Live demo: https://agent-commerce-analytics-template.templates.workers.dev/

A Cloudflare Workers template that gives merchants visibility into how AI shopping agents interact with their store. It tracks the full journey from discovery (/llms.txt reads) through browsing (product views) to purchase (checkout attempts), broken down by agent identity, payment network, and verification status, and ships with a React dashboard that surfaces auto-generated insights (top agents, blocked checkouts, product gaps, content quality, revenue headlines). Seeded with bundled demo traffic so the dashboard is populated on first deploy.

Today, Cloudflare Radar tells a merchant: "You got AI bot traffic." Zone Analytics tells them: "These paths were hit."

Nobody tells them: "An agent searched for toddler skis, found the Little Rippers, and bought three items for $199.97."

The analytics Worker collects commerce-specific signals — /llms.txt reads, search queries, verified agent identity, checkout attempts — then generates a merchant dashboard with:

• Auto-generated insights — top agents, blocked checkouts, product gaps, content quality, revenue headlines — all computed from data, not hardcoded
• Agent journey timelines — per-agent session reconstruction showing discover -> browse -> purchase -> return, with timestamps and return visit detection
• Revenue with dollar amounts — estimated revenue per agent, per payment network, average order value, verified vs. unverified split
• Visual discovery funnel — ASCII bar chart showing /llms.txt reads -> product views -> checkout attempts -> successful/blocked split
• Security & trust — verified/unverified agent ratio, blocked checkout rate, flagged unverified agents
• Demand signals — search query -> outcome mapping: which queries converted, which bounced, which had no matching product (product gap detection)
• Content quality signals — is the concise /llms.txt sufficient, or do agents need /llms-full.txt?
• Product performance — per-product conversion rates, estimated revenue, "viewed but didn't purchase" attribution
• "What this unlocks" comparison — side-by-side of current Cloudflare analytics vs. what this dashboard surfaces

Discovery signals           Trust signals
(/llms.txt reads,           (agent identity,
 search queries)             checkout intent)
        \                      /
         v                    v
    Agent Commerce Analytics Worker
    Collects events, computes aggregations,
    auto-generates insights from data
              |
              v
    Merchant Dashboard
    "Here's what agents did in your store"

Deploy straight from the button above, or run locally:

npm install
npm run dev        # Start the Worker + React dashboard locally

Other commands:

npm run build      # Type-check and build the React app
npm run deploy     # Build and deploy to Cloudflare Workers
npm test           # Run the Vitest suite

On first request the Worker seeds bundled demo traffic (5 agents, 19 interactions) so the dashboard is populated out of the box. Set the SEED_DEMO_DATA binding to false once you're ingesting real events.

The bundled demo traffic includes 5 agents with 19 interactions:

Every event is classified along four independent signals so the dashboard can show a real trust gradient rather than a verified / unverified binary:

Those roll up into one of three tiers:

• verified — Web Bot Auth was valid OR a Managed Ruleset KYA token was present (Cloudflare-verified at the edge).
• claimed-only — Matched the signed-agents list OR a Signature-Input header was present but unverified. The agent identifies itself but provides no cryptographic proof we accepted.
• unverified — UA-only. Long-tail traffic claiming an agent identity with no proof.

GET /dashboard/trust returns the rollup. The POST /events/from-headers endpoint derives signals automatically from the request, so deploying behind a Worker that already runs Web Bot Auth or a KYA-emitting ruleset just works.

TODAY: 5 agents made 19 requests. 2 purchased ($329.96 est. revenue).
       1 unverified checkout blocked. 1 search for products you don't carry.
       Top agent: BuyBot Shopping Agent (Visa) — 3 items, $199.97

INSIGHTS
  [$$] Revenue: Est. $329.96 from agent commerce
  [>>] Top agent: BuyBot Shopping Agent — 3 checkouts, est. $199.97
  [<>] Repeat visitor: BuyBot returned searching "toddler ski goggles"
  [!!] Blocked: PriceScout checkout denied (unverified)
  [??] Product gap: "snowboard toddler" — no matching product
  [##] Top product: Little Ripper Ski Package — 33% conversion
  [OK] Content working: 80% of agents needed only concise /llms.txt

This template is the visibility layer of a broader exploration of agent commerce on Cloudflare — the loop where agents discover a merchant's catalog (e.g. via /llms.txt), get verified at the edge, and convert intent into checkout. Those other layers are out of scope for this template; this one stands on its own and can ingest events from any source that POSTs to /events or forwards cf-agent-* headers.

A merchant running no other piece of that loop can still benefit from a commerce-specific analytics surface. It's stronger when composed with the rest: discovery events (/llms.txt reads, search queries), verified-agent signals from an edge ruleset, and demand patterns from an agent orchestrator all flow through the same event pipeline.