Why FMCG's biggest trust problem is the most mundane one
Paul F. Accornero · 03.06.2026 · 11min read
Content
By Paul F. Accornero · Founder, The AI Praxis · Author, The Algorithmic Shopper (St. Martin's Press, 2027) ·
A shopper opens her grocery app on a Tuesday morning. She asks the agent to reorder the weekly basket. The agent does the job. Twelve items arrive that evening. Eleven are correct. One is wrong. The Greek-style yogurt she has bought for four years has been quietly swapped for a private-label substitute the algorithm decided was equivalent.
She returns it. She is not furious. She is mildly annoyed. The next week, she still uses the agent, but she checks the order more carefully before approving it. Two weeks later, she has stopped trusting it for anything other than bottled water and toilet paper.
Multiply this across a market of sixty million households. That is the accuracy paradox in one sentence. The thing that is going to slow the agentic grocery revolution is not the thing the boardroom is worried about.
What the data says
The single biggest fear consumers report about agentic shopping is not data privacy. It is not AI ethics. It is not job displacement. It is not algorithmic bias. It is, by a clear margin, that the agent will simply buy the wrong item.
39.0% of UK consumers and 38.8% of US consumers name this as their primary concern. That is the largest single answer in both markets. It outranks privacy. It outranks ethics. It outranks every existential question the AI commentariat spends its time on.
Only 26% of UK consumers and 30% of US consumers currently report that they trust an AI agent to shop for them. The trust ceiling is real, and it is structural. But the cause of the ceiling is much smaller and much more boring than the industry conversation suggests. The cause is operational reliability at the SKU level.
This is the accuracy paradox. The most-discussed problem in AI commerce is the one consumers are not particularly worried about. The least-discussed problem is the one capping the entire market.
Why this matters
The implication for FMCG and grocery brands is structural, not tactical.
In the world of the human shopper, brand equity buffers operational error. A consumer who loves a brand will forgive a stockout, a packaging variation, a missing promotion. Goodwill compounds across thousands of small disappointments. The brand survives the noise.
In the world of the algorithmic shopper, there is no buffer. The agent does not love anything. The agent runs a probability calculation on every order. If the agent buys the wrong yogurt twice, the consumer reduces the agent's autonomy by a notch. If the agent buys the wrong yogurt four times, the consumer takes the basket back. Brand equity does not save the agent. Operational reliability does.
This means data quality is no longer an IT line item. It is the precondition for the brand existing in the agent channel at all.
A brand with a 95% accurate product feed will quietly take share from a brand with an 85% accurate feed, even if the second brand is the more loved one. Because the consumer never sees the brand competition. The agent does. And the agent picks the brand that does not embarrass it.
The three layers of trust
This is the framework Appinio's data crystallises. Trust in agentic commerce is not one thing. It builds in sequence. The first layer is the gate. The second and third only open when the first is solved.
Layer 1. Functional Trust.
Does the agent buy the right item, at the right price, on time, in the right condition? This is the layer the 39% accuracy fear lives in. It is the bottleneck. The entire current market is stuck here.
Layer 2. Fiduciary Trust.
Is the agent buying for me, or for the platform that built the agent? Is it taking margin I cannot see? Is the substitution logic optimised for my wallet or for the retailer's? Consumers do not ask this question yet, because they have not got past Layer 1. They will. When they do, the conversation moves from product accuracy to platform incentives, and a new tier of brand and retail problems opens up.
Layer 3. Relational Trust.
Does the agent know me? Does it understand my preferences, my routines, my household, the things I have never told it? This is where the agent becomes a personal procurement officer rather than a search box with checkout. Almost nobody is here today. The brands that get here first own the category for a decade.
Most boards spend their AI conversations debating Layer 3 risks (bias, ethics, personalisation, manipulation) while the market is still entirely held back by Layer 1 failure (the agent buys the wrong yogurt).
What the Agent Decision Preference Stack® shows
Across more than one thousand simulated agent decisions and ten product categories, operational factors weigh 30% to 70% more heavily than brand reputation in AI purchase decisions. Data quality. Discoverability. Decisional clarity. Delivery reliability. These four dimensions, the Four D's™, are the new commercial substrate.
Brand reputation has not vanished from the equation. It still matters. But it matters less than the operational layer underneath it, and it matters far less than most marketing teams have been told. The competitive basis has inverted. Operations is now more important than marketing. Not in slogan terms. In structural terms. The agent reads operations as a primary signal and brand as a secondary one.
This is why the accuracy paradox cuts deeper than it first appears. The 39% concern is not a customer-service problem. It is a structural signal that the entire commercial architecture of FMCG has the priority list backwards.
So what...
The boards spending the bulk of their AI energy on regulatory risk, ethics committees, and reputational defence are protecting against a layer of the trust stack that consumers have not yet reached. Meanwhile, the layer that is keeping their addressable market capped at one consumer in four is sitting in operations, supply chain, and master data, where most FMCG companies have under-invested for a decade.
There is a window. The brands that solve Functional Trust in the next eighteen months become the default choices when the agent expands its autonomy. The brands that do not are quietly substituted out before their consumers ever notice. Once a category default is set in the agent, it is structurally hard to dislodge. The first-mover advantage in the agent layer is durable in a way the first-mover advantage in the human layer never was, because the agent does not forget and the agent does not switch on a whim.
Three practical moves for the brand director reading this on a Monday.
1. Audit your product data feed. What percentage of your SKUs have complete, current, machine-readable structured data across every retailer you list with? Most companies discover the answer is below 40%. That is the gap the agent reads first.
2. Measure your substitution rate, not your share. If your product is being substituted at the basket level when the agent picks, your market share number is lying to you. The substitution rate is the new leading indicator.
3. Pick a category to win the agent on, not a market. Trying to be agent-readable across every category in your portfolio at once is operationally impossible. Pick the one where your data is cleanest, your supply chain is most reliable, and your substitution rate is lowest. Win that one. Earn the right to expand.
One more thing
The accuracy paradox is a temporary condition. Layer 1 will be solved, by someone, in some categories, in the next two to three years. When it is, the conversation moves to Layer 2 and Layer 3, and the brands that have not built the data spine, the operational backbone, and the API-readable product catalogue are not in the conversation at all.
The British shopper who returned the wrong yogurt this morning is not going to write a complaint. She is going to quietly reduce the autonomy she gives the agent. When enough shoppers do that, the entire agent commerce sector hits a ceiling that nobody can break through with better marketing.
The brands that break the ceiling will not do it with a creative campaign. They will do it with master data.
This is the part that decides who is on the shelf in 2030.
About the Author
Paul F. Accornero is a scholar-practitioner working on the structural transformation of commerce by AI agents. He is the founder of The AI Praxis, an independent research and advisory platform on agentic commerce, and the author of The Algorithmic Shopper, forthcoming from St. Martin's Press in 2027.
His SSRN portfolio is in the Top 2% of authors worldwide, with 23 distributed working papers on agentic commerce, algorithmic readiness, and the transition from SEO to Agent Intent Optimisation (AIO). He has published in California Management Review Insights (FT50 outlet) and is invited reviewer for the Journal of the Academy of Marketing Science (FT50, CABS 4) and Big Data Society. He guest-lectures on AI strategy & Agentic Commerce at Harvard.
The data in this article comes from The Appinio Agentic Grocery Commerce Report, a transatlantic study of n=2,000 consumers fielded between 18 March 2026 and 25 March 2026 across the United States and United Kingdom.
The frameworks discussed (the Shopper Schism®, the Agent Decision Preference Stack®, the Four D's™, the Three Layers of Trust) are drawn from Paul F. Accornero's forthcoming book The Algorithmic Shopper (St. Martin's Press, April 2027), U.S. Copyright Reg. No. TXu 2-507-027.
Run the Algorithmic Readiness Audit® at algorithmicreadiness.ai to see how your brand is currently rated by the agents that are already in market.
Get facts and figures 🧠
Want to see more data insights? Our reports are just the right thing for you!
