loyalty_program_design v1.0.0

name: loyalty_program_design description: "Design or fix a loyalty program with unit-economics modeling: payback rate, redemption and breakage, tier thresholds from real spend data, earn and burn rules, liability under ASC 606 / IFRS 15, and fraud controls. Use when someone asks to create a points, tier, cashback, stamp, or membership program, price its rewards, set tier thresholds, stop it losing money, or audit an existing program."

Loyalty Program Design

Purpose

Build a program that changes behaviour at a cost the margin can carry. The core job is arithmetic first, perks second: model what points will cost when they are redeemed, set tiers from the customer's own spend distribution, tune earn and burn together, book the liability honestly, and close the fraud holes before launch.

Workflow

1. Inspect before designing

Collect, or state as missing:

• Eligible sales, order count, average order value, gross margin by category.
• Spend distribution for the last 12 months: median, 75th, 90th, 97th percentile per customer.
• Repeat rate and purchase frequency by cohort.
• Return rate and return window.
• For an existing program: points outstanding, age of outstanding points, redemption by cohort, current tier mix, manual adjustments volume.

Do not propose thresholds, earn rates, or rewards before this pass. A redesign that ignores what members already hold will break trust and the liability model at the same time.

Readiness check: if repeat purchase is rare, margin is thin, or the base is tiny, say so plainly. A simple thank you, referral credit, or stamp card may beat a full points program. Name the simpler option and why.

2. Pick the model that fits the behaviour

• Points: frequent, lower value purchases. Members need a first reward within a few purchases.
• Tiers: wide spend spread, status matters, top customers are worth far more than the median.
• Cashback or credit: price is the main reason people buy, simplicity matters most.
• Paid membership: high frequency, a benefit with obvious value such as free delivery, fee covers its cost at realistic use.
• Stamp or visit count: local, in person, visit frequency is the behaviour to move.
• Coalition or partner: your own visit frequency is too low to keep points useful, partners supply earn and burn occasions.

Pick one primary model. Hybrids are allowed only when each part moves a different behaviour.

3. Model the unit economics before naming any perk

Run the calculator for at least three redemption scenarios. Never present a single point estimate as the cost.

bash
python3 ~/workspace/skills/loyalty-program-design/bin/loyalty_calc.py \
  --sales 1000000 --payback 0.05 --redemption 0.75 \
  --cost-factor 0.6 --incremental-margin 18000 --compare

Verified output for that call: face issued 50000, expected redeemed face 37500, expected breakage face 12500, expected reward cost 22500 (2.25 percent of sales), net against 18000 incremental margin is minus 4500. That program loses money as drawn. Fix it by lowering payback, choosing rewards with a lower cost factor, or proving a larger incremental margin. Do not fix it by hoping breakage will be higher.

Definitions to state in every model:

• Payback rate: face value issued divided by eligible spend. 100 points worth 5 dollars, earned at 100 points per 100 dollars spent, is a 5 percent payback.
• Cost factor: real cost per 1.00 of redeemed face. Use 1.0 for cash like rewards. Use the COGS share for own product rewards: at 40 percent gross margin, a 10 dollar retail reward costs about 6 dollars.
• Redemption rate: share of an issued cohort redeemed at maturity. Breakage is 1 minus redemption, for that mature cohort only.
• Incremental margin: margin from purchases the program actually causes, not margin from all member purchases. Members self select, so member versus non member gaps overstate the effect unless you use a holdout, a staggered rollout, or a pre and post comparison with a matched group. State the method or lower confidence.

Rules:

• Reward cost must sit below proven or conservatively estimated incremental margin, with room for platform, support, and fraud losses.
• Expiry, redemption threshold, and how long points sit unspent move cost as much as the earn rate does. Model them, do not treat them as admin details.
• Prefer rewards with high perceived value and low marginal cost (own services, spare capacity, early access) over discounts on products members would buy at full price anyway. Flag cannibalisation when a reward mainly subsidises existing purchases.

4. Design earn and burn as one system

• Clean currency math: pick a ratio a member can do in their head, for example 100 points equals 1 dollar. Avoid odd conversions that hide value.
• First reward fast: set the lowest redemption threshold so a median member reaches it in 2 to 3 normal purchases. A first reward that takes a year teaches members to ignore points.
• Earn on net spend after returns, not gross. Exclude gift card purchases, taxes, and delivery fees unless there is a stated reason to include them.
• Pending period: points move to available only after the return window closes.
• Burn options at several sizes: one small and frequent, one medium, one aspirational. If nothing is worth redeeming, issuance is just growing liability.
• Partnership burn counts: partner rewards widen burn options but need a settlement rate agreed in writing before launch.

5. Set tiers from the spend distribution

• Use the percentiles from step 1. A workable starting shape: middle tier near the 75th percentile of annual spend, top tier near the 90th to 95th. Test the shape against real counts before committing.
• Two or three tiers for most programs. Add a fourth only when the top group is large and distinct enough to behave differently.
• Keep tier currency separate from spendable points, at a visibly different ratio (for example 1 qualifying unit per dollar beside 10 spendable points per dollar). Annual tier resets must never wipe a spendable balance.
• Each tier must differ in a way that changes behaviour: earn multiplier, a real benefit, or access. A tier that changes nothing is overhead and liability with no return.
• State the qualification window (rolling 12 months is common), the downgrade rule, and any soft landing in the member facing terms before launch.
• Check the near miss group: members within 10 to 20 percent of the next threshold are the group most likely to add a purchase. Size that group. If it is near zero, the threshold is wrong.

6. Book the liability and handle breakage honestly

Summary only, finance owns the final treatment. See references/economics-and-controls.md.

• Under ASC 606 and IFRS 15, points that give a material right are a separate performance obligation. Part of the sale price is allocated to the points by standalone selling price and held as a contract liability until redemption, expiry, or breakage recognition.
• Breakage is recognised in proportion to the pattern of rights exercised when the estimate is reliable and a significant reversal is not probable. It is not recognised as instant profit at issuance.
• Re estimate redemption and standalone selling price on a set cadence. Changes create true up adjustments, so program changes need finance sign off before they are announced to members.
• Never design for breakage as the profit plan. A program that only works if members fail to redeem will fail in public first.

7. Build fraud controls into the design

Cover all four abuse types, detail in references/economics-and-controls.md:

• Accrual abuse: double posting, points on returned goods, self referral, bonus farming with fake accounts.
• Redemption abuse: stolen balances, pooling across fake accounts, double spend across channels.
• Account takeover: credential stuffing against points rich accounts, contact detail change then redeem.
• Insider abuse: manual balance changes, overrides, fake awards.

Minimum controls: pending points until returns clear, automatic reversal on return or cancellation including tier credit, earn caps per transaction and per day, redemption holds and manual review above a set value, extra verification for redemption after a contact or device change, role based access and an audit log for every manual adjustment, velocity and anomaly alerts, transfer limits or no transfers by default.

8. Write the member terms before launch

Terms must state in plain language: what earns, what does not, when points become available, expiry rule, tier window and downgrade rule, transfer rule, what happens on returns, how the program can change and how much notice members get, and how to dispute a balance. Give real notice before a devaluation and let members redeem at the old rate during the notice window where the law or policy requires it. Flag CARD Act gift card distinctions and state unclaimed property treatment for legal review. This skill is not legal advice.

9. Measure after launch

Track weekly for the first quarter, then monthly: enrolment rate, active member rate, redemption rate by cohort, cost as percent of sales versus model, incremental margin versus holdout or matched group, tier movement and near miss conversion, breakage estimate versus actual, fraud flags and confirmed losses, support contacts per 1000 members. Kill or fix a reward nobody redeems in two cycles. Report cost honestly even when the news is bad.

Output Contract

Deliver a design pack with:

1. Inputs used, and inputs missing, with the source period for spend data.
2. Chosen model and why, plus the simpler option rejected and why.
3. Economics table: at least three redemption scenarios from the calculator, with payback, cost factor, reward cost, cost as percent of sales, and the incremental margin method.
4. Earn and burn spec: currency ratio, earn rules and exclusions, pending period, redemption options and thresholds.
5. Tier table (if tiered): thresholds tied to named percentiles, qualification window, downgrade rule, per tier differences, near miss group size.
6. Liability note: contract liability approach, breakage method, re estimation cadence, finance sign off status.
7. Fraud control list mapped to the four abuse types.
8. Member terms checklist with expiry, changes, and dispute process.
9. Measurement plan with the metrics above and a review date.

Operating Rules

1. No invented numbers. Thresholds come from the client's spend data, costs from the calculator with stated inputs, or are labelled as illustrative examples.
2. Never ship a design whose only path to profit is high breakage.
3. Never tune earn without checking burn, or burn without checking liability.
4. No tier, reward, or rule that changes no behaviour.
5. State the causal method behind any lift claim, or downgrade the claim to correlation.
6. Program changes that alter point value, expiry, or tier status need finance and legal review plus member notice. Flag them, do not present them as settled.
7. Keep the structure explainable in two sentences by a member. If it needs a diagram to understand, simplify it.