How to Build a Loyalty Points and Referral Rewards System: Point Rules, Anti-Abuse and Accounting Recognition

Three Years of Punch Cards, 11% Redemption

A six-store bakery chain ran a paper punch card for three years: buy ten, get one free. Plenty of stamps got collected, but only 11% of cards ever came back for redemption, and nobody could answer which customers only show up during promotions, or what a referred customer is actually worth. Ninety days after moving the programme online, the numbers meant something for the first time: transactions tied to a member had a 23% higher average ticket than unidentified ones, referral codes converted first-time buyers at 2.4x the rate of paid traffic, and 4.7% of those referrals turned out to be people referring themselves.

Who Should Build This, and Who Should Wait

Good fit

• Average ticket NT$200–3,000 with a natural repeat cycle inside 60 days: retail, food and beverage, beauty, pet, fitness.
• More than 3,000 identifiable members already (mobile number or LINE UID), enough to segment.
• Gross margin above 35%, so you can spend 3–5% on rewards without eating the business.
• Two or more channels (in-store POS, website, LINE) that need one member record and one point balance.
• Products with genuine repeat demand: consumables, refills, class packs, seasonal returns.

Wait, for now

• Once-in-a-lifetime services such as wedding photography, home renovation or funeral services. Referral rewards work here; accumulating points does not.
• Fewer than 500 monthly active members. The sample is too small and your segments will be noise.
• Gross margin under 20%, such as low-margin wholesale. Points come straight out of net profit.
• In-store POS is a standalone terminal with no API, so offline spend cannot be written back. Customers will only notice that store purchases earn nothing.
• Nobody owns the monthly job of reading the numbers, rotating rewards and adjusting rules. An unmanaged points programme is just a liability that compounds.

Four Routes and What They Cost

The pragmatic sequence is to validate with SaaS or the LINE punch card first. If member sign-up on transactions cannot reach 25% in three months, do not rush into custom work. If it can, then pull the points engine in-house.

Four Phases From Rules to Launch

• Phase 1: Points economics and data audit (5–8 working days). Work out how much you issue, how much gets redeemed and what it costs, before anyone draws a screen. Deliverables: a points economics model (Google Sheets, with three scenarios for issuance rate, redemption rate and liability ceiling), a tier table, a duplicate-rate report on existing member data. Tools: Google Sheets, Notion.
• Phase 2: Flow and interface design (8–12 days). Map sign-up, earning, tier changes, redemption, referral and expiry as a state diagram, and define for each state who can change it and what record that leaves. Deliverables: high-fidelity Figma screens (LINE LIFF member card, website member centre, in-store staff view), notification copy table, draft member terms. Tools: Figma, FigJam.
• Phase 3: Points engine and integrations (20–35 days). Build the point ledger (every change is a row; never edit a balance field directly), expiry batch jobs, tier rules, referral codes and attribution, and back-office audit. Deliverables: a system running real transactions, API spec as a Postman collection. Tools: Laravel, Redis (distributed lock against double deduction), ECPay or TapPay, LINE Login/LIFF/Messaging API, cloud POS webhooks.
• Phase 4: Single-store pilot and launch (10–15 days). Run one store for a full 30 days of real transactions and reconcile the point ledger against POS sales daily. Deliverables: a one-page staff SOP, a Metabase dashboard, Sentry and UptimeRobot monitoring.

Real Costs: Three Tiers and the Hidden Items

• Entry, NT$180,000–320,000. LINE LIFF member card, mobile binding, a single point balance, reward catalogue management, back office. Excludes POS integration and referrals.
• Mid, NT$400,000–750,000. Adds tiers, expiry batch jobs, referral codes and attribution, two-way integration with one cloud POS, online payments, GA4 and dashboards.
• Advanced, NT$900,000 and up. Shared point pool across brands, stored value, cross-channel redemption, two-way sync with ERP and CRM, tier-driven marketing automation.

The line items most people forget

• Points liability: the most expensive and the most commonly missed. Unredeemed points are a liability, typically accrued at 60–80% of face value based on historical redemption. At NT$30M annual revenue and a 3% issuance rate, that is roughly NT$900,000 a year sitting in your P&L. It is not a marketing budget.
• Payments: per the ECPay fee schedule, standard merchants pay 2.75% on domestic credit cards plus NT$1 per order, with 5% VAT on top. TapPay negotiates rates by scale and risk.
• LINE broadcasts: the mid-tier plan is NT$800/month including 3,000 messages, with overage billed on top. Expiry reminders and birthday rewards are the biggest volume driver.
• SMS OTP: NT$0.7–1.2 per message, used at sign-up and whenever a member changes their phone number.
• POS integration: NT$30,000–80,000 per system, plus a middleware layer for older terminals.
• Hosting and Redis: NT$1,500–5,000 per month. Annual maintenance: 15–20% of build cost.

What Clients Imagine vs What Happens

• Imagined: point rules can be tweaked any time. Actual: one change means converting historical balances, publishing a notice, retraining support and absorbing complaints, a small project starting at NT$30,000. Argue the rules out before launch.
• Imagined: POS integration is just an API call. Actual: plenty of stores still run standalone POS with no cloud API. Switching to cloud POS or writing middleware routinely consumes 30% of total project hours.
• Imagined: launch a referral programme and people will share it. Actual: without a designed share entry point and trigger moment, share rates sit below 3%. What works is putting the share button on the page right after a satisfying transaction, not deep inside the member centre.
• Imagined: points are free marketing. Actual: points are a liability, not a discount. You accrue on issuance and release on redemption; with no expiry, that liability has no ceiling.
• Imagined: sign-up rates will grow by themselves. Actual: without staff scripts and a KPI, transaction sign-up stalls around 15%. One sentence from a cashier beats three pop-ups in an app.

Six Traps and How to Avoid Them

• Point value never defined. Fix: state plainly in the member terms that one point offsets NT$1, cannot be cashed out and cannot be transferred, so it is not treated as prepayment or a gift certificate.
• Referral rewards built as multi-level commissions. Fix: keep it to one level, and base the reward on the referred person's actual spend rather than headcount. Multi-level commission structures whose main compensation comes from recruiting rather than selling can fall under Taiwan's Multi-Level Marketing Supervision Act and its filing requirements.
• Self-referral farming. Fix: block when any two of device fingerprint, card BIN, shipping address and IP match, and only credit the reward after the referred person's first purchase clears a seven-day cooling-off window.
• Points that never expire. Fix: set a 12–24 month validity, push a reminder 30 days before expiry, and deduct on a first-in-first-out basis. Without expiry the liability only grows in one direction.
• No explicit consent record for member data. Fix: put an independent, non-pre-ticked consent checkbox in the sign-up flow and store the timestamp and terms version, in line with the notification and specified-purpose requirements of Taiwan's Personal Data Protection Act.
• No audit trail on point movements. Fix: use a ledger rather than a balance field, recording operator, reason code and before and after balance on every row, with two-person approval on any manual credit. Internal point theft happens almost exclusively in systems with no log.

Success Metrics and the 90-Day Roadmap

• Day 30: sign-up and data hygiene. Member-identified transaction rate at or above 35%, sign-up completion at or above 60%, zero reconciliation variance between the point ledger and POS sales, zero unresolved alerts. Do not look at revenue yet.
• Day 60: behaviour and referrals. Average-ticket gap between members and non-members, redemption rate (healthy band is 20–35%; too low means the rewards are unappealing, too high means you are issuing too generously), referral share rate at or above 8%, fraud block rate and false-positive rate.
• Day 90: money. Lift in 90-day repeat-purchase rate against baseline (a reasonable target is 10–15%), acquisition cost of referred customers versus paid CAC, points liability as a share of revenue (keep it under 4%), and retire the three least-redeemed rewards. Only once those three numbers exist should you discuss issuing more.

Decision Checklist

• ☐ Do you have at least 500 monthly active members?
• ☐ Do you have 3,000 or more identifiable members (mobile or LINE UID)?
• ☐ Does the product have a natural repeat cycle inside 60 days?
• ☐ Is gross margin above 35%?
• ☐ Can you calculate your current 90-day repeat-purchase rate as a baseline?
• ☐ Does your in-store POS expose a cloud API for writing spend back?
• ☐ Have you decided what one point is worth?
• ☐ Do you accept that points must carry an expiry?
• ☐ Is someone responsible each month for rotating rewards and tuning rules?
• ☐ Have you budgeted reward cost at 3–5% of revenue?
• ☐ Does your accountant know a points liability must be accrued?
• ☐ Is the referral programme confirmed as single-level and spend-based?
• ☐ Do cashiers have a script and a KPI for signing members up?
• ☐ Have the member terms and consent flow been reviewed by someone qualified?

Nine or more ticks and the mid tier usually pays for itself. Fewer than six and you should first establish a repeat-purchase baseline and get POS data flowing before revisiting this table.

FAQ

What should one point be worth, and what issuance rate is safe?

The common Taiwanese retail design is one point per NT$100 spent, redeemable at NT$1, which is a 1% issuance rate. Food and beverage and beauty, with higher repeat frequency, usually land at 2–5%. Do not copy competitors; work backwards from margin. Cap affordable reward cost at 8–12% of gross margin, then divide by expected redemption rate (most brands land at 20–35%) to get your issuance rate. A low redemption rate does not mean you saved money; it means the rewards are unattractive while the liability still sits on your books.

Do points really have to expire?

From a finance standpoint, effectively yes. Without expiry the liability only accumulates, and five years later you are carrying a large cost that can be claimed at any moment. Twelve to twenty-four months is the accepted range in practice, and three things matter: write the expiry rule into the member terms, send a proactive reminder 30 days out, and deduct first-in-first-out. If your points are actually a top-up balance that can be cashed out, that is closer to prepayment or a gift certificate, with different accounting and compliance treatment, so talk to an accountant before you design it.

Could a referral programme be treated as multi-level marketing?

Usually not, if two conditions hold: it stays single-level (A refers B and earns; when B refers C, A earns nothing), and the reward is calculated on the referred person's actual spend rather than on how many people were recruited. Once you introduce multi-tier commissions, or the main compensation comes from recruiting others rather than selling products, you can fall within the Multi-Level Marketing Supervision Act and its pre-filing obligation. Drawing that line at design time is far cheaper than fixing it afterwards.

Can we do this if the store POS has no API?

Yes, but accept a degraded experience. Three common approaches: staff manually enter the member phone number and amount on a tablet, which is fastest but error-prone; middleware reads a scheduled POS export and credits points on a T+1 basis, so the customer sees nothing at the counter; or you switch to a cloud POS such as iCHEF or Dudoo, which solves it in one move but means new hardware and retraining. Our advice is to run manual entry for three months and let the sign-up rate decide whether the POS spend is justified.

How long before repeat-purchase rates move?

Data in 30 days, behaviour in 60, finance in 90. In month one, expect only that transactions are being tied to members. Month two is when the average-ticket gap between members and non-members becomes visible. The 90-day repeat rate needs until the end of month three to have enough sample. If nothing has moved after three months, the problem is usually not the point rules but the product having no reason to be bought again, and issuing more points at that stage only accelerates the loss.

Next Step

ScriptWalker's Loyalty Points and Referral Programme build starts at NT$180,000 and includes the points economics model, draft member terms and consent flow, and an MVP that runs real transactions. If you already hand out paper punch cards or use the LINE punch card, send us three months of transaction detail (de-identified is fine) and we will return a free points economics review estimating your sensible issuance rate, expected redemption rate and likely liability range.

• Email: [email protected]
• Phone: 0916-224-047
• LINE: @ufv9089p