The Machine Will Ask Why: FMCG Brands, the Agentic Economy and the 2026 Holiday Test

FMCG has lost its growth algorithm. The agentic economy is about to test the one asset that was supposed to protect it. Brands that enter the 2026 holiday season without an AI-layer baseline will have no comparable reference point for reading what the season did to their position.

For two decades, FMCG boards could rely on a familiar formula. A little volume, a little price, some mix, and margin expansion on top. That formula has broken, and the industry is now openly debating whether it will return.

This article argues that a second force will compound the damage. AI assistants and shopping agents are becoming the first evaluator of the brand. They do not carry the emotional memory that brand investment was built to create. They reason from attributes, price and evidence, and they explain their choices. The premium a brand commands must now survive a machine asking why it deserves to be paid.

The algorithm that stopped working

FFA's review of first-half 2026 results across the 55 largest FMCG companies sets out the problem plainly. Aggregate volume has turned positive again at around 1.6%, and pricing has settled near 3%. Yet roughly six in ten of those companies are still growing slower than their categories.

The longer view is starker. FFA's analysis of Euromonitor data puts annual FMCG volume growth over 2020 to 2024 about 110 basis points below its 2014 to 2020 rate. Aggregate recovery and company performance are diverging: the market is growing again while most large players lose share within it.

The forward view is harder still. Consumers have learned to trade down and wait for promotion. Retailers are consolidating, and private label is spreading fastest in the slowest categories. Easing input costs remove the justification for price increases rather than restoring pricing power. In several categories, a flat or negative price outlook for 2027 is the realistic case.

We agree with that diagnosis. Our concern is with what it leaves out. Every one of those pressures operates at a point of choice, and the point of choice is moving.

The agentic accelerant

Shoppers increasingly ask an AI assistant what to buy before they reach a shelf or a product page. Agents are beginning to complete the purchase as well. In the agentic economy, the brand's first audience is a model that compares, reasons and recommends on the shopper's behalf.

That model behaves differently from a human in an aisle. It holds no affection for a pack design and has not watched forty years of advertising. It weighs what it can retrieve and justify: specification, price per unit, reviews, availability and the stated preferences of the person it serves. Asked for the best option, it will often name the one it can defend most easily.

For categories already sliding toward private label, this is an accelerant. A consumer trained to trade down now has an assistant that can do the trading down for them, at scale, and give a reasoned explanation each time. The habit the industry fears is being automated.

What happens to the role of brand

Predictions of the death of brands have a long and poor record. We are not adding to them. Our claim is narrower and more testable: the job brand does is being re-examined by a new evaluator, and much of the old job will not pass.

Brand has historically worked as a shortcut. It saved the shopper from thinking. An agent does the thinking anyway, so the shortcut loses value unless the brand also supplies reasons the agent can find and repeat. A brand that cannot be articulated by a machine risks being treated as a price premium without a case.

AIVO's research shows how wide this gap can be. In WP-2026-14, across more than 20,000 multi-turn probes covering over 200 brands, AI systems displaced a brand they had already cited in 87.3% of cases before reaching a recommendation. We call this the Linkage Gap. Being known by the model is common. Being chosen by it is rare.

That finding has a defined scope. It measures what AI systems recommend inside a controlled evaluation environment. It does not by itself show how many purchases follow those recommendations, or how much revenue moves as a result. Recommendation is the measured outcome; commercial impact has to be modelled separately and tested against sales data. Holding that distinction is what makes the measurement credible.

The brands that hold their position will be those whose premium survives the question an agent eventually asks: why this one, at this price?

Why the holiday season is the stress test

Black Friday falls on 27 November this year, less than seven weeks away. The holiday period concentrates major FMCG decisions into a compressed trading window. Gifting, entertaining and stock-up purchases all peak together.

Three conditions make this season an emerging stress test. Shoppers have spent three years learning to wait for promotion, so they arrive primed to compare. Promotional intensity gives AI systems more price signals to weigh and more reasons to suggest a switch. And AI is now part of the shopping journey for a growing share of consumers, mostly at the research and comparison stage. Agents that complete FMCG purchases on their own remain early, but they are being built and tested now.

The outcome will begin to appear in January sell-through data. Sell-through alone will not explain why a brand gained or lost ground. Price, promotion, distribution, availability and competitor activity all contribute. AI-mediated recommendation is another possible influence, and one most FMCG brands do not yet measure systematically.

An October baseline cannot prove how much revenue AI will cost a brand. It can establish something more immediate: where the brand stands inside AI-mediated decisions before the season begins, and whether that position moves during the weeks that matter most.

The case for a Brand Alpha baseline now

AIVO Brand Alpha measures how AI systems describe, compare and recommend brands within defined category and purchase scenarios. It complements established brand valuation frameworks such as BrandZ, Interbrand and Brand Finance by measuring a different source of competitive advantage: whether AI systems recommend the brand when a customer asks them to choose.

Each brand is scored out of 100. The headline is the AI Share Index, which compares a brand's share of AI recommendations with its share of category sales in the same market. The score is 50 times that ratio, capped at 100, so 50 is a calibrated parity point rather than a 50% recommendation rate. Because of the cap, any brand recommended at twice its sales share or more scores 100; the top score marks a ceiling, not the size of the lead. A score above 50 means AI systems recommend the brand more often than its market share would predict. Scores are comparable only within a stated category, market, set of AI systems, scenario set and measurement period.

Beneath the score sit measures of two different kinds. AI Win-Rate is measured directly: how often the brand is selected head to head at the decision stage. L.E.V. (LLM Equity Valuation) and Revenue at Risk are modelled from that measurement. L.E.V. expresses the brand's AI-layer position as a dollar value. Revenue at Risk estimates the revenue exposed to AI displacement:

\text{RaR} = \text{CAR} \times \text{APIR} \times \text{ADR}

CAR is category-relevant annual revenue. APIR is the share of category purchases in which AI plays a role; where no measured figure exists, it is shown as a scenario range. ADR is the measured rate at which AI systems displace the brand. The result is an estimate of gross revenue exposed within a defined window, not a forecast of revenue lost.

A baseline taken before the season gives the brand a comparable AI-layer reference point. Measured again in January, it shows whether the brand's position inside AI decisions held, improved or slipped. For that comparison to mean something, both measurements use the same documented protocol and scenario set, and any change in the AI systems or test conditions between them is recorded and its effect on comparability disclosed. Linking that movement to commercial results means reading it against sell-through, share, pricing, promotion and distribution data, which brands already hold. The baseline also leaves time to act on what it finds before the peak trading weeks.

The cost of waiting

FFA's analysis ends on a fair warning: companies that wait for the old model to return will spend years explaining a widening gap. We would add that some of that gap will open in a channel most FMCG brands do not yet measure.

Trust, habit and sensory preference will keep their hold on shoppers. The competitive advantage of a brand will depend increasingly on whether its value can also be expressed as a reason an AI system can recognise and defend. Which brands can do that is now an empirical question, and the holiday season will put it to every FMCG brand this year. A brand can measure its AI position in January whether or not it measured in October. What it cannot do in January is recover a pre-season benchmark, or act on one before the peak weeks have passed.

Method note

The AIVO Brand Alpha methodology, version 2.0, is published as a DOI-anchored working paper under AIVO Standard (doi:10.5281/zenodo.23183825). The L.E.V. framework is set out in a separate working paper (Zenodo record 22908818).

Linkage Gap findings are from AIVO working paper WP-2026-14. AIVO Standard papers are DOI-anchored and openly available for critique; they are not peer-reviewed. Market figures in the second section are drawn from FFA's H1 2026 review of the 55 largest FMCG companies and its Euromonitor-based growth algorithm analysis (FFA, NIQ Insider 2026 keynote); they are not AIVO data.