A couple in Vancouver asks their AI agent to find a weekend getaway in Whistler that's dog-friendly, has a hot tub, and is under $400 per night. The agent queries a data layer, matches them to Creekside Lodge, and presents the recommendation with details: dogs of any size welcome, private hot tub on the deck, $340 per night this weekend, ten-minute walk to the village. The couple books directly on the lodge's website.
Now: who gets credit for that booking?
The agent made the recommendation. The data layer provided the structured information that made the match possible. The lodge provided the data in the first place and delivered the product. The couple made the decision. In the current web, this question has messy but workable answers — UTM parameters in the URL, referral cookies, promo codes. The booking came from Google Ads, or from a TripAdvisor click, or from an Instagram story link. The attribution chain is imperfect but traceable.
In agent commerce, the attribution chain is invisible. The couple didn't click a link in a search result. They didn't follow a referral URL. They asked their agent, got a recommendation, and booked directly. There's no UTM parameter. There's no cookie. There's no trace of which data source powered the recommendation — unless the system is designed to track it.
Why attribution matters
Attribution isn't an academic question. It determines who captures value and who gets paid. In the current digital economy, attribution drives billions of dollars:
Google makes money because it can prove (or at least argue convincingly) that its ads drove a click that led to a conversion. An advertiser who can't attribute sales to Google Ads stops spending on Google Ads. Attribution is the foundation of the entire performance advertising industry.
OTAs make money because every booking flows through their platform, creating automatic attribution. When a guest books through Booking.com, the attribution is inherent in the transaction — the OTA processed it, so the OTA earns its commission. No tracking required.
Affiliate marketers make money because their unique referral links create an attribution trail. A travel blogger who sends readers to a hotel booking page with an affiliate link earns a commission on every conversion that trail produces.
If agent commerce can't solve attribution, it can't build a sustainable economic model. Data layers can't charge for matches they can't prove they made. Businesses can't know which data channels drive real customers. Agents can't optimize their recommendations without knowing which ones actually led to outcomes. The entire value chain depends on knowing who did what and who deserves credit.
The agent commerce attribution chain
In agent-mediated commerce, the attribution chain has four distinct links, and each needs to be tracked:
1. The query. A consumer expresses intent to their agent: "Find me a dog-friendly hotel in Whistler under $400." This is the origin of the attribution chain — specific, timestamped, tied to a consumer profile.
2. The match. The agent queries a data layer and receives structured data about Creekside Lodge. The data layer logs this as a match event: match ID, timestamp, which business was returned, which query parameters were matched, and the confidence level. This is the data layer's contribution — it connected the consumer's intent with the business's data.
3. The recommendation. The agent presents Creekside Lodge to the consumer, possibly alongside two other options. The agent logs which businesses it recommended, in what order, with what framing. This is the agent's contribution — it curated and presented the options.
4. The outcome. The consumer books at Creekside Lodge. The lodge confirms a new booking that matches the parameters of the recommendation — same dates, same room type, same guest name or email. This is the outcome confirmation — the proof that the attribution chain produced a real transaction.
Each link in the chain is necessary. Without the query, there's no intent to trace. Without the match, there's no proof the data layer contributed. Without the recommendation, there's no proof the agent presented the option. Without the outcome, there's no proof the recommendation converted.
How the technical architecture works
Solving attribution in agent commerce requires a lightweight but robust tracking system. Here's how the pieces fit together.
Match IDs. When a data layer returns a business match to an agent query, it generates a unique match ID — a random identifier tied to the specific query, the specific business returned, and the timestamp. This match ID travels with the recommendation through the rest of the chain.
match_id: m_7f3a9b2e
query: "dog-friendly hotel, Whistler, <$400"
business: creekside-lodge-whistler
matched_on: [pet_policy, price, location, availability]
timestamp: 2026-02-19T14:23:07Z
data_source: pawlo-mcpRecommendation tokens. When the agent presents the recommendation to the consumer, it includes the match ID in a recommendation token — a lightweight identifier that the consumer's booking action can reference. This could be embedded in a booking link, passed as a reference code, or logged by the agent alongside the consumer's action.
Confirmation callbacks. When the business confirms an outcome — a booking, an appointment, a purchase — it can match the outcome against known recommendation tokens. "We received a booking from Jane Chen for February 22-24 — does this match a recent recommendation?" The data layer checks: yes, match ID m_7f3a9b2e recommended Creekside Lodge to a query matching this profile on this date range. Attribution confirmed.
Outcome tracking. The confirmed attribution is logged as a completed chain: query → match → recommendation → outcome. Each participant in the chain — the data layer that provided the match, the agent that made the recommendation, and the business that delivered the service — has a verifiable record of their contribution.
The multi-touch reality
Attribution in agent commerce isn't always a clean single chain. A consumer might ask three different agents about Whistler hotels. Two of them might recommend Creekside Lodge, using data from the same data layer. The consumer might also see the lodge mentioned in a blog post, or get a recommendation from a friend. Who gets credit?
The answer is the same one digital marketing arrived at after years of debate: multi-touch attribution. The data layer gets credit for providing the match data that powered the recommendations. Each agent that recommended the lodge gets partial credit based on timing and context. The friend's recommendation and the blog post exist outside the tracked chain and don't factor into the agent attribution model.
This isn't perfectly fair, but it's functional. The alternative — no attribution at all — is worse for everyone. A system that gives 60% credit to the first agent that recommended the lodge and 40% to the second, with the data layer receiving a match fee regardless, is imperfect but workable. It's vastly better than a system where the couple books directly, the lodge has no idea which channel drove the booking, and neither the data layer nor the agents can demonstrate their value.
Why transparent attribution benefits everyone
For businesses: Attribution tells you which data channels actually drive customers to your door. If agent-mediated recommendations consistently produce bookings that convert and guests that return, you know your structured data investment is working. If a particular data channel produces matches that don't convert — the agent recommended you, but the consumer didn't book — you can investigate whether your data is misrepresenting something. Attribution is feedback, and feedback is how businesses improve.
For agents: Attribution tells you which recommendations actually work. An agent that can track its recommendation-to-outcome conversion rate can optimize: maybe its confidence threshold is too low and it's recommending businesses that don't convert, or maybe it's not surfacing the right details in its presentation. Without attribution, the agent is guessing. With it, the agent is learning.
For data layers: Attribution proves value. A data layer that can show a business: "Last month, our data powered 47 agent recommendations for your property, 31 of which converted to bookings" has a clear value proposition. The business can compare this against its OTA commission costs, its Google Ads spend, and its other acquisition channels. Transparent attribution makes the data layer's value visible and measurable.
For consumers: Attribution doesn't directly affect the consumer's experience, but it indirectly improves it. When agents can track which recommendations lead to good outcomes, they get better at recommending. When businesses can see which data channels work, they invest more in accurate, detailed data. When data layers can prove their value, they invest more in data quality. The entire system gets better when attribution works.
The privacy balance
Attribution in agent commerce has to work without compromising consumer privacy. The system doesn't need to know the consumer's identity to track attribution — it needs to know that a query led to a match, a match led to a recommendation, and a recommendation led to an outcome.
Match IDs are anonymous. They link a query to a business, not a query to a person. Recommendation tokens are ephemeral — they exist to confirm the attribution chain and then expire. Outcome confirmations can be aggregated and anonymized: the data layer knows that 31 of 47 recommendations converted last month, but it doesn't need to know who those 31 people were.
The attribution system tracks the flow of information, not the identity of individuals. This is fundamentally different from the cookie-based tracking of the current web, where attribution requires following a specific person across websites and building a behavioral profile. Agent attribution tracks the chain of data → match → recommendation → outcome, and each link in the chain can be verified without identifying the consumer.
This is possible because the agent acts on the consumer's behalf. The agent knows who the consumer is, but the data layer doesn't need to. The business knows who booked, but the data layer only needs to know that the booking happened. The attribution chain is verified through match IDs and outcome confirmation, not through personal data.
Solving attribution in agent commerce isn't just a technical challenge — it's a prerequisite for the entire economy to function. Without attribution, data layers can't prove value, agents can't optimize, and businesses can't evaluate channels. With attribution, every participant in the chain can demonstrate their contribution, measure their performance, and improve. The businesses, agents, and data layers that invest in transparent attribution now will build the trust infrastructure that agent commerce needs to scale.