Revenue Attribution Methodology

The headline figure on your dashboard is AI-influenced revenue, not a guaranteed or net-new amount. It is the total value of orders where a product the shopper saw recommended by the AI agent was purchased within 30 minutes of that recommendation. This is a last-touch model: the most recent qualifying AI interaction before the order receives credit for the full order value.

When an order arrives by webhook, we match its line-item products against recommendation events for the same store and shopper identity within a 30-minute window ending at the order time. If at least one ordered product was recommended in that window, the order is tagged ai_last_touch and its full total is recorded as attributed revenue. If there is no match, the order is tagged none and contributes no attributed revenue. The dashboard compares AI-influenced revenue for its stated date range with the listed monthly plan price. This is a revenue-to-plan-price ratio, not profit-based ROI or net gain. A ratio is shown only when the revenue and reference price share the same currency.

Last-touch attribution is simple, transparent, and stable, but it over-credits the agent: some of those shoppers would have purchased anyway, and crediting the entire order (rather than only the recommended line) is generous. Treat the revenue-to-plan-price ratio as a directional indicator of value, not a measured promise of incremental profit. We deliberately do not inflate it with multi-touch weighting that would be impossible to audit.

The optional holdout is off by default. When enabled, a deterministic share of visitors does not see the chat launcher, conversation, teaser or proactive chat. Exposed visitors can see the assistant. Standalone promo bars, exit-intent popups, recommendation carousels, product Q&A and scarcity badges are not withheld. The report compares linked-order conversion rates between the arms after at least 100 visitors per arm and at least one linked order. This threshold is a display safeguard, not a statistical-significance test. Cart-token linking currently supports Shopify; WooCommerce-only stores do not get a lift figure. Consent, missing tokens, sample balance and other interventions can bias the estimate. Treat it as directional evidence, not guaranteed causal revenue or profit.

An order following a recovery email is an associated outcome, not proof the email caused it. Isolating incremental email lift requires a suitable randomized no-send control among comparable eligible shoppers, with the same purchase suppression and observation window. The chat-only holdout does not isolate recovery-email effects. This page does not claim that a built-in email no-send experiment is available.

We do not count orders with no qualifying recommendation in the window, orders outside the selected date range, or test/cancelled orders excluded by your platform. The number is not a billing input - your plan price is flat and independent of attributed revenue. Attribution is for your reporting and decision-making only.

Every attributed order stores its attribution source and the conversation that influenced it, so the figure is auditable against your store's order log. If you want the underlying rows, export them from the Orders and Reports surfaces, or request the raw attribution export through support. We will always show you how a number was produced.