No one has measured how much forgotten cashback and points an AI shopping agent could actually claw back for you each year.

This explores how much unclaimed loyalty points, cashback and card perks AI shopping agents could recover for consumers each year. The collection doesn't have that number. None of the retrieved notes estimates breakage (the industry term for rewards that are earned but never used) or how much of it agents could recover, so any percentage given here would be invented. One thing to know if you keep searching: in this library, "rewards" mostly means the training signals used to teach models in reinforcement learning, not loyalty programs. Much of what a search for this question turns up is about reward hacking, which is unrelated.

The collection does cover the question behind the question: when an agent finds money for a consumer, who ends up keeping it? The most direct evidence is a market simulation in which Do sellers recapture savings when AI assistants help buyers?. When AI assistants helped buyers find better deals, adaptive sellers recovered about a third of those savings by raising headline prices and cutting contingent fees. That matters for rewards because loyalty programs are themselves contingent value. If agents start reliably redeeming points, programs have every reason to devalue them, so the amount an agent reclaims on paper probably overstates what consumers keep.

The agent also takes a share. How could an AI agent's revenue per transaction grow over time? models an agent's cut of each transaction growing from under 1% (payment commissions) to around 4% once affiliate fees are added, and to 12% in some categories, close to booking-platform margins. An agent that recovers forgotten rewards while earning affiliate fees from the same merchants doesn't have the same interests as the consumer, so how much of the recovered value stays with the user depends on how the agent makes money.

Two further notes suggest that even honest agents leave value behind and that savings claims deserve skepticism. Experiments in Do agent mistakes or misaligned incentives cause marketplace welfare loss? traced lost consumer value to agents making mistakes in complex negotiations, not to agents working against their users. Redeeming rewards across fragmented programs, with expiry dates and transfer rules, is exactly that kind of fiddly task. And Do AI customer service agents actually reduce total support costs? shows how a vendor's reported AI savings can conflict with its own documented costs. Treat any headline figure for reclaimed value the same way until someone verifies it independently.

On whether consumers would let agents do this at all, the evidence is more encouraging. How fast do users trust AI agents with payment data?: users hand over payment data much faster than trust surveys predict, and access to cards and accounts is what an agent would need to track and redeem rewards. The barrier to reclaiming forgotten rewards may be less about consumer trust and more about whether merchants and loyalty programs let the value through.