The next time two people see different prices for the same product, the explanation may not be demand, inventory, or a weekend promotion. It may be an algorithm quietly deciding what each customer is willing to pay.
AI-driven pricing is moving from experiment to everyday commerce. Retailers and ecommerce teams are using machine learning for personalised offers, recommendations, demand forecasts, and dynamic pricing. The technology is powerful. The trust model around it is still half-built.
That gap is where the trouble starts.
A Deloitte-FICCI report published today calls for clearer governance around personalised pricing, recommendation engines, automated customer engagement, and AI-generated advertising. It also makes the part many companies would rather avoid explicit: the business using the system remains accountable for the consumer outcome.
[INSIGHT] Personalisation feels helpful until the customer realises the system knows more about their willingness to pay than they do.
The Price Is Not the Problem
Consumers already accept that prices change. Flights, hotels, rideshares, concert tickets, and food delivery have trained people to expect movement. The complaint is not always that the number went up. It is that the reason feels hidden.
There is a difference between dynamic pricing and opaque pricing. A hotel charging more during a major event has a visible explanation. An ecommerce site quietly charging one customer more because their browsing history suggests urgency has a different moral texture, even if both systems can be described as optimisation.
This is why AI-driven pricing is a brand problem, not just a revenue problem. A checkout page is also a statement about the relationship between a company and its customer. If the customer believes the company is testing their limits rather than serving their needs, the transaction can succeed while the relationship fails.
That failure rarely appears in the same dashboard as conversion rate. It shows up later, in lower repeat purchase, more complaints, more price comparison, and a growing instinct to use an intermediary that feels more neutral. I wrote about a similar measurement gap in the AI search measurement crisis. The number teams celebrate is often the number that hides the damage.
AI-Driven Pricing Needs A Reason
The first rule for using AI in pricing is simple: every material price difference needs a human-readable reason.
That does not mean exposing model weights or publishing a technical paper at checkout. It means being able to answer basic questions without hiding behind the phrase "the system decided." What changed? Which inputs mattered? Is the difference based on inventory, timing, location, membership, delivery cost, or an inferred willingness to pay?
If a company cannot explain the answer to its own customer support team, it is not ready to make the decision at scale.
The Deloitte-FICCI report recommends clearer risk classifications, disclosure, traceability, auditability, human oversight, and escalation paths. Those sound like governance terms, but they map to ordinary customer moments. Someone sees a different offer. Someone challenges a recommendation. Someone gets a price that appears inconsistent. The company needs a record of what happened and a person who can intervene.
The companies that treat those moments as edge cases will eventually discover that edge cases are where reputation lives.
The Personalisation Trap
Personalisation is usually sold as convenience. Show people what matters. Remove friction. Make the experience more relevant. Those benefits are real, but they are not free.
Every personalised system makes a judgment about what a person is likely to want, afford, or tolerate. Most customers accept that judgment when it improves relevance. They become uncomfortable when the same hidden judgment affects access, price, or urgency.
The line is not fixed. A grocery app suggesting a cheaper substitute is helpful. A finance product changing an offer based on inferred vulnerability deserves far more scrutiny. A travel site using location to estimate demand may be ordinary. A health product using sensitive signals to charge more is a different category entirely.
The technology may be similar. The social consequence is not.
That is why a single company-wide AI policy is rarely enough. Pricing, advertising, customer support, and recommendation systems do not create the same level of risk. The governance needs to follow the use case, not just the vendor contract.
The same mistake is appearing in marketing automation. Teams assume that if a model performs well on average, the system is safe. But averages flatten the people who receive the worst outcome. The failure of AI marketing automation is often less about a bad model than a weak operating discipline around the model.
Trust Has A Memory
A customer may forgive a bad recommendation. They are less likely to forgive the feeling that a company manipulated them.
That distinction matters because trust compounds in both directions. A transparent explanation can turn a confusing price into a reasonable one. A hidden rule can turn a small price difference into evidence that every other interaction is probably engineered too.
The same applies to AI-generated advertising. The Deloitte-FICCI report points to disclosure and claim substantiation as the creator economy expands. India’s influencer marketing industry was estimated at Rs 3,000 to 3,500 crore in 2025 and is projected to reach Rs 4,500 to 5,000 crore by 2027, according to the report. It also cites a 2025 ASCI finding that 69 percent of top influencers reviewed failed to meet disclosure requirements.
Those figures point to a larger issue. Consumers are not just evaluating whether a message is persuasive. They are evaluating whether the message is honest about its origin and incentives. A hidden sponsored post and a hidden personalised price are different events, but they create the same suspicion: what else is being withheld?
Brands that keep treating disclosure as a compliance checkbox will miss the strategic point. Transparency is not an apology added after the system is built. It is part of the experience design.
The New Brand Advantage
The competitive advantage will not belong to the company with the most aggressive pricing model. It will belong to the company that can use personalisation without making people feel profiled, sorted, or cornered.
That requires a few decisions that are not especially glamorous.
Keep a plain-language explanation for every high-impact automated decision. Give customer-facing teams a real escalation path instead of a script that says the algorithm is working as intended. Test price and offer systems across customer groups before release, then keep testing them after behaviour changes. Store enough evidence to reconstruct a decision months later.
Most important, set boundaries before the revenue team asks for them. Do not use sensitive attributes simply because the model can. Do not turn urgency signals into a quiet penalty. Do not call a price personalised if the customer has no meaningful way to understand the difference.
This is where brand strategy and AI governance stop being separate conversations. A brand promise says what customers should expect from a company. Governance decides whether the machines keep that promise when nobody is watching.
The companies that get this right will not sound like they are defending an algorithm. They will sound like they designed the system around a human being who deserves an answer.
The market is about to find out which brands believe that, and which ones only believed it before the margin opportunity appeared.