Incrementality Measurement for AI Shopping Agents
As AI agents begin shopping and transacting for people, merchants will need to know whether those agents actually created new demand or simply inserted themselves into purchases that would have happened anyway. Build a measurement layer that estimates each customer’s likely baseline behavior, then calculates how much an agent increased purchase frequency, basket size, subscription adoption, category expansion, or customer reactivation beyond that baseline. This matters now because agents, platforms, and commerce networks will all want to be paid for the revenue they influence, but attribution alone cannot prove economic value. Imagine an agent generating $14 million in merchant sales while the system shows that $8 million would have happened anyway and $6 million was truly incremental.

Description
No additional description provided.
The Pitch
AI commerce will make the old attribution problem worse at exactly the moment the stakes get larger. When a human clicks an ad, there is at least some visible trail of intent. When an agent buys on someone’s behalf, the merchant may see only a transaction wrapped in machine logic. The agent can claim influence, the platform can claim distribution, and the merchant can be left paying for a sale that was already inevitable. The valuable company is not the one that records who touched the order, but the one that can say how much of the order should economically count.
The deeper shift is that merchants will be forced to separate access from persuasion. Today, many intermediaries are paid because they control the doorway to demand. But agents could become the doorway to almost everything: replenishment, comparison shopping, subscription management, gifting, travel, insurance, household procurement. If every doorway charges as if it created the purchase, margins get taxed by entities that may only be routing existing intent. A neutral incrementality layer becomes a kind of commercial immune system, helping merchants distinguish useful demand generation from expensive toll collection.
The most interesting customers for this product would not be chasing perfect measurement; they would be trying to preserve bargaining power. Once agents can steer consumer choices, merchants will face a new class of counterparties that can plausibly say, “We brought you the customer.” The merchant’s response cannot be philosophical. It needs an empirical counterweight. Even an imperfect but trusted estimate of baseline behavior changes the negotiation, because it turns agent compensation from a story about gross sales into a discussion about lift.
This also has the unusual property of becoming more valuable as the ecosystem gets messier. Each new agent, wallet, shopping interface, browser assistant, marketplace, and loyalty layer adds another party claiming influence over the same customer. The result is not just measurement complexity, but political complexity: no participant can be trusted to grade its own homework. A third-party system that merchants, agents, and commerce platforms can all reference could become part of the financial plumbing of AI-mediated trade.
The lasting business insight is that agent commerce will not only create new ways to buy; it will create new arguments over who deserves credit. Wherever money is redistributed based on invisible influence, the party that defines “incremental” sits close to the profit pool.