Fix the share of each member's spend you can afford to give back, pick points, stamps or tiers to fit it, then have AI draft the rules and show its arithmetic. Check every figure in a spreadsheet before customers see it. Start with one simple design and add tiers only after 90 days of real data.
AI is good at three jobs here: drafting a rulebook from your numbers, attacking it the way a customer or a tired shop assistant would, and writing the messages. It is unreliable at the thing that matters most, which is arithmetic across stacked rules. So every sum below is worked by hand, and the sample draft comes with the mistakes I would expect to find in it.
Fix the reward budget before you choose a design
Searching for loyalty examples returns long lists of big-brand schemes. They rest on data and margins you do not have, so borrow the mechanism and ignore the numbers. Work from your own. For the illustrations here, take a small independent homeware and gift shop: $360,000 a year in sales, a 45% gross margin (what is left after the cost of the goods), an average basket of $36, and about 2,400 customers with an email address or phone number on file. Say 800 of them join in year one and spend about $180 each over the year, which is $144,000 of member spend.
The one formula worth memorising: a reward that costs r% of spend only pays for itself if members spend at least r ÷ (margin − r) more than they would have without it. At a 5% reward and a 45% margin that is 5 ÷ 40, or 12.5% more. The reason it exceeds 5% is that every $100 of spend brings $45 of margin, and giving back $5 leaves $40. To earn $45 again you need $112.50 of spend. Extra sales only deliver what is left after the reward.
A starting rule of mine, and only my judgement rather than an industry standard: keep the effective reward cost below about a tenth of your gross margin on member sales until a pilot shows what members do. At a 45% margin that is 4.5% of spend, which is the budget I give the AI below.
Points, stamps or tiers: what each costs a shop like this
| Design | How it works (illustrative) | Reward cost, share of member spend | Break-even lift in spend | Best when | Main risk |
|---|---|---|---|---|---|
| Points | 1 point per $1; 100 points = a $5 voucher | 4.0% if 80% of points are redeemed (5.0% if all are) | 9.8% (12.5% if all redeemed) | Basket sizes vary a lot, as with gifts and homeware | Unredeemed points build up as a liability; stacked promotions make the sums messy |
| Stamps | One stamp per visit of $15 or more; 8 stamps = a free item up to $12 | About 2.3%: the item costs $6.60 at cost price against $288 of spend | About 5.4% | Frequent visits of similar size | Rewards visits rather than spend; paper cards get lost or copied |
| Points with two tiers | Silver ($200 a year) earns 1.25 points per $1; Gold ($500) earns 1.5 | About 4.9% across my member mix | About 12.3% | Customer value varies widely and you want to recognise the best | Gold is 5% of members but takes about 31% of the reward budget |
The tier row deserves a closer look because it is the least intuitive. Split the 800 members into 560 Bronze spending $85 a year, 200 Silver spending $300 and 40 Gold spending $900, which totals $143,600. At 80% redemption the rewards cost $1,904 for Bronze, $3,000 for Silver and $2,160 for Gold: $7,064 in all, against $5,744 for flat points on the same spending. Tiers add $1,320 of cost before anyone changes behaviour, and the group that gets the most is the group that needed the least persuasion. At 4.9% the tiered design also breaks the 4.5% budget set above, which is a second reason to start with plain points. That does not make tiers wrong; it means a top tier needs a reason beyond a richer earn rate, such as early access to new stock or free gift wrapping, which costs far less than extra points.
My default for a first scheme at a shop like this is plain points with a 100-point minimum, no tiers, and a stamp card on paper if you want to test the idea before buying anything. Testing three rewards on a barbershop's own numbers follows the same logic for a service business, and building a loyalty programme with AI personalisation covers the later step of tailoring offers.
The prompt that produces rules with arithmetic you can check
Give the AI totals and averages, never a customer list. Tell it the goal, the budget and the constraints, and make it show every calculation on its own line. The sentence that earns its place is the last one: asking it to ask you for a missing number instead of guessing.
You are helping the owner of a small independent homeware and gift shop
design a customer loyalty programme.
Facts (use only these): annual sales $360,000; gross margin 45%; average
basket $36; about 800 customers expected to join in year one, spending
about $180 each per year; the till can award points per dollar; we can
send email and text.
Goal: more repeat visits from existing customers, not deeper discounting.
Constraints: total reward cost under 4.5% of member spend; rewards are
vouchers or free items under $12; part-time staff need rules they can
explain in one sentence.
Task: draft three designs (points, stamps, points with two tiers). For each
give: the rules in plain English; one worked customer example; the reward
cost as a percentage of member spend with every calculation on its own
line; the break-even increase in member spend at a 45% margin; and two ways
a customer or staff member could abuse it.
List your assumptions at the end. If you need a number I did not give you,
ask me instead of guessing.An AI-drafted rulebook and the five corrections it needed
Below is the kind of draft a general-purpose chat assistant produces for the points-with-tiers option. It is illustrative, written to show typical slips rather than quoted from a real run.
PROGRAMME: [Shop] Rewards (points with tiers)
Earn: 1 point per $1 on full-price items, before tax.
Redeem: 100 points = a $5 voucher, used in store or online.
Welcome bonus: 50 points ($5 value) on joining.
Birthday: double points on purchases in your birthday month.
Expiry: points never expire.
Tiers (12-month spend): Silver $200+ earns 1.25 points per $1.
Gold $500+ earns 1.5 points per $1 and free gift wrapping.
Cost: base reward rate 5%, about 5.5% with tiers and bonuses.
Break-even: members need to spend about 5.5% more for the
programme to pay for itself.
Abuse risks: staff scanning their own cards; points on returned items.| Line in the draft | What is wrong | My fix |
|---|---|---|
| "50 points ($5 value)" | At 100 points = $5, 50 points is $2.50. The draft mixed up points and dollars. | Relabel as $2.50, and award the bonus after the first purchase so fake sign-ups earn nothing. |
| "Points never expire" | Your liability only grows, and nothing nudges a return visit. | Expire points after 12 months without a purchase, with a message 30 days before and the rule in the terms. |
| "About 5.5% with tiers and bonuses" | No calculation shown. On my member mix tiers alone cost 4.9% at 80% redemption (6.1% if every point is redeemed), before the welcome and birthday extras. | Require line-by-line sums, and cost the extras separately. |
| "Break-even about 5.5%" | It treats the reward cost as the break-even lift. The right test is r ÷ (margin − r), about 12.3% here. | Put the formula in the prompt and confirm it in a spreadsheet. |
| "On full-price items" | Sensible, but gift cards are not excluded. A customer could earn points buying a gift card and earn again when it is spent. | Exclude gift cards and vouchers from earning, and cap points per transaction at 500. |
Note what the corrections have in common. None is a failure of writing; the draft reads well. They are failures of unit handling and of arithmetic that was never shown. That is why the prompt demands visible sums, and why a human still owns the numbers. A second pass in a fresh chat, asked only to audit the arithmetic line by line, catches some slips, but it can repeat the first chat's mistakes, so the spreadsheet is the referee.
Reward cost line by line, with redemption as the swing factor
| Step | Working | Result |
|---|---|---|
| Member spend in a year | 800 × $180 | $144,000 |
| Points issued | 1 per $1 | 144,000 points |
| Face value of those points | 144,000 ÷ 100 × $5 | $7,200 (5.0%) |
| Cost if 80% are redeemed | $7,200 × 0.8 | $5,760 (4.0%) |
| Gross margin on member spend | 45% × $144,000 | $64,800 |
| Rewards as a share of that margin | $5,760 ÷ $64,800 | 8.9% |
| Wallet-card tool, yearly | $25 × 12 | $300 |
| Spend needed to break even | ($64,800 + $300) ÷ (0.45 − 0.04) | $158,780, which is 10.3% more |
Redemption is the number you cannot know in advance, so test the design at 100% as well as at your expected figure, then replace both with data. At 100% redemption the cost is 5.0% of spend, which is over the 4.5% budget, so a pilot at that level means either accepting the risk for a quarter or trimming the reward (100 points = $4.50 brings the worst case to 4.5%). The break-even lift is 12.5% at 100% redemption (13.0% with the tool), 9.8% at 80% (10.3%) and 5.9% at 50% (6.4%). Break-even depends on the tool too. A software plan that costs $199 a month adds $2,388 a year, and the same sum then needs $163,873 of member spend, a 13.8% lift. Cheap tooling is not a detail when the margin is thin.
Is a 10% lift believable? Ask what your regulars do now. If they already visit every week, a scheme can hardly raise visits by a tenth. If they come twice a year, a reason to return in a quiet month may. Your till data answers this better than any template.
Cost per reward: a voucher, a free item or a perk
The same programme can pay out in different currencies, and they cost you different amounts per reward. Using the homeware shop's 45% margin, where a $12 item costs $6.60 to buy in:
| Reward | What the member sees | Cost to you per reward | Watch for |
|---|---|---|---|
| $5 voucher | "$5 off your next visit" | $5.00 of margin given up | Simple to explain; make sure it cannot be combined with sale prices |
| Free item up to $12 | "A free gift worth up to $12" | Up to $6.60 at cost price | Looks bigger than the voucher but costs more; cap the item's retail value |
| Free gift wrapping (a Gold perk) | "Wrapped free" | About $0.60 of paper and ribbon plus a few minutes of staff time (illustrative) | Cheap, but only valued by people who give gifts |
| Early access to new stock | "First look before it goes on the shelf" | Close to nothing | Needs stock you can hold back for a day |
| Double-points Saturday | "Twice the points today" | Doubles the earn rate for one day: 4% becomes about 8% of that day's member spend | Use on quiet days only, or you subsidise purchases that would have happened anyway |
When you compare "cost per reward", also compare what the member had to do to earn it. A $5 voucher takes $100 of spending in the points design. The free item takes eight visits, about $288. That is why the stamp card works out cheaper as a share of spend even though each reward costs more.
Five members, five outcomes: testing the rules on real profiles
Before launch, push a handful of customer types through the rules by hand. The test below uses the AI's tiered draft and, like the tier sums earlier, applies each member's tier rate to the whole year's spend. In a spreadsheet the points column is =ROUNDDOWN(spend*multiplier,0), rewards are =ROUNDDOWN(points/100,0) and the voucher value is rewards times five. Google Sheets AI prompts shows how to have an assistant build the sheet, provided you still check one row by hand.
| Member (illustrative) | Visits | Spend | Tier | Points | Rewards | Voucher value | Share of spend |
|---|---|---|---|---|---|---|---|
| Occasional shopper | 3 | $90 | Bronze, 1x | 90 | 0 | $0 | 0% |
| Regular | 10 | $360 | Silver, 1.25x | 450 | 4 | $20 | 5.6% |
| December gift buyer | 2 | $250 | Silver, 1.25x | 312 | 3 | $15 | 6.0% |
| Top customer | 30 | $1,500 | Gold, 1.5x | 2,250 | 22 | $110 | 7.3% |
| Bulk buyer, one order | 1 | $3,000 | Gold, 1.5x | 4,500 | 45 | $225 | 7.5% |
Three findings. First, without a welcome bonus the occasional shopper never reaches 100 points, so she gets nothing; the draft's 50-point bonus changes that, since 140 points buys her a $5 voucher on $90 of spending (5.6%), which is the bonus showing up in the cost. Cost it separately and pair it with a "you are 10 points away" message. Second, the tiers push everyone else above the 5% face value of plain points: the regular and the gift buyer land at 5.6% and 6.0%, and the top customer at 7.3%, which is the tier effect again. Third, the bulk buyer shows why a cap matters: limiting points to 500 per transaction cuts that single order's reward from $225 to $25 (0.8% of the order). It also exposes a loophole, because if status counts from the order itself, one large order buys Gold for a year. Count tier spend only from completed, unrefunded purchases, and decide whether status starts on the next visit.
The same sum for a low-margin staple: a pet-food counter
Now run the same sum for a shop selling a 15 kg bag of dog food at $48 on a 30% margin, so the bag costs $33.60. Suppose a loyal customer buys one every five weeks. The obvious reward, "buy ten bags, get the next one free", looks generous and sounds like every other scheme. Run the sum: ten bags is $480 of spending, the free bag costs $33.60 at cost price, so the reward is 7.0% of spend. Break-even is 0.07 ÷ (0.30 − 0.07) = 30.4%. Customers who already buy on a schedule cannot buy a third more dog food, so the scheme would be a straight giveaway.
Change the reward and the picture changes. A $5 voucher after ten bags costs 1.0% of spend and needs only a 3.6% lift. A free delivery, or a small treat bought in at cost price, costs even less. The lesson carries to any shop selling staples: the thinner the margin and the steadier the purchasing, the cheaper the reward must be, and the more the programme should aim at reasons to buy additional products, not at rewarding purchases that would have happened anyway.
Attack the rules before a customer does
Ask the AI to play two roles in turn: a customer trying to earn the most points for the least spend, and a part-time assistant who wants to give friends free rewards. Ask for ten abuses each, then keep the five you find plausible.
| Abuse | What happens | Patch |
|---|---|---|
| Buy, collect points, return | The refund goes back to the card but the points stay | Reverse points on any refund |
| Gift-card loop | Points on the purchase, and again on the spend | No points on gift cards or vouchers |
| Staff using their own cards | Friends' purchases credited to one account | Staff accounts excluded or capped; a weekly report of the ten biggest earners |
| Fake sign-ups for the welcome bonus | Dozens of throwaway emails | Bonus after a first purchase; one account per phone number |
| Stacking with sale prices or codes | Double discounts on one item | No points on discounted items or when a code is used |
| Voucher swapping | Vouchers passed between friends | Non-transferable; one voucher per transaction |
Write the terms in plain English: points have no cash value, how expiry works, how refunds affect points, and that you may change the scheme with notice. Collect marketing consent separately from the sign-up. Rules about points, vouchers and customer data differ between places, so have the final wording and your privacy notice checked by a solicitor or adviser.
Twelve weeks of messages: five drafts with character counts
Start with the sentence staff will say at the till, since the prompt asked for rules a part-time assistant could explain in one breath: "Every dollar earns a point, 100 points is $5 off, and points last a year from your last visit; shall I sign you up?" Then give the AI the rules, two lines of your shop's own social posts as a tone sample, and a limit of 160 characters for texts. Ask for three variants of each, pick one and edit it. Keep a margin because a long name or a points figure lengthens the text. Send marketing texts only to members who agreed to them, and keep the opt-out on every one.
| Moment | Channel | Draft (illustrative) | Characters |
|---|---|---|---|
| On joining | Text | Welcome to [shop] Rewards, [first name]! Earn 1 point per $1; 100 points = $5 off. Your first buy adds 50 bonus points. Reply STOP to opt out. | 142 |
| Close to a reward | Text | [first name], you're 18 points from a $5 reward. Pop in this week and it's yours. [shop] | 88 |
| 30 days before expiry | Text | Your [shop] points ([points]) expire on [date]. Use them in store or online before then. Reply STOP to opt out. | 111 |
| Tier upgrade | Text | Congratulations [first name]: you've reached Silver. You now earn 1.25 points per $1 on full-price buys. [shop] | 111 |
| Quiet for 90 days | Text | Hi [first name], your [points] points are still waiting at [shop]. New [range] arrived this week. Reply STOP to opt out. | 120 |
Only say "last chance" when it is true. If the expiry text goes out and the points are still valid a week later, members learn to ignore you. The quiet-member message fits into a wider win-back flow, which loyalty programme automation for shops sets out.
Where the programme lives: paper card, wallet card or till
- Paper stamp cards. Close to free and the quickest way to test the idea. You collect no data and cards get lost, so treat paper as a pilot.
- Wallet stamp cards. Loopy Loyalty's pricing page lists a Starter plan at $25 a month ($250 a year) for one card design, one location and three sub-users, with unlimited customers, a 15-day trial and Apple Wallet and Google Wallet support. Its staff app is called Stamper, so expect stamp-style cards and check before assuming it can run points.
- Points on the till. Square's loyalty page says customers can earn points per visit, per amount spent or per item or category, including when they pay cash, and describes VIP tiers, with a 30-day free trial. The page sends you to a plan comparison rather than quoting a price, so check that before budgeting. Its sample tier multipliers are 1.5x, 2x and 3x. On my numbers a 3x tier costs 12% of a member's spend (3 × 5% × 0.8) and needs a 36% lift to break even, so treat vendor examples as demonstrations, not recommendations.
- Online-store loyalty apps. Smile.io lists Growth at $199 a month for 2,500 orders a month and works with Shopify and BigCommerce. Built for online stores, it is probably more than a shop with a small website needs, and the sum above shows what its fee does to break-even.
A reasonable path is paper or wallet cards for one quarter, then your till's own scheme once the rules have stopped changing. Setting prices and promotions with AI matters here too, because a loyalty reward and a sale discount must not stack by accident.
Reading the first 90 days and deciding what to change
| Measure | How to work it out | What to do with it |
|---|---|---|
| Enrolment rate | Transactions linked to a member ÷ all transactions | Low after 8 weeks usually means staff are not asking; add a till prompt |
| Repeat visits | Members' visits in the 90 days after joining against the same members' visits in the 90 days before | Compare members with themselves, not with non-members, who are mostly less loyal to begin with |
| Spend per member | Monthly spend after joining against before | The number that decides break-even |
| Redemption rate | Rewards redeemed ÷ rewards earned | Replace your 80% assumption with the real figure |
| Reward cost | Voucher value redeemed ÷ member spend | Compare with your 4.5% budget |
| Points outstanding | Points issued minus redeemed minus expired | This is what you owe; track it monthly |
At day 90 compare the lift in spend per member with your break-even figure (10.3% in the example). My decision rule, offered as a judgement, not a standard: if the lift meets break-even, keep the scheme and consider a tier; if it reaches somewhere between half and all of break-even, adjust the earn rate or the redemption minimum and give it another quarter; if it falls below half, stop or redesign rather than keep paying for it. Weekly till reports make this quick, and a weekly retail KPI sheet built from till data shows how to set one up.
Questions shop owners ask before launching a scheme
How many customers does a shop need before a scheme is worth it?
There is no magic number. What matters is how often your regulars already visit and what margin you earn. Run the break-even sum with your own figures: if members would need to spend 10% or 15% more to cover the rewards and your regulars already visit weekly, the scheme probably will not pay. If they visit twice a year, it might.
Should points expire?
Expiry caps how much you owe members in unredeemed points and gives them a reason to return, but badly explained expiry annoys people. A common compromise is expiry after a period with no purchase, stated in the terms and reminded by message 30 days beforehand. Rules about vouchers and points differ by place, so have the wording checked by your adviser.
Can I test a scheme on paper cards first?
Yes, and it is a sensible way to start. The sums are identical, a stamp card costs almost nothing to print, and eight to twelve weeks of real visits tell you more than any forecast. Number the cards or use a distinctive stamp so they cannot be copied, then move to your till or a wallet card once the rules settle.
Does AI need my customer list to design the scheme?
No. Give it totals and averages: annual sales, margin, average basket, expected members and their average spend. Keep names, emails and phone numbers out of any chat tool. If you later want AI to analyse purchase history, export it without personal details first and check the privacy settings of the plan you use.
Further reads
- Restaurant Loyalty Programme: Simple Rewards and AI-Written Offers — The same reward maths applied to a restaurant's repeat diners
- Gym Referral Programme: Set Rewards and Test the Maths With AI — Price a referral reward the way this tutorial prices points
- Examples of AI in Retail: Eight Jobs a Small Shop Can Copy — Eight shop jobs AI can take on, each with an example
- What AI Costs a Small Retail Shop in Its First Year — What AI tools cost a small shop across its first year
- AI Software for Retail Stores: Five Checks Before You Buy — Five checks before you buy software that claims to use AI
- AI for Laundromats: Upsell Wash-Dry-Fold From Your Transaction Data — Export your payment data, group customers by load size, price a bag and its delivery, then text a wash-dry-fold offer and test it against a hold-back group.
- Comic Book Store Software: Pull Lists, Back Issues and AI Listings — Set up pull lists, list back issues from phone photos and send a weekly new-release email, with AI drafting the words and you checking the facts.
- How AI Can Help in Sales: A Gift Shop's Bundle and Follow-Up Plan — A gift shop's AI sales plan built on its own till data: three priced bundles, gift-note wording and four follow-up messages for one holiday month.
- Baby Store Marketing Ideas: Gift Lists and New-Parent Offers With AI — A gift-list page, a four-email welcome series and milestone offers for an independent baby shop, with prompts and a claims check for safety wording.
- AI Footfall Cameras for a Retail Store: What They Count and Privacy — What a footfall camera really counts, how to turn the numbers into a better rota, and the privacy checks and signage to settle before you install one.
- Can AI Help a Convenience Store Drive Sales? Where to Start — A convenience store owner's guide to AI sales tests: find high-margin lines in your till export, run one bundle test first, and read the result honestly.
- How AI Helps Marketing in a Small Supermarket: 5 Jobs and the Limits — Offers emails, social posts, local-event content, review replies and loyalty messages: what AI drafts well in a small supermarket and what you must still check.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Square Loyalty product page and help article; Loopy Loyalty pricing page; Smile.io pricing page (all checked October 2026). Shop figures are illustrative.