01 · the challenge
The challenge
A small founder-led GTM team was getting 50–80 inbound enquiries a week through three forms and a shared inbox. Replies often took hours. High-intent prospects sometimes waited two days. The team wanted speed without losing the human voice that converted.
02 · our approach
Our approach
- 01
Mapped intents — pricing, demo, integration question, partnership, recruiter — and what a great human reply looked like for each.
- 02
Wired n8n to read the form webhook, classify intent with a small OpenAI call, and look up enrichment via Clearbit + the team's HubSpot.
- 03
Built a draft generator grounded in the company's voice doc, past wins, pricing, and calendar links. Each draft posted to Slack with one-click approve.
- 04
Added a confidence threshold — high-confidence replies auto-send; ambiguous ones wait for the human.
03 · what we delivered
What we delivered
- Production n8n workflow, version-controlled in git
- OpenAI eval suite — 60 golden examples, run on every prompt change
- Slack approval inbox + history
- HubSpot CRM enrichment + opportunity creation
- Looker Studio dashboard tracking response time and booking rate
- Runbook + Loom walkthrough for the GTM team
04 · reflection
“We didn't replace the founder's voice. We taught it to scale.”
The hardest part wasn't the agent. It was getting the voice doc tight enough that drafts felt like the founder, not a chatbot. Six iterations and a Friday afternoon of pair-writing later — the team trusts the auto-send.