reads facings, position, POS materials from the photo) → Class 3 (Contract Compliance Scoring Agent compares observed state against contract terms) → Class 5 (Next Best Action Agent surfaces the correction recommendation to the rep during the visit).
Related scenarios — primarily other framework classes
Two scenarios where NBA surfaces the output — but the engine lives in Class 3.
Next Best Action Agent delivers these to the rep’s app, but the ranking logic — which outlets, which customers, why them — is driven by Scoring & Prioritization Agent (Class 3). Full treatment coming in the Class 3 spoke.
Related scenario 1
Visit prioritization
A rep covers hundreds of outlets. AI ranks which stores need attention first — based on growth potential and risk of losing position. The rep’s route stops being geographic rotation and becomes intervention where it matters most.
Primary driver: Class 3 (Store Potential Scoring Agent). NBA’s role: deliver the ranked list to the rep in Mobile Touch. Full treatment on the Class 3 spoke (coming soon).
Related scenario 2
Numeric distribution expansion
A producer wants to expand distribution of new products. AI targets the customer population most likely to adopt them, based on historical purchase patterns — and tracks both introduction and on-shelf retention, with transparent bonus settlement for distributor reps.
Primary driver: Class 3 (customer population scoring + historical success pattern matching). NBA’s role: surface the targeted customer list per rep, with progress and bonus tracking. Full treatment on the Class 3 spoke (coming soon).
Next Best Action is one class. The framework has five.
Asseco Platform AI Agents Framework groups thirteen specialized AI agents into five classes — each one answering a different question FMCG teams face every day. Next Best Action is Class 5 — the recommendation layer. See how the full chain works, from raw data to concrete action.
Class 1
Data Unification Agent
One clean dataset from scattered sources.
Class 2
Recognition Agent
Every photo becomes structured data.
Class 3
Scoring & Prioritization Agent
Every store ranked, every moment.
Class 4
Context Agent
Walk in already informed.
Class 5 — You are here
Next Best Action Agent
What to do right now, in this store.
Next Best Action Agent draws on outputs from the other four classes — unified data (Class 1), recognized shelf and menu states (Class 2), store and outlet rankings (Class 3), and assembled visit context (Class 4). That’s how one recommendation earns its ranking: it’s backed by the full chain.
Benchmarked by analysts, validated by the industry.
Next Best Action Agent builds on two POI Best-in-Class distinctions (Guided Selling and AI/Machine Learning), runs under ISO/IEC 27001:2022 certified infrastructure with AI-based solutions explicitly in scope, and is part of a platform named Representative Vendor in the Gartner Market Guide for Retail Execution Management.
POI Best-in-Class — Guided Selling 2025
Next Best Action Agent’s Guided Selling instance (deployed in Sales & Retail Execution, RAO feature) received the POI Best-in-Class distinction for Guided Selling in the 2025 Consumer Goods Enterprise Planning & Retail Execution Vendor Panorama Report. Independent benchmark: this class of agent is among the strongest on the market.
POI Best-in-Class — AI/Machine Learning
POI also recognized Asseco Platform with a Best-in-Class distinction for AI/Machine Learning — the technology foundation that Next Best Action Agent (and every other AI agent in the platform) relies on. Two POI distinctions directly validate this class.
ISO/IEC 27001:2022 — AI-based in scope
Asseco Business Solutions is certified to ISO/IEC 27001:2022 — the international standard for information security management. AI-based solutions are explicitly included in the certification scope. Valid through February 2029.
Common questions about Next Best Action Agent
Next Best Action Agent is Class 5 in the Asseco Platform AI Agents Framework — an AI-driven engine that analyzes data from each specific store and ranks the three most impactful actions for the field rep during the visit. In Asseco Platform it appears as three specific agents: In-store Recommendation Agent (Retail Image Recognition), Guided Selling Agent (Sales & Retail Execution, RAO feature), and Order Recommendation Agent (Third Party Field Execution).
No. Next Best Action Agent ranks and recommends — the rep decides. Each recommendation carries a justification (why this action, why now, what the data says). Reps can accept, modify, or skip any recommendation — and every skipped recommendation can carry a reason code, so managers see why the agent’s suggestion was overridden and the system can learn.
Five sources feed the recommendation: transaction history (from Trade Data Hub), shelf state (from Shelf Recognition Agent in Retail Image Recognition), contract terms (from the CRM/CMS), store profile and demographic data, and patterns from similar outlets. The richer the data, the better the ranking — but even a subset delivers value from day one.
A generic assistant answers whatever you ask. Next Best Action Agent is narrowly specialized: it ranks field sales actions based on operational data. It delivers the recommendation in the app the rep already uses (Mobile Touch), with a specific justification. This design trades generality for reliability — the rep gets an answer to one specific question, always grounded in the same data sources.
Both. The use cases above describe retail (traditional trade, convenience, chains). For HoReCa, Next Best Action Agent combines with Menu Recognition and Pre-Negotiation Briefing Agent — the same logic applies: rank actions per outlet based on data. In HoReCa, that data includes menu presence, listings, and competitor exposure on menus.
Deployment time depends on which sub-agent you start with. Guided Selling Agent (RAO feature in SRE) can go live in weeks once Sales & Retail Execution is in place. In-store Recommendation Agent requires Retail Image Recognition deployed. Order Recommendation Agent runs in Third Party Field Execution. Full Class 5 coverage scales with the other four classes in the framework.
Yes. Every recommendation is logged with its outcome: accepted, modified, skipped (with reason code). Head office sees acceptance rates per store, per rep, per product category — and tracks whether accepted recommendations correlate with business outcomes (order volume, shelf share, contract compliance). The agent becomes measurable as a system, not a black box.
Asseco Business Solutions is certified to ISO/IEC 27001:2022, with AI-based solutions explicitly in the certification scope. Data handling, retention, and access control follow the same policies that cover the rest of the platform.
Ready to see Next Best Action Agent running on your data?
Book a 30-minute demo with our field execution specialists. We’ll configure a live recommendation flow based on your industry (beverages, food & grocery, personal care, HoReCa) and show you how the top 3 actions get ranked for a real store profile.
Part of the AI Agents Framework · 9× POI Best-in-Class 2025 · ISO/IEC 27001:2022 with AI in scope
