Automating Lead Qualification with AI: Filtering and Routing Inquiries in Real Time

Sixty form submissions, nineteen real enquiries

A landscaping company got 60 enquiries in a month. After the owner went through them by hand, 19 were people who wanted work done. The rest were price shoppers who never replied, three people asking for services they don't offer, two job applications, and a scattering of spam.

He was spending about two hours a week on the 41, in the same inbox as the 19.

That's the problem worth solving. But the solution isn't as clever as the vendor deck suggests, and getting it wrong costs more than the admin time it saves.

Where the real loss is

Before the scoring, the response time.

The well-known lead response research in sales is consistent on one point: the odds of qualifying a lead drop sharply after the first few minutes, and attempts made within an hour are far more effective than attempts made the next day. Local service buyers behave the same way. Someone requesting a quote for a leaking roof is calling three companies and booking the first one that answers.

So the highest-value automation in this entire category has nothing to do with scoring. It's this: when a form is submitted, send an immediate message that confirms receipt, sets an expectation, and gives the person a way to reach you now.

If you automate only that, you'll beat most of your competitors.

What the scoring is genuinely for

Once speed is handled, scoring helps with a narrower job: deciding what to do next, and in what order.

For a local service business, a workable model uses four signals, and none of them require buying enrichment data.

Specificity of the request. "I need a quote to repaint a 3-bed semi in Oakville before we move in on the 20th" is a different lead from "do you do painting". The first has described a job; the second is browsing.

Timeline language. "Next month", "before the 20th", "as soon as possible" versus "sometime in the spring" or "just looking at options". This is the single most predictive signal in most inboxes.

Local fit. Is the address in your service area, and does the described job match what you actually do at a price you can deliver?

Channel and detail. A form with a phone number and a message beats a one-line enquiry with no contact details. A phone call beats both.

Note what's missing: budget and authority. For residential services, budget is usually unknown until you talk to someone, and guessing it from a form produces confident, wrong decisions. Treat those as things to establish on the call, not to score.

The mistake that costs money

The tempting design is: score low, deprioritise, discard.

For local services that's expensive, because of a specific behaviour: people research for weeks and then buy in a hurry. Someone who fills in a form in March asking general questions may be the customer who books in May, and the business that answered them properly in March is the one they call.

So the output of scoring should be a routing decision, not a rejection.

• High intent, in area, defined job → call within the hour, and put it at the top of the day.
• Unclear intent, in area → an immediate acknowledgement plus a short, useful second email that answers the obvious question (price range, lead time), then a single follow-up a few days later.
• Out of area or out of scope → a polite reply with a referral if you have one. Cheap goodwill, and it takes thirty seconds.
• Genuine spam → discard. This is the only category where deleting is correct.

Nothing gets thrown away except spam.

Where AI actually helps, and where it doesn't

Good uses:

• Reading messy messages. Classifying "need someone to look at me bathroom, water coming through ceiling, can you come today" as an emergency job in area is exactly what a language model does well, and it's tedious for a human.
• Drafting the acknowledgement. Different wording for a quote request, an emergency, and a question, in your voice.
• Extracting structured detail. Pulling suburb, service type, and timing out of a paragraph into a form field so whoever calls knows what they're calling about.
• Summarising the week. "Eleven enquiries needed a call back, two are still unanswered, one is a repeat customer."

Poor uses:

• Enrichment data. For residential local services there's often nothing to enrich. Someone submitting a personal Gmail address has no firmographic record, and the tools that claim to supply one are frequently wrong.
• Judging affordability. Not reliably possible from text, and the failure mode is discriminatory in ways you'd rather avoid.
• Writing to the customer without a review. An auto-reply promising a callout time you can't keep is worse than no reply at all.
• Replacing the call. No model closes a job. It gets a person onto the phone sooner.

A realistic build, in order

Step 1. Instant acknowledgement. Automatic, within a minute, for every enquiry. Confirms receipt, states when you'll respond, gives the phone number. One template per enquiry type, or one well-written generic.

Step 2. Notification routing. Emergency language, or a job description matching your highest-value work, sends a notification to a phone, not to an inbox nobody checks at 6pm.

Step 3. Structured capture. Whatever the form is, capture name, phone, address, service, and timing, reliably. Four fields, all on the form, not parsed from a paragraph. This step alone often doubles the usefulness of everything downstream.

Step 4. Classification and tagging. Assign the enquiry type and the intent level, so the inbox can be sorted by what needs a call today.

Step 5. Follow-up sequence for the unclear ones. Two gentle touches, then stop. Most businesses stop at zero.

If you're getting fewer than thirty enquiries a week

Read this bit before buying anything.

Below about thirty enquiries a week, the scoring layer is overkill. The manual version, one person clearing the inbox twice a day and calling the defined jobs first, beats a half-configured AI system, and it will stay reliable.

Do steps 1 and 2 anyway. The instant acknowledgement and the phone notification are worth more than everything else combined, they cost almost nothing, and they don't break.

Measure it

Three numbers, monthly:

• Median time to first response. Aim for under 15 minutes during working hours.
• Percentage of enquiries answered within 24 hours. The honest measure of whether the system works.
• Enquiries to booked jobs. If this doesn't improve, the routing is wrong, or the follow-up is where people are lost.

A scoring system that can't be tied to one of those three numbers is a hobby.

Where to start

Time your own response to the next enquiry. If it's more than an hour, fix that before you consider scoring, and check whether your form is capturing a phone number.

A free digital presence audit on Webcheckhub scores conversion readiness alongside Google presence and website health, which shows whether you're losing people at the enquiry stage or after it. When you want the intake and follow-up built and wired into your phone, Solutions Elites implements enquiry automation for small businesses.