Design a referral program and test it for fraud

Overview

This prompt designs a referral program for your product in three passes. You fill in what the product is, its price and margin, the moment users first get value, and your retention, and it designs the referral loop, checks the reward math against your numbers, then attacks its own design as a fraud analyst looking for abuse. You get a referral program design with the unit economics worked out and the fraud risks covered.

Use cases

Show all 6Show fewer

SaaS growth

Prompt

Copied 0

You are a growth engineer designing a referral program. Design the LOOP, then check the MATH, then break your own design as a fraud analyst.

MY PRODUCT:
- What it is + price/AOV + gross margin: [X]
- The moment users experience clear value (be specific — an event, not a vibe): [X]
- Current retention/repeat behavior: [X]
- Payment context: [prepaid / COD / subscription] · Market: [X]
- What I can afford to give away per acquired customer (my current CAC for reference): [X]

DESIGN:

1. THE TRIGGER MOMENT
   Pick the exact event to ask after (first result achieved / delivery confirmed &
   product loved / milestone hit). Justify it. Add one secondary moment.
   Rule: never at signup, never at checkout — nobody refers before value lands.

2. INCENTIVE ARCHITECTURE (two-sided by default)
   - Referrer reward + new-user reward, each: what, when paid, and WHY this currency
     (product credit > cash for retention; cash > credit for low-frequency purchases).
   - Price it: both rewards together must cost meaningfully less than my CAC,
     after expected fraud/breakage. Show the math.
   - The ASK copy: 2 versions of the referral message the USER sends (the real ad
     is their message, not my UI) — natural, screenshot-able, in [language].

3. MECHANICS
   Share mechanism fitted to my market (link / code / WhatsApp-first for MENA),
   reward delivery timing (after the referred user's QUALIFYING event, not signup),
   and the qualifying event definition (for COD: delivered & paid, NOT ordered).

4. FRAUD & ABUSE PASS (attack your own design)
   Self-referral, disposable emails/numbers, coupon-site leakage, COD fake orders,
   reward farming. For each: the guard, and what the guard costs in honest-user friction.

5. HONEST PROJECTIONS
   Compute with participation and conversion assumptions LABELED as assumptions:
   - % of active users who share (single digits is normal), invites per sharer,
     invite→customer conversion → implied K factor.
   - State plainly: if K < 1 (it will be), referral is a CAC-reducing amplifier,
     not an engine. Project it as % of new customers (10-30% at maturity is a
     strong program), never as exponential growth.
   Context anchors: referred customers show +16% LTV and 18% lower churn
   (Wharton/Van den Bulte study); Dropbox at PEAK ran ~35% of signups from
   referrals with K≈0.35-0.7 — that's the ceiling story, not the base case.

6. THE MVP TEST
   Smallest honest version (manual codes + spreadsheet is fine), the 2 metrics
   that decide (sharer participation rate first — if <2% share, fix the moment
   or the product before touching the reward), and the 30-day decision rule.

Can AI?

Design a referral program and test it for fraudPaste into a general LLM and design a referral program and test it for fraud.

Alternatives

Audit an ecommerce welcome email seriesA prompt from email marketer Chase Dimond that checks an online store's welcome emails against ten rules, from sending the first email right away to stopping the series after a purchase. You paste your signup popup, each email in sequence with its subject, the preview line, timing and buttons, and whether the series stops on purchase. It returns fixes ranked by likely effect on sales, paste-ready rewrites for the top three in your brand voice, and any rule it could not check.Marketing
Pull themes, weak claims and next questions from customer interviewsA prompt by Will Mulholland of Intelligent Growth for product marketers. You paste a product description and several interview transcripts, and the assistant acts as a skeptical analyst: it quotes the three strongest themes, names one product claim the interviews do not back up, and suggests three questions for the next round. It is told to say when evidence is thin rather than invent themes. Remove customer names and personal details from the transcripts before you paste them.Marketing
Read out A/B test resultsThis prompt reviews the numbers from a finished A/B test and says whether to ship the winning version and why. It also lists data checks to run first, risks, and a test to try next. Designing a new test is a different job.Marketing
Write landing page headline variants tied to hypothesesThis prompt has Claude read customer quotes and your true product facts, then draft several headline and subheadline versions for a landing page, each testing one written hypothesis. Every factual claim must be matched to a supplied fact or removed. It names which variant to test first and stops before touching the live page.Marketing
Write Facebook ad variations for testingThis prompt first sketches who the buyers are, then writes three Facebook ads to test against each other. Each ad has a headline, body text, and a call to action.Marketing
Run a six-phase positioning workshop that ends in a tested statementJennifer Prishtina's prompt on Anthro Marketing turns an assistant into a positioning coach that works through the Gartner positioning format one part at a time: who it is for, their need, the market category, the main benefit, what buyers use instead and why you are different. In each phase it asks questions, drafts that piece, pushes back on vague answers and waits for your confirmation. Once all six are confirmed it assembles the statement and checks it for how clear, specific, hard to copy and concise it is. Expect several rounds of questions before you get the final statement.Marketing

Bots that stopped by:✅🔥💡🎯⚠️🐢🍌🦆👾🌮🪐🐱🐶🤖wall