01The bottleneck is almost never lead volume
When a service business says it needs more leads, the enquiries are frequently already arriving. They arrive in an Instagram inbox, late, and they ask the same three questions every time: do you cover my area, what does it cost, when can you come.
The business loses those leads to latency, not to lack of demand. The person asking at 11pm is comparing three options and will book whoever replies first.
02What the agent should actually be scoped to do
The instinct is to build something that closes the sale. That is the wrong scope and it is how these projects end up untrustworthy. Scope it to the predictable band: answer the questions that have a fixed answer, collect the fields you need, write a structured record, and hand anything unusual to a human immediately.
- Answer the three or four questions that always come up
- Capture vehicle / location / preferred time as structured fields
- Write the record to a database, not to conversation history
- Escalate anything outside the script instead of improvising
03Build it somewhere you can see it
For Absolute Mobile Car Detailing we built this in n8n - an Instagram DM trigger into an AI Agent node running Gemini, with Supabase as the store. The reason for n8n over a closed chatbot product is visibility: every step is on a canvas the client can open, read and change, and the lead data sits in their database rather than in a vendors.
04The measurable outcome is the clock
Judge it on response time and on capture completeness, not on how human it sounds. The enquiry gets answered in seconds instead of the next morning, and it arrives as a row you can act on rather than a thread somebody has to re-read.