An AI agent for lead qualification

The number that decides this

Pull your last twenty enquiries and look at the gap between when each arrived and when someone replied. In most companies of this size the median is measured in hours and the tail runs into days — enquiries that arrived Friday afternoon and got an answer Monday, if at all.

That gap is what an agent closes. Not the quality of the eventual conversation, which is your team’s job and should stay that way. Just the interval, during which a prospect is contacting whoever else came up in their search.

What the agent does

• Replies in minutes, by whichever channel the enquiry arrived — form, email or phone.
• Asks the qualifying questions your team already asks: what the problem is, what scale, what timeframe, who else is involved in the decision.
• Answers what it can answer factually — what you do, what you do not do, published prices, how the process runs.
• Writes the record into the CRM with the fields filled, so the first human contact starts from context rather than from scratch.
• Books the conversation where that is the sensible next step.

What it must not do

Decide whether the lead is worth pursuing. Strategic fit is a judgment about your business, not a property of the enquiry. An agent filtering on its own criteria will discard the unusual enquiry that turns into your best customer, and nobody will ever know it happened.

Invent a price. Where price is published and unconditional, the agent says it. Where it depends on scope, it explains what it depends on and books the call. An improvised number becomes an expectation someone has to walk back, which is a worse start than saying “it depends, here is on what”.

Score leads by its own logic. Collect the facts, apply whatever rules your team already trusts. A score the agent invented gets overridden by everyone within a fortnight, which leaves you with a field nobody reads.

Where it gets awkward

The prospect who wants a human immediately. Some do, and pushing them through a qualification script costs the lead. The rule should be that any request for a person routes to one, and the agent keeps whatever it has gathered so far rather than discarding it.

Enquiries that are actually support. Existing customers frequently arrive through the contact form. The agent needs to recognise a known email address and route accordingly, or a customer with a problem ends up being sold to.

What it connects to

The CRM is the one that matters, with write access — this use case is substantially about record quality. Forms and the shared inbox for intake, and the calendar where qualification ends in a booked conversation.

How to tell whether it worked

• Median and worst-case time to first response.
• Share of enquiries that received any response at all.
• Share of records with the qualifying fields actually filled.
• Number of enquiries arriving outside office hours that got an answer anyway.

The third one compounds quietly. A lead handed over with empty fields costs the salesperson a call to establish what the enquiry already said.