AI for identifying your best customer type, from your own books

6 days ago
3 min read

Updated: 4 days ago

Introduction

Ask an owner to describe their ideal customer and you will get a description of the customer they enjoy working with. Ask the accounts to describe it and you get a different answer, because the profitable customer and the pleasant customer are not reliably the same person.

This matters because everything downstream depends on the answer: which enquiries to chase, where to advertise, what to say, which segments to leave alone. Getting it wrong is expensive in a way that is hard to see, since the business still grows, just more slowly and at a lower margin than it needed to. The evidence to settle it is already in your invoices and job records.

1. AI for identifying your best customer type is a question about profit per customer

Not revenue and not preference.

Rank every customer by gross profit contribution over two or three years, not by turnover. Then look at what the top group has in common. That commonality is your answer, whether or not you like it.

2. Include the cost of serving them

The part that changes the ranking.

Site visits, revisions, support calls, chasing payment, meetings, rework. Two customers at the same revenue and nominal margin can differ enormously once the servicing effort is counted, and the difference is frequently large enough to reverse their order.

3. Look for attributes you can actually target

The finding has to be usable.

Sector, size, location type, how they found you, what they bought first, whether they have an in-house equivalent function. These are things you can look for in an enquiry. "Reasonable people" is a true observation and not a targetable one.

4. Check retention alongside profitability

Duration multiplies everything.

A moderately profitable customer who stays six years is worth more than a highly profitable one who buys once. Ranking by profit per year of relationship, rather than profit per job, tends to identify a different and better segment.

5. Test the segment against your losses too

Where you lose is informative.

Look at which enquiry types you fail to convert and which customers leave. A segment you convert well and retain badly is not your best segment, and profitability alone will not show you that.

6. Expect the answer to be narrower than you want

This is the useful part.

The result is often a specific and uncomfortably small segment — one sector, one size band, one kind of job. Owners resist this because it appears to shrink the market. It does not shrink the market; it concentrates the effort that was previously spread across segments that never paid.

7. Verify by looking at actual accounts

Never act on the summary alone.

Pull the ten accounts the analysis names as best and read their histories. Some will turn out to be one-off projects, misallocated costs, or a relationship that has since ended. The pattern only counts if the individual cases support it.

8. Turn the profile into a scoring rule for enquiries

The point of the exercise.

Two or three attributes, checkable at first contact, that raise or lower how hard you pursue an enquiry. Without this step the analysis is a report, and reports do not change what gets quoted on Tuesday.

9. Redo it annually, because it moves

Segments age.

Input costs, competition and your own capabilities all change, and a segment that was your best three years ago may now be your most contested and least profitable. An annual repeat is straightforward once the first one is done.

Keep the segments you are exiting in view rather than cutting them abruptly. Some low-margin work covers fixed costs or feeds better work later, and the ranking alone will not tell you which.

Conclusion

Rank customers by profit contribution net of servicing cost, because that answer is usually not the one you would have guessed.

Count the cost of serving each customer, look for attributes you can actually identify in an enquiry, weigh retention alongside margin because duration multiplies everything, test the profile against your losses as well as your wins, accept that the answer will be narrower than feels comfortable, verify by reading the individual accounts the analysis names, convert the profile into a scoring rule your team can apply at first contact, and repeat the whole exercise annually.