If you’ve been following pricing-regulation news this year, you might conclude that legal risk comes down to how advanced your model is: simpler systems are safe while sophisticated ones are exposed. I hear a version of this in almost every enterprise conversation, usually as some worry about “black box” AI. I understand why, but the statutes that actually passed in 2026 show why.

The most significant of those is New York’s One Fair Price Act. Along with the laws that followed in other states, it drew a line with one input: an individual consumer’s personal data. Retailers evaluating explainable AI pricing need to understand that distinction, because it changes which questions are worth asking a vendor.

What the One Fair Price act actually targets

New York’s One Fair Price Act (S.8623B/A.9349B) passed the State Legislature in June 2026 and awaits Governor Hochul’s signature, with a decision due by the end of the year. It targets the use of personal data, such as browsing history, location, or inferred income, to set tailored pricing. 

The act expressly preserves legitimate discounts, loyalty programs, and coupons. It also marks a shift from New York’s earlier approach. The Algorithmic Pricing Disclosure Act, in force since November 2025, allows personalized pricing as long as the retailer displays the line ‘THIS PRICE WAS SET BY AN ALGORITHM USING YOUR PERSONAL DATA.’ The One Fair Price Act would remove that disclosure route, since it bans the practice outright rather than just requiring a label.

Maryland and Connecticut passed related measures earlier in 2026. New Jersey followed, signing a measure on grocery and food-delivery pricing on July 23, 2026, though it does not take effect until August 2027.

The line is drawn around whether an individual’s personal data is used to individualize a price, and has little to do with how sophisticated the model is. A basic rules engine fed browsing histories is in scope; a demand model trained on aggregate transaction data is not.

That is why the “black box versus transparent” debate is the wrong frame for regulatory compliance, even though interpretability and explainability matter enormously for a different reason. What matters instead is what data crosses the boundary into the engine.

7Learnings runs on SKU-level transaction volumes, historical demand, inventory positions, and seasonal patterns, and does not ingest cookies, device identifiers, browsing histories, or individual customer profiles. 

One caveat: your obligations depend on your whole pricing stack rather than any single component. If a recommendation feeds a downstream personalization layer that uses customer-level data, the combined system is what a regulator assesses, and that is a conversation for your legal team.

Infographic outlining new pricing laws and AI vendor selection criteria.

Why "trust the algorithm" is not good enough

When the logic behind a price is opaque, no one can defend a change to finance or merchandising, and adoption stalls even when the model is sound. “Trust the algorithm” is not an answer a Head of Pricing can give a CFO, an auditor, or a regulator.

The accountability gap

A pricing decision the team cannot explain is a risk rather than an efficiency. Automating pricing is meant to produce better decisions at scale, but a recommendation nobody can account for just transfers the risk onto whoever approved it. The gap between “the model said so” and “here is why the price moved” is exactly what internal stakeholders, and now regulators, are pressing on.

When opacity stalls adoption

Opacity kills trust before the model gets a fair test. A price stuck on a zero-sales forecast or a competitor rule producing an obviously wrong number are small failures on their own, but with no way to see what happened, they read as proof the system can’t be trusted. 

Teams override recommendations, revert to manual work, and the investment never pays back. Explainability is what keeps a good model from failing on contact with the people meant to use it.

What explainable AI pricing actually means

Explainability is often reduced to a confidence score, so it is worth defining concretely. Explainable AI pricing means the platform surfaces which factors drove a specific recommendation, and the projected impact on profit and margin, rather than returning a bare number.

Seeing the drivers behind a recommended price

In practice, a merchandiser can open any SKU and see the reasoning in sequence: the demand forecast, the price elasticity, the stock position, the competitive context, and how each moved the final number. 

At 7Learnings, this lives in what we call the Rule Funnel, a step-by-step view of a price passing through every filter from forecast to recommendation, in the interface itself. This kind of feature importance view — showing which input variables had the most influence on the final price — is what separates responsible AI systems from black boxes. 

Explainability that lives in a PDF written after the fact is different from explainability you can pull up in a meeting for your CFO. For more on spotting rules dressed up as AI, see our guide on how to tell what is actually under the hood.

Explainability is more than a confidence score

A confidence score tells you the model is sure, but not why. “The model is 92% confident” is no more defensible to finance than a bare number. Real explainability names the drivers behind the recommendation and lets a human interrogate them. That’s the difference between a system you can operate with accountability and one you can only defer to.

How to prove the models are actually working

Explainability answers the first question. The second is whether the model actually works, and finance will ask it regardless of the regulatory climate. Vendor accuracy claims don’t settle that. Measured lift in a controlled test does.

Run an AB test to measure real lift

The design is straightforward: price part of a comparable assortment with the tool, price the rest with your existing process, then compare profit, revenue, and sell-through over the same window. A verified result carries far more weight than a projection.

When off-price retailer Outletcity ran predictive pricing against its end-of-season sell-off, including handling for broken size runs, it delivered 26% higher profit, 11% higher revenue, and 5% higher sales against a control group. The ROI was measured live. It’s a European result, not a US-specific promise, but the method is the point: the numbers came from a live test, not a lab projection.

This is also where human-in-the-loop AI governance and approval workflows earn their place. When merchandisers actually review and approve recommendations, that produces a record of who approved what and when, which matters as much for an audit as for internal trust.

Backtesting and simulating before prices go live

Backtesting a model against historical periods shows whether its recommendations would have improved past results, and simulating the forecasted impact of a price before it goes live lets a team see the projected effect on profit, margin, and sell-through in advance. This kind of ongoing model monitoring turns the decision to run a live test from a leap of faith into a calculated next step.

The questions to ask a vendor about their models

If you take one practical thing from this, make it a short checklist you can put to any AI pricing vendor, including us. Every question is structural, not a swipe at a named competitor. If a vendor can’t answer clearly, that is itself the answer.

  • What data crosses the boundary into your pricing engine, and does any of it identify an individual consumer?
  • Does the interface show which factors drove a specific recommendation and how they moved the price?
  • How do you measure lift, and can we run an AB test against our existing process to see it?
  • How do you handle private-label goods that have no competitor price to reference?
  • Who makes the final decision, is that approval real and recorded, and is our data isolated from other retailers’ rather than pooled into a shared model?

We built 7Learnings to answer each of these directly, and the reason rule-based tools and genuine machine learning algorithms answer them so differently is architectural rather than cosmetic.

For teams that want to see what a goal-driven, explainable approach to price optimization looks like in practice, our work on target steering and our predictive pricing guide both go further, as does our view of what retailers most often get wrong about AI.

Demand to see the reasoning and the proof

Explainable recommendations and provable lift are the real bar for AI pricing. Any tool that can’t meet both shouldn’t be trusted with your margin. The regulatory moment only reinforces it. 

The retailers handling the new pricing laws well are the ones who can answer for their pricing based on how their system is built: a clean data boundary keeps them clear of the surveillance-pricing rules, explainability lets them defend every number, and isolated data keeps them out of the collusion cases. Compliance is a property of the architecture more than of the policy binder.

Seeing why a model recommended a price, and the A/B-tested lift behind it, is the only real way to judge an AI pricing tool. Book a 7Learnings demo and put both the reasoning and the proof against your own assortment.

Frequently asked questions about explainable AI pricing

What is explainable AI in pricing?

Explainable AI pricing is an approach in which the system shows which factors drove a specific price recommendation, such as the demand forecast, price elasticity, stock position, and competitive context, along with the projected impact on profit and margin. 

Instead of returning a bare number or a confidence score, it lets a merchandiser trace how the recommendation was reached. The purpose is accountability: a price a team can explain is one it can defend to finance, to merchandising, and if needed to an auditor or regulator.

How can I tell if an AI pricing model is actually working?

The reliable test is measured lift in a controlled AB test rather than a vendor’s accuracy claim. Price a portion of a comparable assortment with the tool, price the rest with your current process, and compare profit, revenue, and sell-through over the same period. 

A verified result, such as Outletcity’s AB-tested 26% profit increase, carries far more weight than a projection. Backtesting and pre-launch simulation build confidence, but a live controlled test against a genuine control group is the standard worth holding out for.

What is the difference between explainable AI pricing and a black-box model?

A black-box model returns a price without showing its reasoning, so the team has to take the number on trust. Explainable AI pricing surfaces the drivers behind each recommendation and the projected impact, so a human can see why the price moved and interrogate it. 

The practical difference shows up when someone asks you to justify a price. With a black box, you cannot, but an explainable system lets you pull the reasoning up in the meeting. Explainability is about accountability, which is why it matters even where regulation does not require it.

Does New York’s One Fair Price Act ban AI or dynamic pricing?

No. The One Fair Price Act targets surveillance pricing, meaning the use of an individual’s personal data to set a price tailored to that person. Dynamic pricing based on aggregate market signals, such as supply and demand, inventory, competitor prices, and seasonality, is preserved, and the statutes carve those factors out explicitly. 

A model that runs on aggregate transaction data rather than individual personal data is aimed at a different practice from the one the law restricts. How the law applies to your specific setup is a question for your counsel.

Can AI pricing recommendations be audited?

Yes, if the system is built for it. Auditing a price recommendation requires two things: a record of which factors drove it, and a record of who reviewed and approved it. An explainable platform provides the first through its reasoning trail and traceability logs, and the second through a human-in-the-loop approval log. 

This is increasingly what auditors, boards, and regulators making informal enquiries expect to see, and it is far easier to produce from a system that captures the reasoning as it prices than to reconstruct after the fact.