Revenue attribution

Revenue attribution answers the question every price objection hides behind: what did the AI employee actually contribute? Instead of "you had 93 conversations", the platform records every business outcome — a WooCommerce order, a Calendly booking, a lead marked won — and links orders back to the conversations that drove them, in two honest tiers that are never summed together.

The two tiers

Tier	Requires	Window
Direct	The assistant recommended a product, the visitor clicked its card, and an order containing that product followed. Every direct euro has a visible chain of evidence: conversation → click event → order line.	7 days (configurable)
Influenced	The same visitor had a conversation before ordering, but the specific product can't be tied to specific advice. Reported as context, never as proof.	30 days (configurable)

An order that qualifies for both tiers counts once — as direct. When several conversations precede an order, the last touch wins for influenced; for direct, the click-bearing conversation wins even when a later click-less one exists (evidence beats recency). No identity match means the order stays unmatched — there is no fuzzy guessing, by design.

How an order is linked to a conversation

Identity keys, strongest first:

1. Conversation cookie — at checkout the WordPress plugin stamps the widget's own first-party cookies (pb_anon_id, pitchbar_conv_id) onto the order as meta. Same browser, deterministic.
2. Visitor id — the widget's 1-year anonymous id, matched through the visitor record.
3. Logged-in customer — the plugin's ShopperToken already carries wp_user_id onto the conversation; the order's customer id matches across devices.
4. Hashed e-mail — the order's billing e-mail as a sha256 hash, compared against captured leads. The raw address never leaves the shop and is never stored on the outcome row.

Honesty rules, built in

• Direct and influenced totals are never added into one figure.
• Refunds and cancellations reduce or reverse attributed value automatically; the original value stays on the row for audit.
• A recorded match is never rewritten by a webhook retry, a later conversation, or a window change — windows apply to new orders only.
• Unmatched is a correct answer, not a failure.

Requirements

• WooCommerce shops need the WordPress plugin v2.1.0+, which adds the order hooks and the checkout identity stamping.
• Calendly bookings and leads marked won in the inbox create conversation-linked outcomes automatically — no extra setup.

Reading the numbers (Phase A)

Phase A collects silently — there is no dashboard yet. The audit surface is an artisan command, run per agent:

php artisan attribution:report --agent=<uuid> --days=30

It prints per-tier totals, the count of unmatched and reversed orders, booking / won-lead counts, and one evidence line per direct match so each attributed order can be verified by hand against the real shop before any client sees a number.

Configuration

Env	Default	Meaning
ATTRIBUTION_DIRECT_WINDOW_DAYS	7	Max days between the product click and the order.
ATTRIBUTION_INFLUENCED_WINDOW_DAYS	30	Max days between the conversation and the order.