AI in FMCG Commerce: Turning Board Pressure Into Business Outcomes

Fig. AI use cases in FMCG mapped to business outcomes.

The last row is the least written about and, for manufacturers and distributors, the closest to revenue. Ordering suggestions, personalized pricing shown at the point of purchase, and search that understands trade buyers all depend on one thing: connected commerce data. That dependency will recur throughout this article.

PepsiCo illustrates how quickly the leaders are moving along the whole table at once. In 2025 it became the first major food and beverage company to deploy Salesforce's Agentforce AI agents at enterprise scale, spanning customer service, field sales, marketing, and supply chain across a business serving some five million retail locations, while expanding its AWS partnership to power PepGenX, its internal generative AI platform. Notice what the deployment is designed to do: remove friction from ordering, improve on-shelf availability, and sharpen promotions—commerce outcomes, fed by connected commerce data.

Two intros deserve a section of their own—both from beverages, both operating at a scale the rest of the industry studies. Before them, though, it's worth seeing how AI in consumer goods plays out beyond the order book, because the generative side of the story carries both the best productivity numbers and the clearest warning.

Unilever has rebuilt product content production around AI digital twins—pixel-perfect 3D versions of its products created with NVIDIA Omniverse. The company reports up to 55% cost savings and 65% faster content turnaround, with double the click-through rate and three times longer attention. One TRESemmé campaign in Thailand cut content creation costs by 87% while lifting purchase intent 5%.

Nestlé applies the same digital-twin idea to packaging visuals in marketing, and has gone further upstream: with IBM Research it built a generative AI tool that proposes novel high-barrier packaging materials using a chemical language model. Coca-Cola, meanwhile, launched Project Fizzion with Adobe in May 2025—a design intelligence system that encodes brand rules into machine-readable "StyleIDs" and claims content production up to ten times faster.

The caution comes from the same company. Coca-Cola's AI-generated 2024 Christmas ad drew widespread criticism as "soulless" and "devoid of any actual creativity", and the 2025 follow-up attracted backlash again despite technical improvements.

Generative output without judgment—about brand, audience, and context—cuts both ways. Fizzion and the Christmas ads came from the same organization, which rules out any simple verdict on the technology.

The real lesson: outcomes depend on the guardrails and data wrapped around a model, rather than the model itself.

If you want evidence that AI in food and beverage industry operations can move revenue rather than demo well, the beverage giants supply it. AB InBev's BEES platform generated $52.5 billion in gross merchandise value in 2025 across 29 markets, with 72% of company revenue captured through digital B2B channels, per its full-year results.

BEES is a B2B super-app through which bars, shops, and restaurants order from AB InBev and its distributors—and its most instructive statistic concerns who places the orders. CEO Michel Doukeris has said that 75% of orders are generated by machine learning: "neither us nor the customer is creating the order; the customer is just accepting the sales proposal that he receives." That is AI applied directly to the ordering problem—next-order prediction, assortment suggestions, replenishment—at the scale of a global route to market.

HEINEKEN has built the equivalent for its sales force. AIDDA, its AI Data-Driven Advisor, recommends next best actions to representatives, turning order-takers into advisers. By 2024 the company had scaled it to eight markets, where it's used by more than 3,700 sales reps and telesales agents supporting 490,000 customers. In Mexico, HEINEKEN reports AIDDA lifted gross profit by 3–4%. Its Promo Advisor tackles the FMCG-specific problem of simulating and planning promotions, and in late 2025 the company added CoBrain, a generative AI layer within its Connected Worker program, now deployed across more than 150 breweries. CoBrain is also a measure of how far AI in food industry manufacturing now reaches beyond the order book, into production itself. All of it is HEINEKEN's own development. And none of it would run without the years HEINEKEN spent beforehand digitizing and connecting the commerce and route-to-market systems the AI reads from and writes to.

That sequencing shows up in our own client work too, in the same industry. Virto Commerce is HEINEKEN's commerce partner for its mobile-first B2B route-to-market platform, launched as an MVP in Singapore and extended across more than 25 countries—the connected ordering layer that this kind of sales AI presupposes, and a separate effort from HEINEKEN's in-house AI products.

The pattern repeats at different scales. When Dutch beverage wholesaler De Klok Dranken replaced its Magento-based storefront with a composable platform integrated into its retained SAP ERP, roughly 80% of its 4,000+ corporate customers moved to digital ordering—the adoption base any future ordering intelligence would learn from.

And Lavazza's Benelux dealer Bluespresso consolidated thousands of customer-specific price lists into one unified B2B and B2C store, turning personalized pricing from a spreadsheet exercise into structured data.

None of these is an AI case study. Each is the precondition for one.

The prize for getting past them is real: Bain estimates that scaling AI could improve consumer products operating margins by 3–5 percentage points. Yet most FMCG and CPG organizations hit the same three walls—organizational complexity, fragmented data, and technology estates that resist connection.

Consider what a typical FMCG manufacturer looks like from the inside:

• multiple operating companies,
• regional divisions with their own P&Ls,
• dozens of brands accumulated partly through acquisition,
• a technology stack assembled the same way, and
• an independent distributor ecosystem beyond which the company often has no visibility at all.

Every one of those seams fragments the data an AI initiative needs.

• Product information lives in regional spreadsheets and disconnected PIM systems;
• orders flow through distributor systems the manufacturer can't see;
• promotions are planned in one tool and evaluated, months later, in another.
• A forecasting model fed from that estate produces forecasts as fragmented as its inputs.

Which is why so many AI programs begin somewhere unexpected: with commerce modernization.

The common triggers are familiar—

• a legacy platform constraining growth,
• an AI mandate from the board,
• entry into new markets or channels,
• commercial operations still run manually.

Each is really the same trigger: the existing systems can't supply what comes next.

A composable commerce platform sits on top of the ERP systems FMCG companies already run—SAP, Microsoft Dynamics, Oracle—rather than replacing them.

It unifies the three data domains AI in FMCG depends on:

• product (clean, complete, consistently structured catalog data),
• price (contract and customer-specific pricing as structured rules rather than spreadsheets), and
• orders (every channel and market flowing through one connected layer).

Because it's API-first, whatever consumes that data next—a forecasting model, an ordering assistant, an autonomous purchasing agent—can connect without another integration project. And because it's modular, modernization proceeds without a big-bang replacement: one market, one storefront, one process at a time.

To be clear about what this is and isn't: a platform like Virto Commerce is the foundation that makes AI usable in FMCG commerce, rather than the AI itself. FMCG companies that treat that foundation as the first AI investment tend to be the ones whose later AI investments produce numbers a board will accept.

Three developments deserve a place on any FMCG planning horizon, and each rewards early data groundwork.

1. Autonomous ordering and machine customers. BEES's 75% machine-generated orders are a preview: the buyer accepts a proposal rather than composing an order. The next step—AI agents that place orders within rules a customer sets—is the direction BCG describes for agentic AI in consumer goods. Agents can only buy from systems that expose clean product, price, and availability data programmatically.
2. Adaptive trend and demand intelligence. Models that read sell-out, social, and market signals continuously will compress the distance between a demand signal and a commercial response—provided the channel data exists to read.
3. Findability in AI search. As trade buyers ask AI assistants what to stock and where to buy it, brands and distributors face a new visibility question: does your product data surface in generated answers? Winning there starts with structured, complete, machine-readable catalogs.

Each of these is a topic in its own right, and we'll treat them separately in forthcoming articles.

Not with a model. The organizations inside Bain's 6% share a sequence, and it's short enough to keep on one page:

1. Pick a use case tied to a revenue or cost KPI—incremental sales from order suggestions, margin from promotion optimization, forecast-driven waste reduction—and name the number it must move.
2. Audit product-data quality first. If the catalog is inconsistent across markets, fix that before training anything on it.
3. Connect commerce to the ERP. Orders, pricing, and availability need to flow through one layer AI can read.
4. Buy proven capability rather than building it. FMCG leaders increasingly adopt established foundations instead of maintaining fragile in-house builds—the pattern across every example in this article.
5. Measure the business outcome, never the pilot. A pilot that "worked" but moved no KPI is a cost.

The task facing FMCG executives is less about adding AI to the company than about building a commerce foundation that lets the business keep adopting it—as models improve, as agents start placing orders, as discovery moves into AI assistants, and as products, markets, and channels change around all of it. Boards will keep asking. The teams with good answers will be the ones who invested first in the data and connectivity everything else runs on.

If that's the conversation happening in your business, start with the FMCG commerce foundation and let the AI roadmap build on it.