Flex · · 12 min read

Flex: I Built a Quoting Bot in 27 Minutes

Not pretty. Not production. But a working quote-intake flow for a roofer, built in under half an hour. Here's the raw recording and what I'd fix next.

Source material Live demo
AI in Crayon · ep
27 minutes. One quoting bot.

Friday flex. I recorded myself building an AI quote-intake bot for a local roofer, start to finish, no edits.

What I built

  • Customer fills out a short form (address, roof age, concerns)
  • Bot asks follow-up questions tailored to their answers
  • Bot generates a preliminary quote range based on their roof type and zip code
  • Everything gets emailed to the roofer’s sales team

What worked

Claude (via the API) handled the conversation beautifully. The prompt was maybe 400 words. Total AI cost per lead: under $0.05.

What didn’t

  • The initial form had too many fields. Users bounced at step 3.
  • The quote range was too wide on the first run — had to give the model more specific pricing rules.
  • No rate limiting. A bot could’ve spammed it.

What I’d fix next

  • Trim the form to 3 fields max before starting the AI conversation
  • Feed the model the roofer’s actual past-project pricing as context
  • Add basic abuse protection (Cloudflare Turnstile, rate limits)

The Monday test

You don’t need 27 minutes. You need 10. Open the AI tool you already pay for, describe one specific workflow your team does manually, and ask it to help you design an intake form for that workflow. Don’t build it. Just draft it. You’ll learn more in 10 minutes of drafting than in three months of reading about AI.

Keep going

AI in Crayon · ep
The four rungs
Up next
Translated Strategy

From 'I Use ChatGPT Sometimes' to AI-First Operations

The practical ladder from curiosity to leverage. What changes at each rung, what to spend on, and where most operators get stuck (hint: it's step two).