A batch of inbound leads used to take us two to three hours to work through. It now takes five minutes.
This is our own pipeline, not a client’s. Leads from our contact forms were read one at a time: does the company fit, can this person buy, and what should the first message say? Replies went out late, and the time came out of work that needed a person.
| Before | After | |
|---|---|---|
| A batch of leads | 2–3 hours | 5 minutes |
| Qualification work | 15–20 hours a week | Automated |
| First message per qualified lead | About 30 minutes | Drafted automatically |
In total, that’s more than 25 hours a week back.
The rules
Two checks decide whether a lead qualifies: the company has fewer than 60 employees, and the contact has the authority to buy. GPT-4 applies them and writes down its reasoning for every lead, so when a decision looks wrong we can see why and adjust the rule.
The message
For qualified leads, GPT-4 drafts a first message that mentions the prospect’s industry, the process they told us they want automated, and how they prefer to be contacted, written in our voice.
Running it
Someone on the team sends a command to a Telegram bot. n8n works through the batch in sequence and posts progress back to the chat, so there’s no dashboard to open.
Stack used
- n8n
- OpenAI GPT-4
- Google Sheets
- Telegram
Related: custom ai solutions, n8n consulting and ai for small business