Wildcard

Wildcard — A trace-only AI referral revenue attribution layer over connected ecommerce merchants, with an unusually complete causal join: for each confirmed AI-referred session the record links the specific tracked prompt that drove the visit, the citation/recommendation that earned the click, the landing surface, and the matched order and revenue. Crucially the dataset also stores its own negatives — Wildcard explicitly records what could NOT be observed (missing referrers, cross-device journeys, unobservable views) and discloses them alongside every result, and it deliberately excludes Shop…

Wildcard offers (Alternative, Fundamental) — Request further research — no session, order or prompt volume is published and the dashboard figures are marked illustrative. Given $130K raised and 5 employees the panel is early, but the per-record grain is the point: each row carries prompt, model, surface, citation, product, order and revenue, so even a modest session count yields a high-value joined corpus..

Prompt-to-purchase ground truth — end-to-end labelled examples of (buyer prompt → model recommendation/citation → click → product → paid order), which is the reward signal shopping-agent and retrieval-ranking models most lack and which no analytics-only competitor can produce because they see arrival, not the prompt behind it. The disclosed-unobservable accounting is separately valuable as a calibration dataset for attribution-bias correction across the industry (how much of AI influence is structurally invisible to last-click, measured rather than assumed). Also: per-model/surface revenue per session for channel-value models, and product-level AI-demand signals for merchandising.

The data is with 2 years of history.

Coverage spans US; Application Software; alternative, fundamental; equities.