The Missing Layer Between AI Recommendations and Retail Decisions

A repricing recommendation flashes on a merchant’s dashboard. The model has years of data behind it, the number looks reasonable, and still the decision stalls. The merchant overrides it. Then does the same thing the following week.

That pattern is showing up across retailers that have already spent heavily on artificial intelligence. Deloitte’s 2026 State of AI in the Enterprise report found two-thirds of organizations are already capturing productivity gains from AI, but only about a third are using it to genuinely transform how the business runs. McKinsey’s 2026 AI Trust Maturity Survey found a similar gap. Only about a third of organizations have reached a mature level of responsible AI governance — i.e., the practices that make a system trustworthy enough to act on its own in the first place.

Here’s the odd part: The better the models get, the more often people override them. That isn’t a data problem. It’s a decision problem. Retailers have spent a decade making AI smarter and almost no time rethinking how people and machines decide together.

AI accuracy was never really the issue. Decision accountability is. Accountability is something you design into an operating model, not something you train into a model. That’s the shift the retailers ahead of this curve have already made. They spend less energy tuning the model and more building the layer around it so the merchant isn’t left to reconstruct its logic on faith.

Building the Trust Layer

Three things make it work in practice:

1. Explainability inside the workflow. Not a separate dashboard nobody opens, but the reasoning delivered right where the decision happens — a pricing lead seeing the demand signals, competitor moves, and margin guardrails behind a recommendation as they act on it. At one omnichannel apparel retailer I worked with, the pricing team was overriding close to two-thirds of automated repricing recommendations, mostly because the reasoning behind them lived in a separate analytics dashboard nobody had time to open mid-shift. Once that context (demand signals, competitor pricing, margin floors) moved into the repricing tool itself, the override rate dropped below 30 percent within a single quarter. Put the reasoning in context, and the friction starts to fall away.
2. Human control by design. Automation changes oversight; it doesn’t remove it. Some calls run on their own inside clear guardrails, while others need a sign-off. That line has to be drawn deliberately. Ship-from-store fulfillment is a clean example. Optimized purely for delivery speed, a system can route an order through three facilities to shave half a day off arrival, at a cost that erases the margin on the sale. Automation still makes sense here. What it needs is guardrails that weigh cost-to-serve alongside speed, as well as a habit of treating overrides as data instead of defeat.
3. Shared accountability. Retail decisions cross legal, compliance, brand, and finance, so the record behind them — how the call was made, what data drove it, who could have stepped in — has to be shared from day one. Retail media is where this shows up fastest. An ad-placement engine optimized purely for auction revenue can start competing with a merchant’s own control of the brand, surfacing a competitor’s sponsored listing on a hero product’s page or discounting placement in ways marketing never signed off on. Without a shared record of who set which guardrail, that fight replays every quarter instead of getting resolved once. Done well, this kind of governance isn’t a brake; it’s what lets leaders roll a model out across categories and regions instead of re-arguing it in every market.

The Economic Cost of Mistrust

Skip any piece of that layer and the old pattern comes right back — except now it’s harder to see because the retailer has already paid for the platform. This mistrust never shows up as a line item, which is why it goes unmanaged. But it has a P&L. The repricing tool nobody trusts becomes a margin decision made late or made on gut feel instead of signal. The fulfillment guardrail nobody built is cost-to-serve eating into a sale’s margin, order after order, with no one catching it. The retail media fight nobody resolved is the same negotiation, and the same lost placement revenue, repeating every quarter instead of getting settled once.

Underneath all three sits the priciest line: The platform and the model are already paid for, and mistrust parked in advisory mode means none of that spend earns its keep. Multiply that across a dozen initiatives and it’s a real drag on EBITDA.

Retailers are overinvesting in models and underinvesting in operational trust. The next point of model accuracy is worth far less than the first point of trust.

Where Retail AI is Heading

This is about to get more urgent. Retail is shifting from AI that recommends to AI that acts, such as agentic commerce, where systems propose the repricing, the fulfillment routing, and the ad placement, then execute inside set limits. You can’t hand a decision to an agent you don’t trust to explain itself. Operational trust decides whether tomorrow’s AI is even possible.

For retail leaders, this changes where the attention goes. The job isn’t mainly about squeezing more performance out of the model. It’s about building the layer around it, including explainability in every workflow, clear lines for what runs on its own vs. what needs a person, and a shared record of who’s accountable for what. It also changes how you keep score. The real measure is adoption, how much faster decisions get made because of it.

Retailers that crack this spend less time defending their models and more time improving them, because teams use recommendations they understand and can push back on. The ones that don’t stay stuck in pilots that never deliver.

A competitive advantage in retail won’t come from who has the best AI model, but from who builds a layer of trust strong enough to let merchants act on the model without a second thought.

Vasanth Menon is Retail leader at Brillio, a Bain Capital company specializing in digital transformation solutions.