Introduction

The story you've heard about AI in retail and FMCG is mostly correct. Adoption is up sharply, vendors are everywhere, and the leaders in the space are pulling ahead of the rest. The story you haven't heard as often is that for most enterprises, the AI spending hasn't shown up in revenue.

The gap is real and measurable. Industry research published in 2024 and 2025 puts it cleanly: while adoption has nearly doubled in two years, only about one in twenty companies is currently generating substantial bottom-line value from AI. The technology is being deployed widely. The revenue impact is concentrated narrowly.

This article is for the CIOs, CDOs, and commercial leaders trying to figure out where AI actually moves revenue in retail and FMCG and why so many investments don't. We'll look at the gap between AI adoption and AI revenue impact, the three decision categories where AI consistently lifts revenue, and one engagement that shows what a revenue-tied AI deployment looks like in practice. The framing throughout is operational rather than aspirational.

The Promise vs The Reality: AI in Retail and FMCG Today

AI adoption in retail and FMCG has moved from experiment to enterprise priority faster than any prior wave of enterprise technology. A 2024 McKinsey survey found that 71% of CPG leaders had adopted AI in at least one business function, up from 42% the year before. Industry analysis suggests that 90% of retail executives are now experimenting with generative AI, and roughly two-thirds have piloted gen AI in at least one workflow.

The market sizing reinforces this. Industry analysis of digital and AI transformation in CPG estimates the technology could create up to $660 billion in annual value across the industry by 2030. For a single $10 billion food and beverage company, the modeled upside ranges from $810 million to $1.6 billion in addressable value, most of it concentrated in customer and channel management.

Then there's the reality.

A 2025 global study of 1,250 companies found that only 5% are generating enough bottom-line value from AI to drive shareholder returns. A separate 2024 study found that 74% of companies struggle to scale AI beyond pilots. Industry research from KPMG showed that nearly half of CPG companies are still struggling to integrate their data effectively, the prerequisite for any AI investment to produce useful output.

The picture that emerges is consistent. AI adoption is broad, market potential is large, and yet for most companies, the technology hasn't translated into measurable revenue impact yet. Among the small share that is generating revenue from AI, the same research shows they're seeing roughly twice the revenue growth and 40% more cost savings than the laggards.

That gap between most companies and the few that capture real value is what this post is about.

Where AI Actually Moves Revenue: A Taxonomy

The companies generating real revenue impact from AI share a specific approach. They apply AI to a smaller set of decisions, with a much sharper definition of what those decisions are.

Across retail and FMCG, three decision categories consistently produce measurable revenue impact when AI is applied well.

1. Pricing Decisions

Pricing is the most direct revenue lever in retail and FMCG, and the one where AI most cleanly maps to a measurable outcome. A pricing model evaluates elasticity, competitive position, and historical response. It recommends a price point. The recommendation either gets implemented or it doesn't. Revenue moves up or down based on the recommendation.

The directness of this loop matters. Pricing decisions get made, prices get set, and the market response is observable within weeks. Compared to most enterprise AI use cases, the signal-to-noise ratio is high. Industry research on revenue growth management indicates that AI-driven pricing optimization is one of the most profitable applications of AI in CPG, with documented margin and revenue gains.

2. Trade Marketing and Promotional Spend

Most FMCG companies spend between 15% and 25% of revenue on trade marketing and promotions. Most of them can't accurately attribute which portion of that spend produced incremental revenue and which produced none.

AI applied to trade promotion optimization addresses this directly. Machine learning models that combine sales data, promotional history, competitor activity, and shopper behavior can identify which promotions worked, which didn't, and where the next marketing dollar should go. The decision being changed is specific, which promotion to run, when, at what depth and the outcome is measurable in lift.

3. Distribution and Inventory Decisions

The third category is operational. Demand forecasting, inventory rebalancing, and distribution optimization don't generate revenue directly, but they reduce the cost side of every revenue dollar and prevent stockouts that suppress revenue. A 2024 benchmark study found that AI applied well to forecasting can reduce error rates by up to 65% and improve supply chain efficiency by 20%.

Example: Unilever's use of weather-pattern AI for ice cream demand forecasting reportedly drove a 30% sales increase in key markets, simply by ensuring product availability matched temperature-driven demand more accurately.

In all three categories, the pattern is consistent. AI is tied to a specific decision. The decision has an owner. The outcome is observable on a short cycle. Revenue moves.

Why Most AI Investments Don't Hit Revenue

If AI lifts revenue when applied to specific decisions, the natural question is why so many enterprise AI investments don't produce revenue impact. The pattern across recent industry research, and what we see across engagements, points at a single root cause.

Most AI gets deployed horizontally rather than vertically. A horizontal deployment looks like this. An enterprise stands up a data lake, integrates a generative AI platform, gives employees access to a copilot interface, and waits for value to emerge. The technology works. Productivity goes up at the margins. Some workflows get faster. But no specific revenue-bearing decision changes, because the AI isn't connected to a decision at all, it's connected to a search box.

A vertical deployment looks different. AI is tied to one decision, one decision-owner, one outcome metric. The model output flows into the workflow where the decision actually gets made. When the recommendation is acted on, the result gets measured. When it's ignored, the system learns why.

Recent analysis of where AI value concentrates makes this explicit: roughly 70% of AI's measurable value sits in core business functions where decisions are made — pricing, marketing, supply chain, customer engagement — not in the horizontal productivity tools that absorb most of the spend.

The companies generating revenue from AI have figured out that the architecture of the deployment matters more than the sophistication of the model.

A Look at One Pattern in Practice

Below is a walkthrough of one Finzarc engagement that illustrates a vertical AI deployment in FMCG.

The Context

A global FMCG company operating across multiple consumer categories: personal care, food, household faced exactly the gap described above. The company had data. It had analysts. It had a BI platform. What it didn't have was a clear path from analytics output to the pricing, marketing, and distribution decisions its brand teams made every week.

The specific business challenge was the annual operational planning cycle. Each year, CXOs needed to build the AOP for the coming financial year, making revenue, margin, distribution, and spending commitments across the flagship brand portfolio. The challenge had four observable parts:

• Scattered data. No single source of truth. Sales data, marketing spend, distribution data, and consumer panel data lived in different systems.
• Limited visibility. Brand teams couldn't drill down by geography, pack size, and timeline in a single view.
• Impact blind spots. Attributing the actual revenue impact of brand strategies was largely guesswork.
• Inefficient planning. The AOP was effectively a manual exercise driven by historical performance, with limited ability to test scenarios.

The combined effect was a decision system in which the most consequential commercial decisions of the year: pricing changes, marketing allocation, distribution shifts were being made on stale, fragmented data, with little ability to predict outcomes.

The Build

The work focused on building what we called a Brand Optimizer, a vertical AI deployment tied directly to the pricing, marketing, and distribution decisions that determined the year's commercial outcomes.

The system had four capabilities, all designed to live inside the workflow brand teams were already running.

• Brand Overview consolidated brand health, event mapping, and strategic recommendations across the portfolio into a single view. Instead of pulling data from five systems before a planning meeting, the meeting opened with the view already up.
• Brand Analytics enabled drill-down by geography, pack range, and timeline. Brand teams could move from a portfolio-level question to a SKU-level diagnosis without leaving the interface.
• Scenario Simulator forecasted revenue outcomes with high accuracy across tested scenarios. Before committing AOP budget, CXOs could test pricing, distribution, and marketing scenarios against the model and see projected outcomes.
• GenAI Chatbot replaced hours of manual pivot tables and chart-building with conversational queries. A brand manager asking "which packs are losing share in the western region?" got an answer in seconds rather than a slide deck two days later.

The critical architectural decision was that the system wasn't a dashboard. It was tied directly into the AOP planning workflow. Recommendations from the model flowed into the actual decisions being made, with named owners on each decision and measurable outcomes attached.

The Outcomes

Three flagship brands were taken through the full cycle. Each one produced a clear, measurable result tied to a specific decision change.

• A personal-care brand. The Scenario Simulator identified a pricing inflection point the team had missed. A targeted price adjustment moved gross margin from 30% to 38% and produced a +27% revenue lift across the brand.
• A food category brand. Better attribution from the analytics layer surfaced which marketing channels actually produced incremental sales. A 10% increase in spend, redirected to the channels the model flagged, translated to +12% revenue growth.
• A household-care brand. The model flagged a distribution and media mix that was no longer paying back. A redirected strategy cut operations costs by -26% without revenue loss.

In each case, the AI didn't replace the brand team's judgment. It changed the inputs the team's judgment ran on, and tied the team's decision to a measurable outcome with a documented owner.

What This Case Reveals

The Brand Optimizer engagement isn't unique. The pattern represents vertical AI tied to specific revenue decisions is what consistently shows up in retail and FMCG deployments that produce measurable revenue impact.

Three things separated this deployment from the broader pattern of AI investments that don't move revenue.

• The decisions were named in advance. Before any model was built, the specific decisions the system would inform pricing, marketing allocation, distribution were identified, with owners attached.
• The model output flowed into the workflow. The system was integrated with the AOP planning process, not standing apart from it. Brand teams used the model output inside their existing decision cycle.
• Outcomes were measured against the decisions. Every recommendation that was acted on produced a measurable revenue or cost outcome that could be traced back to the model. The system had a feedback loop, which meant it could learn.

These are architectural properties, not algorithmic ones. The model behind the Brand Optimizer wasn't dramatically more sophisticated than what a horizontal AI platform would offer. What was different was the architecture around it.

The Finzarc View

The gap between AI adoption and AI revenue impact in retail and FMCG isn't going to close because the models get better. The models are already good enough. The gap closes when the architecture changes, when AI gets deployed against specific revenue-bearing decisions, with named owners, integrated workflows, and feedback loops on outcomes.

This is the work we focus on at Finzarc. We help retail and FMCG enterprises move from horizontal AI deployments that don't show up in revenue to vertical ones that do. Pricing decisions tied directly to model recommendations. Trade promotion optimization that closes the loop on attribution. Distribution and supply decisions made on forecasts that are accurate enough to act on.

The economics work when the architecture works. Better models won't fix it. Better systems will.