Referral Program Designer

---
name: referral-program-designer
description: Use when designing or revising a referral / advocacy program. Triggers on "referral program", "refer a friend", "advocacy program", "viral loop".
---

# Referral Program Designer

The "refer a friend, both get $20" pattern works for some products and tanks others. A good design respects: (1) who's actually motivated to refer, (2) what they're willing to refer for, (3) what blocks them today, and (4) what protects you from the fraud + low-quality referrals that ruin most programs by month 3.

## Required inputs

1. **Product type** — consumer / prosumer / SMB / mid-market / enterprise
2. **Current acquisition mix** — paid, organic, referrals (current %), other
3. **Unit economics** — CAC, LTV, payback period
4. **Existing organic word-of-mouth** — if you have no NPS / no organic referrals, fix that first before incentivizing
5. **Existing program (if any)** — structure + data
6. **The moment of truth** — when in the lifecycle a user is most likely to refer (post-onboarding, post-value, mid-renewal, etc.)

## Design dimensions

### 1. Who refers
- All users (broad, low conversion)
- Power users / NPS promoters (narrow, high conversion)
- A specific segment (e.g. agencies referring clients, teams referring other teams)

Narrow is usually better. Designing for everyone means designing for no one.

### 2. The incentive
**For the referrer**:
- Cash / credit (high-motivation, fraud-prone, low-retention)
- Product upgrade (medium-motivation, lower-fraud, high-retention)
- Status / recognition (low-motivation but for the right users, high-value)
- Charity donation (moral incentive, polarizing)
- Stacking rewards (every referral compounds)

**For the referred**:
- Discount (transactional, low-quality referrals)
- Extended trial (lower-pressure, higher-quality)
- Pre-loaded value (e.g. credits, content access)

**The mistake to avoid**: equal reward for both = good optics, bad incentive alignment. Skew toward the referrer if you want activation, toward the referee if you want quality.

### 3. The trigger
- Referrer-initiated: they share a link
- System-initiated: post-purchase email, in-app prompt at a moment of delight
- Hybrid: in-app prompt + shareable link

In-app prompts at peak-NPS moments outperform email reminders 3–5x.

### 4. The reward timing
- Immediate on signup: fraud-prone
- After referee's first valued action (active week, paid month): better quality
- After payment: highest quality, longest delay (motivation lags)

Balance: reward big enough on first signup that the referrer cares, with a multiplier when the referee converts.

### 5. Anti-fraud rules
- One reward per unique referee (deduplicated by email + payment method)
- No self-referrals (block obvious patterns)
- Reward cap per referrer per month
- Referee must complete a non-trivial action before reward triggers
- Manual review of top-N referrers per week

### 6. Measurement
- Top-line: % of new signups from referrals
- Quality: LTV / retention / activation rate of referred vs. non-referred
- Cost: CAC of referred users (including reward cost) vs. other channels
- Network effect: how many referrers actually refer (vs. one-time program adoption)

### 7. Kill criteria
The program gets paused / sunset if:
- Referred-user LTV < 0.7x non-referred LTV for 60 days (quality erosion)
- Fraud > 5% of rewards
- Cost-per-referred-user > paid channel CAC
- Program adoption < 5% of eligible users after 90 days

## Output

```
## Program design
- Who refers: [segment]
- Who gets reward: [structure]
- Reward type + amount: [for referrer / for referee]
- Trigger moment: [where in the flow]
- Reward timing: [when paid out]

## Anti-fraud rules
[Bulleted list, each enforceable in product/data]

## Launch plan
- Pilot scope: [N users, M weeks]
- Measurement plan: [what we'll watch, when we'll decide]
- Kill criteria: [the specific thresholds]

## What this program will NOT solve
[If you're hoping referrals fix poor product-market fit, they won't. Be clear about scope.]
```

## Anti-patterns

- Launching to all users at once — fraud and noise overwhelm signal
- Equal cash rewards both sides — looks fair, optimizes for nothing
- "Refer 5 friends" tiered rewards — users do 0 or 1, almost never 5
- Reward at signup (no quality bar) — fraud farm
- No kill criteria — program runs forever even when it stops working

## Tone

- Mechanism-honest. The incentive shapes behavior. Acknowledge what you're optimizing for.
- Realistic. Most referral programs underperform expectations. Plan for that.
- Quality-aware. Volume metrics are vanity; LTV-quality metrics are signal.