Case study · Internal · AI lead-routing

A six-question quiz that routes itself.

The funnel that meets every visitor to ai1off.com — two AI agents, six questions, one personalized recommendation, delivered before the page even reloads.

Built for: AI1offs (our own funnel)
Role: Qualify visitors, route to the right service
Shape: Cloudflare Worker → n8n → two AI agents → CRM
Stack: n8n · Groq · LangChain · Gmail · Sheets · Supabase CRM

The before

Three services, one inbox, a guessing game.

AI1offs sits in front of three offerings — custom apps and AI builds here, document-workflow automation at DocConduit, and WordPress hosting at WyoHost. Every visitor lands with a different problem, and the path that fits them isn't always the page they clicked.

The old answer was a generic contact form, a hopeful inbox, and an evening spent reading between the lines to figure out whether someone needed a webapp, a workflow, or a host. The first half of every discovery call was a re-qualification we should have done up front.

What we needed was a router. Fast enough that visitors would actually fill it out, smart enough that the recommendation was right, and structured enough that whatever it produced could land in the CRM without a human in the loop.

The shape of it

Two phases. Two AI agents. One webhook.

A Cloudflare Worker gates every request through Turnstile, then forwards to n8n. A switch node picks the phase. Each agent has a structured output parser so the response shape is predictable.

Phase 01 · Reveal

"Here's the path that fits you."

The Reveal agent reads the six answers, considers the three service paths, and returns a structured recommendation: which spoke fits, why, and one or two outcomes the visitor could expect. The structured output parser keeps the response shape predictable so the frontend renders it without surprises.

Groq · LLM
LangChain · structured output parser
respond-to-webhook · under 1s

Phase 02 · Email + log

"Here's that recommendation in your inbox."

When the visitor opts into the emailed version, a second agent drafts a personalized recommendation, then the workflow fans out: Gmail sends to the visitor, Sheets logs the interaction, Doug gets a new-lead notification, and a Code node signs the payload before POSTing it to the CRM webhook.

Groq · LLM + parser
Gmail · personalized send
Sheets · analytics log
HMAC-signed POST · /api/webhook/n8n

Outcomes

A discovery call that starts at the right answer.

< 1s From quiz submit to on-screen recommendation.

15 Nodes: two agents, two parsers, one switch, four fan-out targets.

3 → 1 Service paths considered, single recommendation routed.

Cloudflare Worker Turnstile n8n Groq LangChain Gmail Google Sheets Supabase CRM (HMAC)

Try it on yourself

Built for us. Built for you next.

We use this quiz on every visitor to ai1off.com — and we ship the same shape for clients who need to qualify their own visitors before a sales call. If you've got more than one offer and one inbox, this is the pattern.

Take the quiz Book a 30-min build call

All case studies