Corrections, 7 October 2026: The UK partnership distinction has been corrected against ICO guidance. Unsupported response-speed and standard outreach-return claims have been replaced with instructions to measure genuine enquiries, costs and completed jobs. A chosen trial size is an illustration, not a forecast. AI-generated contact details still require independent verification.
How to Use AI to Find New Customers
Most businesses looking for new customers are solving the wrong problem. They don’t have a shortage of prospects — they have a shortage of visibility, or a conversion rate that quietly wastes the enquiries they already get, and neither is fixed by generating more leads.
So this guide starts with a diagnostic that takes ten minutes and frequently redirects the whole exercise. Then it gives you the workflow for whichever answer you got: getting found, mining what you already have, or genuine outbound.
Every prompt is copy-paste ready, and the free routes come before the paid ones because for most small businesses they work better.
Step 1: Diagnose before you do anything
Three numbers tell you which problem you have
Count these for last month: how many enquiries you received, how many you quoted for, and how many became jobs. Then match your pattern below.
Which problem you actually have
| Few enquiries | A visibility problem. Nobody knows you exist. Go to Steps 3 and 4 — mostly free, and the highest return available to you. |
| Plenty, few won | A conversion problem. More leads would add more waste. Fix quoting speed, follow-up and qualification first — Step 6. |
| Good conversion, want more | Genuinely a volume problem. Steps 3 and 4 still come first, then outbound in Step 5 if you sell to businesses. |
| Wrong enquiries | A targeting problem. You’re visible to the wrong people. Step 2 fixes this before anything else. |
Most people arrive at this topic assuming they’re in the third row and turn out to be in the second. If you’re quoting plenty and winning few, buying an outbound tool actively wastes money.
Step 2: Define who you’re actually looking for
Work backwards from your best existing customers
Don’t invent a customer profile — derive it from the customers you already have who were profitable, straightforward and glad they hired you. That’s a real pattern rather than an aspiration.
MY FIVE BEST CUSTOMERS — profitable, easy to work with, happy with the result:
1. [who they were, what the job was, what they paid, how they found me, why it went well]
2. [same]
3. [same]
4. [same]
5. [same]
MY THREE WORST — unprofitable, difficult, or unhappy:
1. [same detail]
2. [same]
3. [same]
Tell me:
1. What the good ones have in common — situation, not demographics
2. What the bad ones have in common, and what warning sign I could spot at enquiry stage
3. Where the good ones came from, and whether that route is repeatable
4. What I should say in my marketing to attract more of the first group
5. What I should say — or stop saying — to attract fewer of the second
Be specific. If the pattern is that my best customers all came through one referral source, say so plainly.
Point 2 earns its place: knowing the warning signs of a bad customer at enquiry stage saves more money than any lead generation does.
Step 3: Mine what you already have
Past customers and referrals, before anything cold
The warmest prospects in your business are people who already bought from you and people they know. Most businesses never systematically contact either, which makes this the cheapest customer acquisition available.
MY BUSINESS: [what you do]
WHO I WANT MORE OF: [from Step 2]
WHAT I’VE DONE FOR THEM: [the typical job]
Rules:
– Under 120 words
– Don’t offer a reward or incentive
– Be specific about who I’m looking for — “someone with a [specific situation]” rather than “anyone who needs [service]”
– Make it easy to say no without awkwardness
– One clear action: what exactly should they do if someone comes to mind
– Sound like a person, not a campaign
Then write a shorter version for text or WhatsApp, and tell me the best time to send this relative to finishing a job.
The specificity instruction is what makes referral requests work. “Do you know anyone who needs a plumber?” gets nothing; “do you know anyone in a Victorian terrace with the same cold-radiator problem you had?” gets a name, because it triggers an actual memory.
CONTEXT: [what they bought, roughly when, what might have changed since]
Rules: under 100 words; acknowledge the gap without guilt-tripping; don’t lead with a discount; give one genuine reason to get back in touch now; ask a question they can answer in one line.
Also suggest one thing I could offer that isn’t a price cut.
Step 4: Get found by people already looking
Answer the questions your buyers are typing
The highest-quality customers are the ones who came looking for you. The way to reach them is to answer, properly, the questions they search before buying — which you already know, because they ask you the same ones on the phone.
QUESTIONS I GET ASKED CONSTANTLY, in customers’ own words: [list 10]
WHAT I DO: [services]
WHERE: [areas]
For each question tell me:
1. Whether it’s worth a page on my site — be honest, some aren’t
2. The title, phrased the way a customer would actually search
3. The three things the page must answer to be genuinely useful
4. Whether I should publish a price on it
5. Whether someone searching this is close to buying or just researching
Then rank them by which would bring me the most actual enquiries — not the most traffic.
Do not give me keyword volumes or search statistics; I’ll check those myself.
That last instruction matters. AI will invent search volumes convincingly, so treat prioritisation as its judgement and verify the numbers with a real tool if you need them.
Step 5: Outbound — only if it genuinely suits you
Fifty researched approaches by hand, before any automation
Cold outreach works for businesses selling meaningful-value services to other businesses. It does not work for most consumer-facing local trades, and automating a message that hasn’t been proven by hand simply distributes failure faster while damaging your domain reputation.
So the sequence is: build a list of fifty genuinely well-matched prospects, research them properly, send by hand, and only automate once you have a message that gets replies.
WHAT I DO: [service, and the specific problem it solves]
WHO I’M CONTACTING: [company, sector, size, role of the person]
WHAT I’VE FOUND OUT ABOUT THEM: [paste what you found — recent news, a job advert, a change, something on their site]
WHY I THINK THEY SPECIFICALLY HAVE THIS PROBLEM: [your reasoning]
Write a first approach that:
– Opens with the specific thing I found, not with who I am
– States in one sentence why I’m contacting them rather than anyone else
– Makes one small, easy ask — not a meeting request
– Is under 90 words
– Has no attachments, no links, no bullet points
– Sounds like one person emailing another
If what I’ve found out about them isn’t actually a good reason to contact them, tell me so rather than writing the email.
That final line is the useful safeguard. If the model can’t find a genuine reason to approach someone, the honest answer is that they’re not a prospect — and hearing that saves you from sending something generic.
Check the entity type before emailing rather than blasting a list, keep a genuine immediate opt-out, don’t use data whose provenance you can’t establish, and read the ICO’s direct marketing guidance. “The tool sent it automatically” is not a defence. Our listicle on tools for finding new customers covers this and the tool landscape in more detail.
Step 6: Fix the leak before you widen the pipe
If you diagnosed a conversion problem, this is your whole guide
Three things convert more enquiries than any lead source, and all three are free.
Speed. Time an identified test enquiry through your form and check that someone receives it, acknowledges it and follows up. Log response time alongside qualified enquiries and completed jobs so you can assess whether the process helps your business.
Follow-up. Most quotes that go quiet were forgotten, not rejected. Set a rule: follow up after three working days and again after ten, then stop.
Qualification. If you’re quoting for people who were never going to buy, ask better questions earlier rather than generating more of them.
MY BEST CUSTOMERS LOOK LIKE: [from Step 2]
MY WORST LOOK LIKE: [from Step 2]
MY TYPICAL PRICE: [range]
WHAT I NEED TO KNOW BEFORE QUOTING: [list]
Give me:
1. Five questions to ask at first contact that would identify a good fit quickly
2. For each, what a concerning answer sounds like
3. One question that politely surfaces budget without asking “what’s your budget”
4. How to decline politely when someone isn’t a fit — actual wording
5. What I should put on my website so poor-fit enquiries don’t reach me at all
Step 7: Track what actually brought them in
One question, asked every time
Ask every new customer how they found you, and write it down. Not in analytics — in a note, next to the job value. Within three months you’ll know which route produces customers rather than which produces clicks.
Most small businesses discover something they didn’t expect: that referrals produce their most profitable work, that a channel they’ve been paying for produces enquiries but no jobs, or that a single search question drives a disproportionate share of their revenue. That knowledge is worth more than any tool on this page, and collecting it costs one sentence per customer.
Then spend where the profitable customers came from, not where the enquiries came from. Those are frequently different channels.
Before you spend money on finding customers
- You’ve counted enquiries, quotes and wins for last month.
- You know which of the four problems you have.
- Your Google Business Profile is complete, if you’re local.
- You’ve asked past customers for referrals — specifically, not generally.
- Your response time is tested, not assumed.
- A follow-up rule exists and is in your diary.
- For outbound: fifty by hand first, and entity types checked.
- You’re recording how every customer found you.
Frequently asked questions
Can AI just generate me a list of leads?
AI-generated prospect names and contact details need verification against real sources. If you use a database, establish where its data came from and whether your planned contact is lawful. UK email rules distinguish corporate subscribers, including LLPs and Scottish partnerships, from sole traders and certain other partnerships; check the ICO guidance before using a list. Use AI to organise research about businesses you have verified, rather than treating its output as confirmed contact data.
How many new customers should I expect from this?
Set a trial period and record qualified enquiries, completed jobs, cost and time for each channel. Referral requests, accurate local listings, useful content and lawful outreach may produce different results in different businesses. For an illustrative outreach trial, you might review 50 verified prospects, but that is a chosen sample size, not a forecast of replies or sales. Compare outcomes before scaling; this guide has not measured a standard two-or-three-conversation return.
Should I offer a referral reward?
Usually not, and the prompt in Step 3 deliberately excludes it. Three reasons. People refer because they want to help someone they like, and introducing money can make the act feel transactional — some will refer less rather than more. It also changes what a referral is worth to the recipient, since a recommendation someone was paid for carries less weight than one freely given, and that can come out awkwardly. And in the UK, if a reward could be read as an incentive attached to a review or endorsement, disclosure obligations arise under consumer protection rules, which is more complication than the scheme is worth for most small businesses. What works better: ask specifically rather than generally, ask at the right moment — right after a job has gone well — and thank people properly when they do refer, which costs nothing and makes them likely to do it again.
What if I’ve genuinely tried everything and nothing works?
Then the problem is probably upstream of marketing, and that’s worth facing rather than buying another tool. Three possibilities, in order of likelihood. Your offer isn’t compelling enough at the price — meaning people understand it and don’t want it, which is a product or positioning problem rather than a promotion one. You’re visible to the wrong people, so enquiries arrive from those who were never going to buy at your price. Or your conversion is failing invisibly: slow replies, unclear quotes, no follow-up. The fastest way to find out is to ask ten people who didn’t buy why they didn’t — genuinely ask, without defending. Most businesses never do this and it’s the single most informative exercise available. AI can help you draft that outreach and analyse the answers, but the answers have to come from real people. If several say the same thing, that’s your actual problem, and no amount of lead generation would have fixed it.
The bottom line
Diagnose first, because most people looking for more customers have a conversion or visibility problem and buying an outbound tool makes it worse. Derive your target customer from the best ones you already have rather than inventing a profile. Then work outward in order of warmth: past customers and referrals today, free local visibility this week, content aimed at real buying questions this month, and cold outbound only if you sell to businesses and only after fifty researched approaches by hand have proven the message. Ask every new customer how they found you, and spend where the profitable ones came from.
UK direct marketing rules differ depending on whether you contact individuals, sole traders, partnerships or limited companies — verify current requirements in the ICO’s direct marketing guidance before any outbound campaign. AI-generated contact data, search volumes and market figures are frequently inaccurate; verify against real sources. Nothing here is legal advice.