Customer segments are the few groups inside your customer list that each need a different message, and with AI they are a Skill job you can finish in one sitting of about an hour. You give the AI your list with names only, no contact details, and it proposes the groups, the rule for each, and the one thing each group should hear from you. Then you correct the draft with what only you know about the people behind the rows.
The output is short on purpose. Three to five groups, each with a name, a rule you can check in your records, and one message. That is enough to decide who gets which email this month.
- Shape
- Skill
- Sitting
- one sitting
The shape: Skill
A Skill is a reusable brief the AI runs the same way every time. This job takes that shape because your customer list never sits still. New people buy, regulars drift away, a service you added last spring starts to pull in a different crowd.
So today has two outputs. The first is the set of groups. The second is the brief that made them, saved where you can find it again: a project in Claude, a custom GPT, or a text file you paste from.
Next time, you export the list again, paste it under the same brief, and compare the new groups to the old ones. A group that grew, shrank or disappeared is news about your business. That is what makes the second run take fifteen minutes instead of an hour.
What you give the AI
The groups are only as true as the list, and the list should carry facts, not feelings. Gather these by name:
- An export of your customer list from wherever it lives: your booking tool, your invoicing app, your email tool, or a spreadsheet.
- First names or initials only. Delete email addresses, phone numbers, street addresses and payment details before you paste. The AI does not need them to propose groups, and they should not leave your machine for this job.
- What each person bought, by service or product name, as it appears on your invoices.
- The date of each purchase, or at least the first and the most recent.
- How many times each person has bought, if your tool counts it.
- The amount spent, as a total per person, if you are comfortable sharing it. Round numbers are fine.
- How each person found you, where you recorded it: referral, search, a directory, a walk-in.
- A one-line note on what your business does and for whom, so the AI reads the list in context.
- The messages you send now, listed by name: the monthly newsletter, the rebooking reminder, the holiday offer. If you send one email to everyone, write that down. It is an input too.
If your list has more than a hundred or so rows, attach the export as a file instead of pasting it, and add one line to the brief: "Count each group with code, not by reading." Counting a long list by eye is where the AI slips.
Anything sensitive stays out. If your business handles health, legal, financial or children's information, do not paste the list at all; describe the groups you suspect in plain words and ask the AI to pressure-test them.
What comes back
The AI returns a short table with one row per proposed group and four columns: the group's name, the rule that puts a customer in it, how many people on the list fall in it, and the one message that group needs from you. Under the table it lists the customers it could not place and the questions it could not answer from the list.
Paste this brief, then paste your list under it:
You are proposing customer segments for a small business.
Use only the list below. Do not invent customers, purchases or data the list does not contain.
The business: [one line on what it does and for whom].
The messages we send now: [list them, or write ONE EMAIL TO EVERYONE].
Propose three to five customer groups.
Each group must be defined by a rule that can be checked in the list, such as what they bought, how often, how recently, or how they found us.
Do not use age, gender or anything the list does not show.
Return a table with these columns:
Group name (plain words, under five words) | The rule | How many on the list | The one message this group needs from us
Rules:
- Place every customer in exactly one group, or list them as unplaced.
If a customer fits two, say which rule wins and why.
- Name the groups after what the customers do, not after a made-up person.
- The message column is one sentence, not a campaign.
- If two groups would get the same message, merge them.
- After the table, list any customers you could not place, by first name or initials.
- After that, list every question you could not answer from the list, so the owner can fill it in.
The list:
[PASTE THE LIST: first name or initials, what they bought, dates, times bought, total spent, how they found us]The "rule that can be checked in the list" line is the important one. Without it, the AI writes personas: invented people with hobbies and fears, none of whom you can find in your records or send an email to.
How to judge it
You can reject a bad set of groups in a minute. It has one of these faults:
- The groups describe people, not behavior. "Busy Brenda, 42, values convenience" is a persona. You cannot sort your list by it. A group you can use reads like "Bought once, more than a year ago."
- The counts do not add up. Add the "how many" column. If it is more or less than your list, customers were double-counted or dropped.
- There are too many groups. Seven groups for a list of ninety people means you will write to none of them. Ask the AI to merge down to the three that need the most different messages.
- Two groups get the same message. Then they are one group.
- The message is a campaign plan. You want "Your next visit is due, here is the link to book," not a four-week sequence with subject lines. The sequence is a separate job.
- It uses data you did not give it. If a group depends on age, income or location and your list has none of those, the AI guessed.
- Nobody is left unplaced and no questions are listed. Real lists are messy. A tidy answer with no loose ends usually means the AI smoothed over the mess.
A good result reads like something you half knew already, with the counts that make it real. One group will usually be bigger than you expected. Another will be smaller.
What stays with you
The AI can sort rows. It cannot see why the people behind them buy, stay or leave.
- Why a group exists. The AI sees that twelve people bought in December and never came back. You know whether they were gift buyers, out-of-town visitors, or unhappy. Each answer means a different message.
- Which group matters most. The AI can count heads. You know which group pays the bills, which one refers friends, and which one costs you more time than it is worth.
- The privacy call. What leaves your machine is your decision. Strip the list before you paste, check your plan's data settings, and follow any rules your industry sets before the AI's.
- The send button. A group is a list of real people. Every message to them goes out under your name, so read it as one of them would before it leaves.
- Keeping the list honest. The groups are only as current as your records. If half your customers book by phone and never make it into the tool, the groups describe the other half.
The groups are a draft until you have read each one and thought of two real customers who belong in it. If you cannot, the group is not real yet.
An editorial example
This is an invented business, used only to show the shape. Say you run a small bike repair shop with one mechanic and a booking page.
You export the customer list from your booking tool, delete the email and phone columns, and keep first names, the service, the booking dates and times, and how many visits each person has made. You send one email to everyone, a seasonal tune-up reminder, so you write that down.
The AI comes back with four groups. "Tune-up regulars" have booked a tune-up two or more times; the message is a reminder when their usual month comes around. "One flat tire" came in once for a puncture and never returned; the message is that you do full tune-ups too. "New bike, first check" booked the new-bike check and nothing since; the message is what to bring in after the first few hundred miles. "Commuters" book repairs on weekday mornings; the message is that you offer same-week drop-off times.
You correct one thing. The AI split "Commuters" out by booking time, but you know most of them are also tune-up regulars, so you merge the two and keep the commuter detail in the reminder. Three groups, three messages, and one email that used to go to everyone now goes three ways.
Where the groups point next
Once you have the groups, two jobs follow.
The first is the sentence each group hears about why you, which is the work of a value proposition. The second is the path one group walks with you from first touch to the next visit, which a customer journey map lays out row by row. If your list leaves the big question open, which is why people buy, a short customer survey is the way to ask them.
Save the brief before you close the chat. The second run is where the Skill pays for itself.
Of every job on this list, social media is the one that comes back every week, and Boomp is software that does that one for you. See ten posts made from your website, free
