Built lead intake bots in n8n, no clients yet. How did you sell your first one?
retained text
For the last month I've been building lead intake bots in n8n. I have two demos, one for real estate agencies and one for equipment rental like excavators and cranes. They run on my own server and work in Telegram, in a chat widget on a website and through a web form. The bot reads what the person wrote and asks only for what's missing, one or two things at a time. It understands voice messages, corrections, several machines in one request and "let's start over". If someone asks about price, it doesn't make up a number. It sends the question to a manager and keeps collecting the request. Angry clients, requests for a real person and complex topics go to a human with a reason. It remembers returning clients but doesn't reuse their old details. The manager gets a card in a Telegram group with all the details, the original message and any photos. The client gets a short PDF. Every lead is saved to Google Sheets along with a log of what the model decided, and the owner gets a weekly report. I spent a lot of time on reliability. Code checks every model answer, because the model breaks its own rules on unusual messages. If the model or the PDF step fails, the manager still gets the raw message and the contact. A watchdog runs every 10 minutes, finds stuck leads and reminds clients who went quiet. There are error alerts, uptime monitoring and daily backups. For testing I wrote 160 conversations with expected answers and a script that replays them against the live workflow without touching production. A bit over half of the steps pass right now. Most failures are the bot asking one more question when the request is already complete. I have no paying clients yet, and that's the part I understand least. So I'd love to hear from people who've done this: 1. What made your first small business client say yes? Did you charge a setup fee plus monthly, or something else? What did they actually care about? 2. How do you get the client's team to actually use it instead of ignoring the cards? 3. What goal did you agree on for a pilot so the client could see it worked? 4. Does your model decide the whole flow, or does it only pull out the data while code decides what happens next? I'm thinking about switching to the second.
document fetched 2026-09-27 00:18 · canonical https://reddit.com/r/AI_Agents/comments/1wqw94y/built_lead_intake_bots_in_n8n_no_clients_yet_how