How does the money AI shopping agents actually move compare to the flood of hype around them?

This explores how much money AI agents actually move through payment systems today, and how that compares with ordinary human-initiated payments. The short answer is that the corpus doesn't contain a direct measurement of agent payment volume next to human payment volume. What it does have is a set of growth numbers that look dramatic, projections about where things could go, and evidence about why real transaction volume is still small.

The most widely quoted growth figure needs a warning label. Adobe reported that traffic referred to retailers by generative AI grew 693.4% over the 2025 holiday season. Lipsman's reanalysis shows that this traffic went from about 0.1% to 0.6% of all ecommerce traffic Does Adobe's AI referral traffic growth statistic hold up under scrutiny?. The percentage is mathematically correct, but it starts from almost zero. It also measures referrals, meaning a person clicking through from a chatbot, not payments that an agent made on its own. It's worth recognizing this pattern whenever you see a striking agentic-commerce growth rate: big multipliers on a tiny base say more about how new the category is than about how big it is.

Forecasts are where the large dollar figures come from. McKinsey projects that agentic commerce could handle up to $1 trillion of US consumer retail revenue by 2030, and $3–5 trillion globally Could agentic commerce reach one trillion dollars by 2030?. That's a scenario about the future, not a measurement of today, and it depends on protocols for agent coordination and payment becoming widely adopted. Jaipuria's model looks at the money flow from the agent's side instead. An agent might start by earning less than 0.5% per transaction in payment fees, then add affiliate fees, and eventually reach booking-platform-like margins, a blended take of roughly 0.8% rising to 4% How could an AI agent's revenue per transaction grow over time?. That tells you how agents could make money as volume grows, not how much volume exists now.

The corpus offers two explanations for why actual volume is lagging, and neither is about the technology. The first is trust. In a Visa-commissioned Harris Poll, 72% of US adults used AI assistants but only 23% trusted AI to carry out payments. That number rose to 61% when a familiar payments brand was attached Does naming a payments brand increase trust in AI transactions?. The second is infrastructure. Fraud systems were built around signals of human identity and intent Does agent-initiated commerce break existing payment infrastructure?, and they rely on behavioral cues like typing rhythm that disappear when an agent does the checkout. That leaves merchants carrying the fraud risk Does agentic commerce shift fraud liability onto merchants?. The IMF argues that the AI's probabilistic decision should be kept separate from the predictable steps of authorizing and settling a payment Should AI agents make payment decisions independently?. In other words, agents may decide what to buy, while the money keeps moving through the same channels it always has.

That last point suggests an unexpected answer to the question. "Agent payments" may never show up as a separate category that you can count. In travel, AI agents haven't replaced the online travel agencies. Those agencies remain the merchant of record behind the scenes, and Expedia's business-to-business segment grew 24% year over year Are AI agents actually removing online travel agencies from bookings?. If agent spending flows through existing intermediaries and card networks, its growth will appear inside their numbers rather than as a separate line next to human payments. That may help explain why a clean comparison is so hard to find.