Quick Answer
Lead scoring is a way to rank every enquiry before you reply to it, so the best job in your inbox gets a response in five minutes and the tyre kicker waits until Friday. You can build a working version in Make.com on the Core plan (about £8.50 a month) using a form trigger, an OpenAI module to read the enquiry, a small scoring rubric, and two routes: fast-track and nurture. For a one-van business handling 20 to 40 enquiries a week, this turns a chaotic inbox into a sensible pipeline without hiring an admin. Leads contacted within five minutes are 21 times more likely to qualify than leads left for 30 minutes (MIT and InsideSales, 2007), and UK trades currently lose roughly £24,000 a year to unanswered enquiries (DigitalX, 2026). Scoring solves both problems at once.
Table of Contents
- Why lead scoring matters for trades right now
- The stack: what you actually need
- The three-dimension scoring framework
- Step 1: Map your lead sources into one inbox
- Step 2: Define your scoring rubric
- Step 3: Build the Make.com scenario
- Step 4: Route hot leads, nurture cold ones
- Step 5: Track conversion and tune the score
- What this actually costs
- Common mistakes that kill a scoring system
- Watch
- What tradespeople are saying
- Frequently asked questions
- My verdict
Make.com
ServiceM8
ZapierWhy lead scoring matters for trades right now
Most one-van businesses I speak to treat the enquiry inbox the same way for every job. A boiler replacement gets the same five-line WhatsApp reply as a "can you quote me for a tap washer?" Both customers wait 20 minutes. The boiler customer rings someone else. The tap washer customer takes up an hour of your evening on a £40 job.
The numbers behind this are stark. MIT and InsideSales tracked more than 15,000 leads across 100 firms and found that responding within five minutes makes a lead 21 times more likely to qualify than responding within 30. Industry research published this year by DigitalX put the average UK trades loss from unanswered or slow-answered calls at around £24,000 a year. The size of that gap, between five minutes and 30, is where lead scoring earns its keep.
For the last 15 years lead scoring lived inside enterprise CRMs. HubSpot, Salesforce, Pipedrive, all charging £40 to £150 per user per month, all of them overkill for a heating engineer with two vans. That changed when Make.com made it possible to wire a form, an AI model and your job management software together for less than a takeaway. The barrier to entry collapsed. Most trades just have not realised yet.
Scoring is not about treating customers differently. It is about treating enquiries differently. A high-value job is allowed to interrupt your day. A "just curious" message can wait until Friday. The goal is to get to the right answer faster, with less friction, and stop the inbox running your week for you.
The stack: what you actually need
You need three things, and one of them you already own. The other two are cheap.
First, a single place all your enquiries arrive. Most trades have leaks here. Calls go to a mobile. Form submissions go to Gmail. Facebook messages go to Messenger. WhatsApp goes to WhatsApp. The website contact form goes to a form provider that emails you. Five inboxes, no single view. Step one is plumbing them all into one place. For most trades that one place is ServiceM8 or whichever job management tool you already use. If you do not have one, a single Gmail label and a Google Sheet works fine to start.
Second, Make.com. Make is a visual automation builder that connects apps together. For our purposes it does three jobs: it watches for new enquiries, it asks an AI model to read each one and score it, and it routes the result. The Core plan at roughly £8.50 a month gives you 10,000 operations, which for a one-van business is more than enough. If you are processing more than 200 enquiries a day you have bigger problems than scoring.
Third, an OpenAI account. The scoring brain in this build is a GPT-5 prompt that reads the enquiry text and outputs three numbers. You pay per request, and at the volumes we are talking about, each scored enquiry costs roughly £0.001 to £0.005. A busy month for a small trades business is under £1.
That is the whole stack. Make.com plus OpenAI plus whatever inbox you already use. Make beats Zapier for this job on cost alone once you cross 1,000 operations a month, and the AI module is genuinely more capable. Zapier is fine if you already pay for it. n8n is fine if you already self-host it. For a fresh build, Make is what I would pick.
The three-dimension scoring framework
Enterprise lead scoring models can have 30 or 40 signals. Page views. Email opens. Job title. Company size. None of that applies to a trades enquiry from a homeowner on a Tuesday morning. You need a simpler frame.
The one I use, and the one I would recommend to any small trades business, is three dimensions: Job size, Urgency, Source quality. Each scores from 0 to 30. Sum them, plus a 10-point qualitative bump from the AI for "this person sounds serious", and you have a score out of 100.
Job size is the most predictive signal you have. A boiler swap is worth 30. A bathroom refurb is worth 30. A leaking tap is worth 5. The AI reads the enquiry text and matches it against your job categories. If the enquiry mentions specific equipment, named brands, or a budget number, that is a buying signal and the score climbs. Vague messages with no detail score low because they almost always do.
Urgency is the second dimension. "No hot water" is a 30. "Quote for a new bathroom in the spring" is a 10. "Just thinking about" is a 5. Urgency is a tone signal as much as a content signal, and the AI handles it better than a keyword filter. The phrase "as soon as possible" can mean today or it can mean within the next month. Reading the surrounding context catches both.
Source quality is the third dimension and the one most trades get wrong. A referral from an existing customer scores 30 because referrals close at three to five times the rate of cold leads. A direct enquiry through your website scores 25 because the person searched, found you, read your work, and chose you. A Checkatrade lead scores lower because the customer is also speaking to three other trades, which means lower close rate and more price pressure. A cold Facebook message scores lowest of all.
| Dimension | 30 points | 15 points | 0 points |
|---|---|---|---|
| Job size | Boiler, bathroom, rewire, system install | Repair work, small install | Tap washer, advice request |
| Urgency | Emergency, no heat, no water | Within the month | Future planning, "thinking about" |
| Source quality | Customer referral, direct website | Google Local Services, Facebook page | Cold lead site, bargain hunters |
You will want to tune the categories for your trade. A roofer scores differently from an electrician. A landlord-focused gas engineer scores landlords highly under source quality regardless of urgency, because the long-term value of a managed property is worth four boiler repairs. Start with the three dimensions above, run it for two months, then adjust.
Step 1: Map your lead sources into one inbox
This is the part most trades skip and then wonder why scoring does not work. Before you build anything in Make, you need every enquiry landing in a place Make can watch. There are three reasonable routes.
The cleanest route is a single web form on your site that catches the contact-page submissions, with phone calls captured by a missed-call autoreply service that drops the caller's number and any recorded message into the same Sheet or CRM. Tradify, ServiceM8 and Jobber all support this. If you cannot put a form on every channel, use a virtual receptionist or an AI receptionist that asks two qualifying questions ("what is the work?" and "when do you need it?") and writes the answers to the same place.
The slightly messier route is to leave the channels where they are and use Make to watch each one. Make has native connectors for Gmail, Outlook, Facebook Pages, Google Sheets, Typeform, Tally, ServiceM8 and most form providers. You build one trigger per channel and route them all into the same scoring branch. More setup, more moving parts, more places for things to fail.
The third route is a phone-only business that uses a missed-call SMS responder. Every missed call triggers a text asking what the customer needs. The reply lands in a single SMS inbox you can watch from Make via a service like Twilio. It works but it is the most fragile of the three.
Step 2: Define your scoring rubric
The scoring brain is a GPT-5 prompt. Get this right and the whole system works. Get it vague and you will tune for months. The trick is to write the rubric the way you would explain it to a new apprentice, in plain English, with examples.
Here is the structure I use. Open with what the AI is doing ("you are scoring an inbound enquiry to a Sheffield-based heating engineer"). Then give the three dimensions with point bands and three real examples per band. Then list the output format you want, which should be a strict JSON object so Make can parse it without breaking.
Examples matter more than instructions. Give the prompt one example of a high-score enquiry and one of a low-score, with the score and the reasoning written out. The AI will pattern-match to those examples far more reliably than to abstract rules. Spend the time on three good examples per dimension and the model will be accurate eight times in ten.
Step 3: Build the Make.com scenario
The scenario itself is six modules. Sign in to Make.com, click Create New Scenario, and lay them out in this order.
- Trigger. Pick the connector that watches your enquiry inbox. For most trades this is Google Sheets "Watch Rows" pointing at the master enquiry sheet. For a form-driven setup, use the Webhook module so the form posts directly to Make.
- Text Parser. Pull the relevant fields out of the trigger payload. Customer name, message body, source, phone number, email. Set the source field manually for each channel so the AI knows where the lead came from.
- OpenAI: Create a Chat Completion. This is the scoring step. System prompt contains the rubric. User prompt contains the parsed enquiry text and source. Set the model to GPT-5 and the response format to JSON object. Temperature 0.2, so the model stays consistent across enquiries.
- JSON: Parse JSON. Take the OpenAI response and turn it into discrete fields. You now have total_score, job_size_score, urgency_score, source_score and reasoning available as variables.
- Router. Three routes. Route A for total_score 75 or higher, the fast-track. Route B for 40 to 74, the standard reply. Route C for under 40, the nurture path or polite decline.
- Action per route. Route A creates a job in ServiceM8 with a "HOT LEAD" tag and sends you a Telegram or WhatsApp alert with the score and reasoning. Route B writes to a callback Sheet that you work through twice a day. Route C sends an auto-reply with a calendar booking link or, for very low scores, a polite "we are not the right fit" message.
Total build time is about three hours if you have done Make before, six hours if you have not. Most of the time goes into the OpenAI prompt and testing it against real enquiries. Save a folder of 20 to 30 historical enquiries from your inbox and use them as test data. If the AI agrees with the score you would have given on 16 of 20, the prompt is good enough to ship.
Step 4: Route hot leads, nurture cold ones
The routing is where the system pays for itself. The hot route needs to interrupt you. The cold route needs to not.
For hot leads, I use Telegram alerts because they push through silent mode on my phone but a WhatsApp Business broadcast works just as well. The alert should include the customer's first name, the score, the one-sentence reasoning from the AI, and a clickable phone number. The whole message is about 40 characters. Time from enquiry submission to phone in my hand is usually 90 seconds.
For standard leads, you do not need an alert. You need a list. Make writes the enquiry to a Google Sheet or a dedicated ServiceM8 view, and you work the list twice a day at 11am and 4pm. This protects the morning and afternoon job windows from constant phone-checking. The standard list is also where the bulk of the work comes from over time, so the discipline of working it consistently matters more than the urgency.
For low-scoring leads, you have two choices. The polite decline is appropriate for enquiries that are clearly outside your service area, your trade, or your minimum job value. Write one template and have Make send it automatically. The nurture path is appropriate for low-urgency, low-job-size enquiries that might mature later. A "thanks for your enquiry, here is a guide to bathroom planning, we'll be in touch in March when most refurb work is booked" auto-reply is perfectly fine and keeps the door open.
Step 5: Track conversion and tune the score
Every system needs a feedback loop or it drifts. The good news is that for scoring, the loop is dead simple: log the score, log whether the lead converted, look at the data once a month.
Add a final Make module that writes every scored enquiry to a Google Sheet with the date, the score, the source, and a blank "converted" column. At the end of each month, go through the sheet and tick the conversion column. Then sort by score and check. If your "hot" band (75+) is converting at over 60 percent, the prompt is calibrated correctly. If it is below 40 percent, the rubric is too generous and needs tightening. If your "cold" band is converting at over 20 percent, you are throwing money away on the polite decline and the score thresholds need adjusting.
You will also notice patterns in the data that nothing else would show you. A specific source you assumed was strong might be converting at 15 percent. A weak-looking source might be converting at 50. Google Local Services Ads in particular surprise people both ways, depending on your trade and your area. The scoring sheet gives you the first honest data on which lead sources actually pay back.
What this actually costs
Honest numbers, in GBP, for a one-van or two-van business.
For comparison, an enterprise CRM with built-in scoring like HubSpot Marketing Hub Starter runs about £15 per user per month for two users, plus an extra £40 per month for the lead scoring tier, plus the connectors to your other tools. You are looking at £80 to £150 a month minimum, and the setup time is similar. The Make.com approach gets you 90 percent of the benefit at 20 percent of the cost, and you actually own the logic so you can change it without paying a consultant.
The hidden cost is your time learning Make. The first scenario takes a weekend. The second takes an hour. By the third you are reusing modules and adding new ones in 20 minutes. Treat the weekend as an investment, not a tax. Every other automation in your business gets cheaper after you have the muscle memory.
Common mistakes that kill a scoring system
I have watched plenty of small businesses build scoring systems and abandon them within six weeks. The reasons repeat.
Mistake one: scoring before centralising. If your enquiries still arrive in five places, the score is incomplete. You see what Make sees and miss what it does not. Fix the inbox first.
Mistake two: too many score bands. Three is the right number. Hot, standard, cold. As soon as you add "warm-hot" and "lukewarm" the routing logic falls apart and you stop trusting the system. Resist the urge.
Mistake three: trusting the AI without testing. Send 30 historical enquiries through the prompt before you go live, score each by hand, and compare. If the AI disagrees with you on more than four, the prompt needs more examples. Do not skip this step. The AI is helping you make commercial decisions and it needs to be calibrated to your actual judgement.
Mistake four: no feedback loop. Without logging conversion, you cannot tell whether the scoring is calibrated. Six months in you will assume it still works because it once worked. Add the logging module on day one.
Mistake five: treating the score as gospel. It is a triage tool, not a court judgement. If a "cold" lead rings the office and says they have just inherited £40,000 and want a heat pump, your apprentice does not need to check the score before booking the visit. Use the score to decide what to do automatically. Use your judgement to override when it matters.
Watch
How to build an AI Lead Scoring System in 19 MIN! | Make.com | ChatGPT | Tally
Step-by-step end-to-end build using Make, OpenAI and a Tally form trigger.
Automate Lead Qualification with AI | Make.com Tutorial for CRM & Slack
Practical walkthrough of qualifying leads with AI and routing them into a CRM.
How to Build An Automated Lead Management System in Make.com (Pipedrive)
End-to-end lead management scenario, easy to adapt for ServiceM8 instead of Pipedrive.
Automate Lead Research with AI & CRM in Make.com
Pulling extra context on a lead from the web before scoring. Optional but useful.
Make.com CRM Automations & How to Build a Scalable Sales System
Broader Make.com sales build, useful for the routing and notification patterns.
Setup AI Lead Scoring With ChatGPT + Zapier + Salesforce
The Zapier version of the same pattern, worth a watch if you already pay for Zapier.
What tradespeople are saying
Frequently asked questions
No. Make.com is visual. You drag modules onto a canvas and connect them. The only typing is the OpenAI prompt, and that is written in plain English. If you can write a careful email to a customer, you can write the prompt.
With a tight rubric and three good examples per band, GPT-5 will agree with your scoring eight times in ten. The two it gets wrong are usually the borderline cases between hot and standard. You catch those when you skim the log. Over time the prompt gets better because you add the edge cases as new examples.
You need a call-to-text fallback. Most VoIP and mobile providers offer a missed-call SMS responder. The text asks the customer to send a brief description of what they need. That description goes into the same scoring pipeline. For higher-end setups, an AI receptionist takes the call live and writes the structured summary straight to Make.
It can send an auto-decline, yes, but you should be careful. UK GDPR gives customers a right to challenge fully automated decisions that affect them. The safest pattern is to have a human glance at the cold list once a week before any decline goes out. Two minutes a week buys you the legal cover.
Yes, and the lower the volume the more important triage becomes. If you get 20 enquiries a week and three of them are hot, knowing which three is the difference between a £6,000 month and a £15,000 month. The Make Core plan covers you with 9,000 operations a month to spare.
Even better. Both have Make.com connectors. Make scores the enquiry first, then creates the job in your system with the score and reasoning attached as a note. Your existing pipeline stays in ServiceM8 or Jobber. Make sits in front, doing the triage your job management software was never built to do.
A CRM pipeline tracks a lead after you have decided to work it. Scoring decides whether to work it, and how fast. Most small CRMs assume every enquiry deserves the same attention. Scoring inverts the assumption and applies your time where it pays back the most.
My verdict
The five-minute response gap is real. The 21x conversion difference is real. The £24,000 of lost revenue per year is conservative for most one-van businesses I work with. Lead scoring is the single highest-ROI automation a small trades business can build in 2026, and Make.com makes it cheaper and easier than enterprise CRMs were five years ago. Spend a weekend on the build, two weeks on the rubric, and a month on the feedback loop. Then forget you ever did it manually. The AI marketing stack and the reviews automation stack are the next two automations I would build, in that order. Together the three of them replace most of what a part-time admin would do, for less than £30 a month.
