AI lead qualification: a conversation with rules, not a form

An assistant can qualify a lead, but the rules are the company's: what it may ask, what it may infer, what it may promise, and when it hands over.

Inbound does not keep office hours. A message lands at eleven at night from someone who compared three providers that afternoon, and the company that answers first with something useful usually stays in the conversation. Handing the first reply to an AI assistant is a reasonable decision, and most companies make it for the speed. The mistake is treating speed as the whole job, then discovering the assistant has been qualifying everyone enthusiastically and nobody usefully.

Qualification is a conversation with rules, and the rules are the company's to write. The assistant asks the structured questions any good consultant would ask: what the buyer needs, by when, and for whom. It recognizes fit and urgency against definitions the company supplied, and it scores nothing silently: when it decides a lead is hot, warm, or early, the reason travels with the verdict. What it may ask, what it may infer, what it may promise, when it hands over: each of these is a written decision, not a default that arrived with the tool.

The failures have a shape. There is the assistant that interrogates like a form, four questions deep before the buyer has said anything real, and the buyer leaves at hello. There is the one that promises a timeline it was never authorized to promise, because moving the conversation along felt helpful in the moment. And there is the qualified lead that reaches sales wearing a score and carrying nothing: no notes, no context, no trace of what the buyer actually said, so the human restarts the conversation the assistant already had.

The practice starts with question sets short enough to earn their answers. Two or three questions, asked in a natural order, each one justified by the last reply, beat a questionnaire every time. Label the assistant honestly as AI, because a buyer who discovers the disguise later stops trusting everything else the company said. And write handover notes for the human who continues, in the buyer's own words where possible, so the conversation transfers instead of restarting.

Two further disciplines keep the system honest. Route the impatient to a person deliberately: a buyer who says so, or signals it in tone, is not a qualification failure, and making them argue with a form is how good leads die. And review mis-scores on a schedule: read the conversations where the assistant called a lead dead that later bought, or hot that went quiet, and correct the rules rather than blaming the model. The rules are the company's, which means the errors are too.

We run our own funnel through an assistant built this way, and the qualification rules were written by us, revised by us, and signed off by the humans who take the handovers. Guidance, not guarantees: response-time benchmarks, conversation-design choices, and what any given assistant may promise are matters for whoever operates your channel and, where consumer-protection rules bite, licensed counsel. Write the rules down, label the machine honestly, and hand over with the context intact.