The winner is the scheme that breaks even on the smallest rise in visits, and one month of your own bookings shows which. Strip the names from the export, then ask an AI assistant to cost a free tenth cut, points and a referral reward. In the worked example, referral wins: it costs you only when someone new sits down.
A month of data prices each scheme precisely but cannot prove how clients will react. That part is an assumption, so the method below squeezes it into one number: how many extra visits a month each scheme must create to pay for itself. Compare that number with what you honestly expect, then run an eight-week pilot with a spending cap.
Export the month and take the names out
Ask your booking app or till for a visit-level export covering the last full calendar month: one row per visit, not a summary report. Summary screens hide the figure the model needs most, which is how often the same person came back. Keep the ten columns below and delete the rest.
| Column | Why the model needs it |
|---|---|
| Visit date and start time | Shows which slots stay empty and which are always full |
| Client ID | Counts how many times each person returned |
| Service | Separates a standard cut from beard trims and add-ons |
| Barber | Matters if pay or commission differs by chair |
| Price paid | Gives revenue and the average ticket |
| Tip | Kept apart so tips do not inflate prices |
| Status | Completed, no-show or cancelled |
| Booking source | Online, phone or walk-in |
| First visit | Yes or no, so new-client numbers are real |
| Discount or voucher | Stops an existing promotion being counted twice |
The assistant has no use for names, phone numbers or email addresses, so take them out before you upload anything. Many apps already export a client number; keep that. If yours exports names only, sort the sheet by name, type 1 in the first ID cell, enter =IF(A3=A2,B2,B2+1) in the cell below (names in column A, IDs in column B), fill it down and delete column A. Keep your original export to yourself. Then use an assistant account where the model-training switch in the privacy settings is off, or a business plan, which does not train on your content by default.
Ask for a cleaning report before any analysis. In the example shop (three chairs, 26 trading days) the raw file had 560 rows. Forty-one were cancellations or no-shows, which left 519 paid visits from 371 different clients at an average of $34 a visit, tips set aside.
A placeholder that looks like your best client
Now take a different shop, to see what goes wrong. Its till logs every anonymous walk-in under one dummy client. Uploaded untouched, the first summary announces that the best client visited 143 times. In truth 143 strangers have been counted as one loyal regular, which shrinks the number of distinct clients and inflates every visits-per-client figure. Make the first output a list of the five highest visit counts. Any ID with more than about ten visits in a month is a placeholder or a data error, not a person.
Spreadsheet clean-up prompts for this kind of mess are in the tutorial on cleaning messy data, and the barbershop booking app comparison covers how different apps log walk-ins.
Profile the month before costing any reward
Attach the cleaned file and paste the prompt below. It asks for facts only, so a wrong figure is easy to catch.
I'm uploading one month of visit data for a three-chair barbershop. Names and phone numbers have been removed; client_id identifies each person.
Do these in order and show your working:
1. Cleaning report: rows received, cancelled or no-show rows removed, duplicate rows removed, and any client_id with more than 10 visits (list them).
2. Paid visits, distinct clients, average price paid per visit (tips excluded) and the average price of a standard cut.
3. Visits per client: how many clients had 1, 2, 3 and 4 or more visits, and what share of all visits each group made.
4. First-visit clients, and the share of visits that were walk-ins.
5. Opening hours are Monday to Friday 9 to 6 and Saturday 8 to 5. Count the unsold 30-minute slots, by weekday and by before or after 3pm.
Do not recommend anything yet. If a figure depends on a guess, say so.The example shop's answer came back like this (illustrative figures):
| Visits in the month | Clients | Visits | Share of visits (rounded) |
|---|---|---|---|
| 1 | 262 | 262 | 50.5% |
| 2 | 79 | 158 | 30.4% |
| 3 | 21 | 63 | 12.1% |
| 4 or more | 9 | 36 | 6.9% |
| Total | 371 | 519 | 100% |
Three findings from that table and the rest of the answer shape everything that follows.
- Concentration. The 109 clients who came more than once (29% of clients) made 257 of the 519 visits, close to half. Any reward tied to visit counts pays mostly this group, for what they already do.
- Walk-ins. 197 visits (38%) had no booking. A scheme that starts with an app sign-up misses most of those unless the till can stamp or credit them on the spot.
- Idle time. 150 half-hour slots stayed unsold from Monday to Wednesday before 3pm, and Saturday had none. A reward redeemed in a quiet slot costs less in lost capacity than one redeemed on a full Saturday.
Two more figures go into the model: 47 first-ever visits (13% of clients), and a standard cut price of $28 against the $34 average ticket, because beard trims and add-ons lift the average.
Write each scheme as a rule a barber can say in one breath
A model needs rules, not ideas. Write each scheme as a precise rule first, because every clause turns into either an assumption or a cost control later.
| Scheme | Said at the chair | Rule as it will be printed |
|---|---|---|
| A. Free tenth cut | "Nine cuts, the tenth is on us." | One stamp per paid visit of $25 or more. Nine stamps earn one free standard cut, Monday to Thursday before 3pm. Card valid 12 months from the first stamp. |
| B. Points | "A point for every dollar, 100 points is $5 off." | 1 point per $1 paid, tips excluded. 100 points = $5 off a later visit. Points expire after six months without a visit. No cash value. |
| C. Referral | "Bring a mate and you both save $10." | A new client who gives your name saves $10 on a first visit. Once that visit is paid, you get $10 off your next one. One reward per new client. |
Notice how much of each rule is really cost control. "Standard cut" stops the reward swallowing a $45 cut and beard trim. "Monday to Thursday before 3pm" aims redemptions at slots the calendar shows as empty. The expiry date becomes the redemption assumption, and "once that visit is paid" makes the referral a pay-on-results reward.
One reward to leave out: stamps or points for a Google review. Google's review rules ban incentives, so keep the scheme about visits and referrals.
Settle who pays before the model runs, because it changes the cost line. If your barbers are salaried, a free cut costs the shop the price of the cut. If they earn 50% commission and the shop still pays them on the full price, the shop carries the whole $28. If they are paid only on money taken, they carry half of it, and the stamp card will quietly stop appearing at their chair. Chair renters are a separate case: a shop-wide card only works if the renters agree to fund it.
One prompt that builds the three-scheme workbook
With the profile confirmed, request the whole model in one go. Tell the assistant to keep every guess on an Assumptions tab and to use formulas rather than typed results, so you can change a number and watch the totals move. ChatGPT and Claude both run code on an uploaded CSV, and Claude can hand back an .xlsx with working formulas; Gemini accepts spreadsheet uploads too. If your assistant only returns tables, ask it to write each formula as text and paste them into your own sheet.
Using the cleaned visit data, build a workbook with an Assumptions tab, one tab per scheme and a Results tab. Put every guess on the Assumptions tab, labelled, and use formulas everywhere else.
Facts from the data: 519 paid visits, 371 clients, average ticket $34, standard cut $28, variable cost per visit $4 (consumables and card fees), 197 walk-in visits.
Scheme A, free tenth cut: one stamp per paid visit of $25 or more; nine stamps earn one free standard cut. Assumptions: 80% of visits earn stamps, 70% of earned free cuts are redeemed.
Scheme B, points: 1 point per $1 paid; 100 points = $5 off. Assumptions: 60% of spend is by members, 70% of earned points are redeemed.
Scheme C, referral: the friend saves $10 on a first visit; the referrer gets $10 off a later visit once the friend has paid. Assumptions: 4, 8 and 14 referred clients a month as three cases; 25% of them would have come anyway; 60% of referred clients return; returners average 5 further visits in the first year.
For A and B, work out the monthly cost, then the extra visits a month needed to break even. An extra visit is worth the ticket minus the variable cost, and extra visits earn stamps or points too. For C, work out the expected value of one referred client over twelve months and the share of windfall referrals at which the scheme stops paying.
Add a sensitivity table: net result a month if visits rise 0%, 3%, 6% and 10%, for A and B.
Finally list every assumption in plain English, ordered by how much it moves the result.Two settings in that prompt do the real work. The $4 variable cost means an extra visit is worth $30, not the full $34, so the assistant cannot flatter a scheme by crediting every added visit with the whole ticket. And the instruction to rank assumptions by influence tells you where to spend your attention: nearly always on redemption and on the rise in visits.
Reading the results: cost, break-even and what you must believe
This is the Results tab for the example shop, after the corrections described in the next section.
| Scheme | Cost a month | Extra visits to break even | What you have to believe |
|---|---|---|---|
| A. Free tenth cut | About $904 (32 free cuts at $28) | 32 a month (6.2% more visits) | A visible card gets regulars to book sooner often enough to add about 32 visits |
| B. Points | About $371 | 13 a month (2.4% more) | Members return slightly sooner, and a small lift is enough |
| C. Referral | $20 per referred client, paid after their first visit | Loses only if more than 83% of referred friends would have come anyway | Enough clients actually tell a friend: 4 to 14 a month |
The sensitivity table answers the question "what if I'm wrong about the lift?" Net result a month, after reward cost:
| Rise in visits | Scheme A net a month | Scheme B net a month |
|---|---|---|
| 0% | -$904 | -$371 |
| 3% (16 visits) | -$464 | +$85 |
| 6% (31 visits) | -$24 | +$541 |
| 10% (52 visits) | +$562 | +$1,149 |
Read these as questions about belief, not forecasts. The card needs a 6% rise before it earns anything, and below that it simply costs money. Points clear their bar at 2.4%, but a points balance is harder to see from the chair than nine stamps on a card, so whether a 3% lift is realistic is your call, and the pilot settles it.
Redemption is the setting that moves the verdict most. Drop the card's redemption from 70% to 50% and its break-even falls to 4.3%; raise it to 90% and it climbs to 8.1%. Points shift from 1.7% to 3.2% across the same range. That is why the prompt asked for assumptions ranked by influence, and why your expiry rule matters more than the reward's size.
A rough decision sequence for the example shop:
- Strike any scheme whose break-even rise is higher than you would swear to. If you cannot picture a card adding 32 visits a month, do not print it.
- Among those left, prefer the one that is cheapest to be wrong about: referral first, then points, then the card.
- Pilot one scheme, not three. Two live at once make any change in visits impossible to credit.
Checking the arithmetic before you trust a number
Assistants slip on arithmetic occasionally and on reading your rules often. In this illustration the first answer came back with figures roughly twice as high as they should be, for reasons that are easy to find once you look for them.
| Line | First pass | Corrected | Why it changed |
|---|---|---|---|
| Visits that earn stamps | 100% | 80% | Walk-ins and clients who decline a card |
| Earned free cuts redeemed | 100% | 70% | Cards get lost, abandoned or expire |
| Value of a free cut | $34 | $28 | The rule says standard cut only |
| Value of an extra visit | $34 | $30 | $4 of consumables and card fees |
| Scheme A cost a month | $1,961 | $904 | Result of the four lines above |
| Scheme A break-even visits | 65 | 32 | Lower cost outweighs the lower visit value |
Two of those errors pull in opposite directions: assuming everyone redeems overstates cost, while valuing an extra visit at the full ticket understates the visits needed. Errors that partly cancel are the dangerous kind, because the answer still looks sensible.
Verify three numbers by hand before anything else:
- Scheme A: 519 visits x 80% x 70% / 9 = 32.3 free cuts, and 32.3 x $28 = $904.
- Scheme B: 519 x $34 = $17,646 of spend, x 60% members x 5% value x 70% redeemed = $371.
- Scheme C: 8 referred clients x $70 expected value = $560, with 8 x $20 = $160 paid out in rewards.
If one of your hand sums disagrees with the sheet by more than rounding, ask the assistant to show the cell and formula behind that line rather than to try again. Then change one assumption on purpose: set redemption to 0% and the card's cost should fall to $0; double the free-cut price and the cost should double. A model that fails either test is wrong somewhere the headline figures hide.
Referral: a reward you pay only when someone new sits down
Referral behaves differently from the other two because nothing is paid until a new client has been in the chair. Take one referred friend. They pay $24 on a first visit after the $10 saving, which leaves $20 once the $4 variable cost is taken off. The referrer's $10 credit comes off a later visit, so the first visit nets $10. If 60% of such clients return and average five more visits in the year at $30 each, that adds $90 on average, and the expected value of a genuinely new client is $100 over twelve months.
Some referred friends would have booked anyway. In the model they are windfalls: they cost $20 in rewards and add no new revenue. With 25% windfalls, one referral is worth 0.75 x $100 minus 0.25 x $20, which is $70. The scheme only stops paying when windfalls pass 83% of referrals, so its downside is far smaller than a card's.
| Referred clients a month | Reward spend | Value over their first twelve months |
|---|---|---|
| 4 | $80 | $280 |
| 8 | $160 | $560 |
| 14 | $280 | $980 |
The card and points figures are monthly flows, while this table counts each cohort's first year, so referral pays back more slowly. Its weak spot is volume: it only works as far as clients actually mention it. Where the month cannot help is the return rate. An export of one month cannot say how many first-timers come back, so pull the month from three months earlier and count how many of its first-timers booked again. If your app does not hold that history, start with a cautious guess and replace it with your own figure after eight weeks.
Three controls keep the scheme honest. Record the referrer at the till in a "referred by" note and match it to their client ID. Pay the credit only after the friend's visit is paid. Refuse a reward when the two clients share a phone number or card. Also ask every new client how they heard about the shop for a month before launch, so you know how many referrals you were already getting. Automated referral reminders can then nudge clients without adding work at the chair.
The invitation to existing clients can be plain:
[first name], know someone who needs a decent cut? If they book a first visit and mention your name, they save $10 and you get $10 off your next cut. No codes, no app. Book here: [link]
Only message clients who have agreed to marketing texts, using the consent your booking app already records; if you are unsure what your records allow, ask a data-protection adviser.
Cap the cost before the first card is printed
A scheme can be sound on paper and still run away from you, so fix four limits in advance. First, a monthly cap on free cuts: set it a little above the model's figure (40, say, against the 32 modelled), and have the assistant compare the stamp log with the cap each week. Second, an end date printed on the card, so the pilot is a trial rather than a promise. Third, a no-stacking line. Fourth, an expiry that fits your clients' rhythm.
The expiry sum is quick. Nine stamps at a visit every two weeks take 18 weeks. At four weeks they take 36, and at five weeks 45, almost the whole year. A client who comes every six weeks needs 54 weeks, so a 12-month card never pays out for them. Your visits-per-client table tells you which group you are rewarding: the 109 clients who came more than once in the month are the only ones likely to finish a card within six months.
Here is the back of the card for scheme A, filled in rather than left as a template:
Tenth cut free. Collect a stamp for every paid visit of $25 or more; staff initial each one. Nine stamps earn one standard cut free, Monday to Thursday before 3pm. One card per client, no cash value, not combined with other offers. Valid for 12 months from the first stamp. We may end this offer with 14 days' notice.
Each sentence is carrying a line of the model: the $25 threshold, the standard cut, the quiet-slot window, the expiry and the end date.
An eight-week pilot and the three numbers to read
Choose one scheme and run it for eight weeks. Do not change prices in the same period. If a price rise is due, announce it first, using the approach in the tutorial on announcing a price rise, and start the pilot afterwards. Otherwise you cannot tell which change moved the numbers.
From the example shop's month, the baseline is 1.40 visits per client (519 visits across 371 clients). Three figures decide the pilot:
- Visits per active client every four weeks. If this does not move well above 1.40, the card or points are paying for visits that were happening anyway.
- Extra visits against break-even. For the card that means 32 more visits a month than the baseline period. Compare with the same weeks last year if you have them, because school holidays and wedding season move visits more than any card does.
- Spend against the cap. Free cuts given, valued at $28, and referral credits issued.
Make the weekly review a five-minute job. Export the week's visits, strip the names and ask: "Compare this week with the baseline figures above. List stamps issued, free cuts redeemed, projected month-end cost against the $1,120 cap and any client with more than three stamps in a day." That last clause catches stamp stuffing, which is the commonest way a card goes wrong.
The stop rule is the part owners skip. If after eight weeks the extra visits sit below break-even, end the scheme or change one rule, not three. A busy-feeling shop is not evidence, because regulars who would have come anyway fill the chairs too.
Use the chair as part of the scheme. The wording below separates a message that nudges from one that announces:
Before: "Join our loyalty programme and earn rewards! T&Cs apply."
After: "[first name], your card is at 6 of 9. Three more cuts and the tenth is on us. Shall I pencil in your next one for three weeks from today?"
The second version is built around the next booking, which is the thing a loyalty scheme is for. For clients who have drifted away, the same data feeds the win-back messages in the tutorial on lapsed clients.
When the month you exported is not a normal one
A single month is a snapshot, and some months mislead. December and the weeks before a wedding season inflate visits and ticket values; a holiday month deflates them. If your month was unusual, export the nearest ordinary one and rerun the prompts, which takes minutes. If it contained a promotion, flag those visits with the discount column so the assistant can treat them separately.
Shops under three months old have too little history for any of this. Use bookings a week instead of a month, keep the card terms short and treat the whole exercise as a first draft. Shops where most visits are anonymous walk-ins can still cost a card, because stamps are earned per visit, but they cannot say who their regulars are. For them a paper card is the sensible start, with a count of cards completed after eight weeks.
If your barbers charge different prices, run the model per chair and total the results; a free cut at one chair can cost a third more than at another. For the background to loyalty design beyond the barbershop, including tiers and personalised offers, the loyalty programme tutorial with AI personalisation covers what this one leaves out.
Questions barbers ask before launching a loyalty card
Do I need loyalty software, or will a paper card do?
A paper card is enough for an eight-week pilot, and it works for walk-ins at the chair. The catch is data: paper leaves no visit trail, so you cannot measure the lift afterwards. If your booking app can stamp or credit visits automatically, use that for the pilot and keep paper as the backup for walk-ins.
Should the reward be a free cut or a dollar discount?
Either can work. A free cut is quicker to say at the chair and feels generous, while a dollar discount is easier to cap and to vary by day. Write both as rules and cost them in the same workbook, because a free standard cut and a $5 discount differ in price and in who tends to redeem them.
What if a client loses a stamp card halfway?
Keep a stamp log at the till as well as the card: client ID, date, stamp count and the barber's initials. A lost card can then be rebuilt in a minute, and the same log feeds the weekly stamp-stuffing check. If your booking app can hold a note or tag per client, record the count there instead. A card you cannot rebuild annoys exactly the regulars it was meant to keep.
Is a loyalty scheme worth it for a one-chair barber?
Only if the chair has gaps. A fully booked barber gains nothing from a card, because it discounts visits that would happen anyway; a referral reward that brings new clients into the quiet weekday slots is the better test. The sums are the same with smaller numbers, but one month of a single chair is noisy, so compare the pilot with the same weeks last year where you can.
Can I run a card and a referral reward together?
Yes, as long as the rules stop stacking: the referred friend's first visit earns a stamp, but no second discount sits on top of the referral one. For the pilot, add one scheme at a time. If you launch two together and visits rise, you will not know which one did it.
Further reads
- What Is the Best AI for a Barber Shop? Start With These Seven Jobs — Seven barbershop jobs where AI saves the most time
- How Salons Use AI to Rebook Clients and Fill Gaps in the Diary — Rebooking prompts that fill the quiet slots a reward should target
- How to Cut Salon No-Shows With AI Reminders and Deposits — Protect the slots your loyalty clients book
- Booksy for Barbers: Costs, Boost Fees and the AI Receptionist Beta — What one popular barber booking app costs before you add rewards
- How Much Time Can AI Save a Barber Each Day? — Where a barber's day goes and what AI can take back
- Gym Referral Programme: Set Rewards and Test the Maths With AI — The same referral sums for a gym, with different retention
- Restaurant Loyalty Programme: Simple Rewards and AI-Written Offers — How a restaurant tests simple rewards on its own sales
- How to Plan a Month of Barbershop Instagram Content With AI — Posts that announce the scheme without sounding like an advert
- Virtual Receptionist for a Barbershop: Human, AI or Cloud Phone? — Three ways a two-chair barbershop can stop missing calls, priced per month, with a sum you can redo using your own call log.
- AI Receptionist for Barbers: Walk-In Queues, Waitlists and Texts — Run a walk-in queue with AI: written queue rules, wait-time estimates, a text-me-when-ready flow and what to do with no-shows.
- Customer Loyalty Programmes for Retail Stores: Can AI Design One? — A small shop's loyalty programme drafted with AI: points, stamps and tiers costed line by line, plus the five corrections the draft needed.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Anthropic support page 'Create and edit files with Claude'; OpenAI help article 'Data analysis with ChatGPT'; Google Gemini Apps Help page on uploading and analysing files. The shop, figures and results in the examples are illustrative.