The layer underneath that compounds.
The whole platform runs on one intent graph: which moments convert, which get abandoned, what each is worth — per agent environment, per lifecycle stage. It's why performance improves with every vendor and every moment, and why a late entrant can't bootstrap it.
The moment-conversion dataset.
Which moments convert. Which get abandoned. What each is worth, which framing works — per agent environment, per lifecycle stage, free through paid. This is pre-checkout intent Stripe never sees and no analytics vendor can reconstruct, because it exists only at the moment of the blocked tool call.
- Your single-tenant analytics are yours — in your dashboard, nobody else’s.
- Cross-tenant learning is privacy-gated aggregates only. Never your raw data.
- The dataset compounds with every vendor and every moment. It cannot be bootstrapped late.
Attribution with nothing to argue about.
An authenticated MCP session maps to a pseudonymous per-user subject, which maps to a Stripe customer, which maps to a completed checkout. Every link in that chain is deterministic. No cookies, no probabilistic modeling, no attribution debate — and for a brand-new trial or free converter, checkout mints the Stripe customer inline. Still deterministic, still cookie-free.
- subject_ref is vendor-hashed and PII-forbidden by contract.
- The checkout outcome is stitched back through Stripe metadata — not inferred.
- Determinism is what makes outcome-based pricing credible at all.
Coarse intents, fine facts, and a decision engine in between.
Cross-tenant learning only aggregates if the vocabulary is shared. Canonical buying intents and an eight-class commercial-trigger taxonomy stay deliberately coarse — that's what makes patterns comparable across vendors. Fine-grained variety lives in namespaced facts. Above both, a Thompson-sampling engine learns which offer framing works — always inside guardrails you set.
- Intents coarse, facts fine — the discipline that keeps network learning real.
- The engine optimizes framing within your messaging constraints, never around them.
- Every experiment is grounded in a recorded moment, like everything else here.
The products are the value. The graph is why it's durable.
Every moment you capture makes the next one convert better. Start accumulating.