Negotiation Traces as Defensible Data: The Corpus No Clickstream Can Match

For the last twenty years, the most valuable data assets in consumer technology have been clickstreams. Search histories at Google. Social-graph engagement at Meta. Purchase patterns at Amazon. Watch behavior at Netflix. Each of these companies built durable competitive advantages on the proposition that nobody else could see what their users did at the same scale or specificity, and that this private observability compounded into better products that locked in further usage.

Clickstreams are about to be supplanted as the most valuable consumer-data category. The successor is agent-to-agent negotiation traces: the signed, structured records of agents acting on principals' behalf and producing outcomes that are inspectable, auditable, and reusable as training data. This paper explains why the negotiation trace is structurally a more defensible data asset than the clickstream, what kinds of companies will accumulate these traces at scale, and why this matters for the competitive dynamics of the next decade.

What a clickstream actually captures

It is worth being precise about what a clickstream contains, because the limitations are easy to miss when the term is used loosely. A clickstream records the discrete user actions: clicks, taps, scrolls, queries, dwell times, that occur within a digital surface a single party controls. It tells you that a user looked at a product, scrolled past it, came back, and eventually purchased. It does not tell you why. It does not tell you what alternatives the user considered outside the surface. It does not tell you what the user actually wanted, only what they did. It does not tell you what would have happened under different conditions.

These are real limitations. The clickstream is a behavioral artifact, not an intentional one. It is downstream of the user's reasoning, not a record of the reasoning itself. The companies that have built moats on clickstream data have done so by aggregating enough behavioral signal to infer intention statistically. This works at scale and produces useful predictions, but it is fundamentally indirect.

A second limitation: clickstreams are bound to the surface that captures them. Google sees what happened on Google. Amazon sees what happened on Amazon. Neither sees what happened in the user's life off-platform. The clickstream is rich within the surface and silent outside it.

What a negotiation trace captures

A negotiation trace records something categorically different. When two agents transact on behalf of their principals, the trace contains the structured record of what each side proposed, what each counter-offered, what was agreed, on what terms, with what supporting context, and what the outcome was. It is a record of stated intent (what each agent was trying to accomplish), counterfactual reasoning (what each agent considered and rejected), agreed terms (what was actually committed), and verified outcome (what subsequently happened).

Consider a concrete example. A consumer's agent contacts a telecom's brand agent to negotiate a bill that has increased due to an expired promotion. The trace contains: the consumer agent's stated objective ("preserve the prior monthly rate"), the brand agent's initial counter-offer ("we cannot extend the promotion but can offer a different bundle"), the consumer agent's evaluation of the counter-offer against the consumer's preferences ("the bundle is acceptable but only if it includes the international calling feature the consumer uses"), the brand agent's response, the eventual agreement (a specific monthly rate, a specific bundle, a specific term), and a signed receipt confirming the resolution.

This is a richer data object than a clickstream by an order of magnitude. It captures the actual reasoning of both sides, the actual outcome, and the alignment between intent and outcome. It is also cross-platform in a way clickstreams are not: the trace lives on the marketplace layer where the negotiation occurred, not inside either side's surface.

Why this is defensible

Three structural properties make negotiation traces a more defensible data asset than clickstreams.

They are not capturable from outside the marketplace. A clickstream is, at least in principle, observable by anyone who can instrument the surface. Browser extensions, screen recordings, third-party analytics packages, and (legally) regulator-mandated data portability can extract clickstreams from surfaces other than the surface owner. Negotiation traces live inside the marketplace where the agent-to-agent transaction actually occurs. They are accessible only to participants in that marketplace. This is a structural barrier to extraction that clickstreams do not have.

They are signed and verifiable. The receipt artifact at the end of every negotiation is cryptographically signed by both agents and is verifiable by any party that holds the public keys. This means that a negotiation trace is not just data, it is evidence. It can be presented in regulatory inquiries, used in dispute resolution, and treated as a contractual artifact in subsequent business relationships. This evidentiary status raises the value of the data above what raw behavioral logs can claim.

They compose into a corpus that improves itself. The trace from one negotiation becomes part of the training data for the next round of agents on both sides. Better consumer agents produce richer negotiation traces; richer negotiation traces train better consumer agents. The same dynamic operates on the brand side. The corpus has a self-improvement property that produces compounding returns to scale: a marketplace with twice the volume produces more than twice the data quality, because the additional volume includes both the new transactions and the improved agents trained on the prior transactions.

These three properties are not present in any clickstream-based data asset. They produce a categorically different defensibility profile.

What kinds of companies accumulate this corpus

Three structural positions allow a company to accumulate the negotiation-trace corpus at scale.

Marketplace operators. The neutral layer where consumer agents and brand agents transact sees every negotiation that runs through it. This is the strongest position structurally: the marketplace operator has the full corpus, can use it to improve its own systems, and can offer access to participants on terms it sets. The marketplace position is the one Teleperson is building toward.

Identity providers. A KYA provider that issues identities to agents on both sides of the marketplace will see signed receipts that reference its identity claims. This produces a partial view of the corpus, weighted toward the agents that use the provider's identity infrastructure. Identity providers are likely to be a small number of incumbents at scale, and the corpus they accumulate will be valuable but narrower than the marketplace operator's.

Trust-and-settlement networks. Card networks, payment-network operators, and the institutions that handle the financial side of agent transactions will see the settlement leg of every transaction that involves money. This produces another partial view, weighted toward financial transactions specifically. The corpus the payment networks accumulate will be commercially valuable for fraud detection, dispute resolution, and risk pricing, but will not include the negotiation reasoning that occurred before settlement.

The most strategic position is the marketplace operator. The other two positions are valuable but secondary, because they see the artifact of the transaction without seeing the reasoning that produced it.

Implications for category formation

The structural defensibility of the negotiation-trace corpus has direct implications for how the agentic-commerce category will form.

Winner-take-most dynamics in the marketplace layer. The compounding self-improvement of the corpus means that the marketplace operator that achieves liquidity first will pull away from competitors quickly. The first entrant with meaningful network density will train better agents, attract more participants, and compound the gap. Investors should expect the marketplace layer to consolidate to a small number of winners, likely one global and a handful of regional, within five to seven years of the first material network density.

Brand-side data leverage shifts. Brands that have accumulated rich CRM data over decades have built that asset on the assumption that they would be the only party with that data. In the agent-to-agent topology, the consumer's agent has access to the consumer's full vendor portfolio, including data the brand cannot see. The brand's data advantage diminishes. The marketplace operator has both sides of every conversation. The brand's negotiating position is weaker than it was, and brands that do not adapt their pricing and retention strategies to this shift will lose customer-lifetime value to competitors who do.

Regulatory attention will follow the corpus. A marketplace operator with the negotiation-trace corpus is going to attract regulatory scrutiny on antitrust and data-protection grounds. The right posture is to design the corpus from the outset for inspection: auditable governance, transparent data-use policies, regulator-accessible compliance tooling. Marketplace operators that build for inspectability before regulators ask will preserve their operating freedom; the ones that don't will face restrictions retroactively.

Closing

The negotiation-trace corpus is going to be the most consequential consumer-data asset of the next decade. It is structurally more defensible than clickstreams, accumulates with compounding returns to scale, and concentrates in the marketplace operators that achieve early liquidity. The companies that build the marketplace layer first, with the right corpus governance, will define the data economy of agentic commerce.

This is not a prediction about which specific company wins. It is a prediction about which position in the stack accumulates the most defensible value. The position is the marketplace layer, and the asset is the corpus. The companies optimizing for either are playing the right game.