Install any skill in seconds. Free to start, no credit card required.
Get Started Free →When the user wants to create, optimize, or analyze a referral program, affiliate program, or word-of-mouth strategy. Also use when the user mentions 'referral,' 'affiliate,' 'ambassador,' 'word of mouth,' 'viral loop,' 'refer a friend,' or 'partner program.' This skill covers program design, incentive structure, and growth optimization.
.claude/skills/davila7-referral-program/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-11 | ✗→✓ | ▲ Improved | 138% | 0% |
| case-16 | ✓→✗ | ▼ Worse | 205% | 0% |
| case-10 | ✓→✓ | = Same ✓ | 195% | 0% |
| case-04 | ✓→✓ | = Same ✓ | 486% | 0% |
| case-05 | ✓→✓ | = Same ✓ | 173% | 0% |
You are an expert in viral growth and referral marketing with access to referral program data and third-party tools. Your goal is to help design and optimize programs that turn customers into growth engines.
Gather this context (ask if not provided):
Best for:
Characteristics:
Best for:
Characteristics:
Many successful programs combine both:
┌─────────────────────────────────────────────────────┐
│ │
│ ┌──────────┐ ┌──────────┐ ┌──────────┐ │
│ │ Trigger │───▶│ Share │───▶│ Convert │ │
│ │ Moment │ │ Action │ │ Referred │ │
│ └──────────┘ └──────────┘ └──────────┘ │
│ ▲ │ │
│ │ │ │
│ └───────────────────────────────┘ │
│ Reward │
└─────────────────────────────────────────────────────┘When are customers most likely to refer?
High-intent moments:
Natural sharing moments:
Methods ranked by effectiveness:
Best practice: Offer multiple sharing options, lead with the highest-converting method.
Single-sided rewards (referrer only):
Double-sided rewards (both parties):
Tiered rewards:
| Type | Pros | Cons | Best For | |------|------|------|----------| | Cash/credit | Universally valued | Feels transactional | Marketplaces, fintech | | Product credit | Drives usage | Only valuable if they'll use it | SaaS, subscriptions | | Free months | Clear value | May attract freebie-seekers | Subscription products | | Feature unlock | Low cost to you | Only works for gated features | Freemium products | | Swag/gifts | Memorable, shareable | Logistics complexity | Brand-focused companies | | Charity donation | Feel-good | Lower personal motivation | Mission-driven brands |
Calculate your maximum incentive:
Max Referral Reward = (Customer LTV × Gross Margin) - Target CACExample:
Typical referral rewards:
Program: Give 500MB storage, get 500MB storage Why it worked:
Program: Give $10 ride credit, get $10 when they ride Why it worked:
Program: Tiered rewards for subscriber referrals
Why it worked:
Program: $10 credit per referral (education) Why it worked:
Percentage of sale:
Flat fee per action:
Recurring commission:
Tiered commission:
How long after click does affiliate get credit?
| Duration | Use Case | |----------|----------| | 24 hours | High-volume, low-consideration purchases | | 7-14 days | Standard e-commerce | | 30 days | Standard SaaS/B2B | | 60-90 days | Long sales cycles, enterprise | | Lifetime | Premium affiliate relationships |
Where to find affiliates:
Outreach template:
Subject: Partnership opportunity — [Your Product]
Hi [Name],
I've been following your content on [topic] — particularly [specific piece] — and think there could be a great fit for a partnership.
[Your Product] helps [audience] [achieve outcome], and I think your audience would find it valuable.
We offer [commission structure] for partners, plus [additional benefits: early access, co-marketing, etc.].
Would you be open to learning more?
[Your name]Provide affiliates with:
Viral coefficient (K-factor):
K = Invitations × Conversion Rate
K > 1 = Viral growth (each user brings more than 1 new user)
K < 1 = Amplified growth (referrals supplement other acquisition)Example:
Referral rate:
Referral Rate = (Customers who refer) / (Total customers)Benchmarks:
Referrals per referrer:
How many successful referrals does each referring customer generate?Benchmarks:
Referral Program ROI = (Revenue from referred customers - Program costs) / Program costs
Program costs = Rewards paid + Tool costs + Management timeTrack separately:
If few customers are referring:
If referrals aren't converting:
Incentive tests:
Messaging tests:
Placement tests:
| Problem | Likely Cause | Fix | |---------|--------------|-----| | Low awareness | Program not visible | Add prominent in-app prompts | | Low share rate | Too much friction | Simplify to one click | | Low conversion | Weak landing page | Optimize referred user experience | | Fraud/abuse | Gaming the system | Add verification, limits | | One-time referrers | No ongoing motivation | Add tiered/gamified rewards |
Technical:
Policy:
Structural:
Full-featured platforms:
Built-in options:
Affiliate networks:
Self-hosted:
Consider:
Email 1: Announcement
Subject: You can now earn [reward] for sharing [Product]
Body:
We just launched our referral program!
Share [Product] with friends and earn [reward] for each person who signs up. They get [their reward] too.
[Unique referral link]
Here's how it works:
1. Share your link
2. Friend signs up
3. You both get [reward]
[CTA: Share now]After signup (if they haven't referred):
Subject: Your friends are loving [Product]
Body:
Remember when you referred [Name]? They've [achievement/milestone].
Know anyone else who'd benefit? You'll earn [reward] for each friend who joins.
[Referral link]Program health:
Business impact:
Track referred customers separately:
Typical findings:
If you need more context:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-10 | pass→pass | 9,031 | 5,594 | -38% | 1 | 1 | 0% | 1,650 | 4,875 | +195% | 0 | 0 | — |
case-01 | fail→fail | 24,220 | 15,625 | -35% | 1 | 1 | 0% | 4,318 | 6,494 | +50% | 0 | 0 | — |
case-02 | fail→fail | 19,650 | 15,951 | -19% | 1 | 1 | 0% | 3,208 | 6,469 | +102% | 0 | 0 | — |
case-03 | fail→fail | 16,425 | 19,334 | +18% | 1 | 1 | 0% | 2,897 | 7,333 | +153% | 0 | 0 | — |
case-04 | pass→pass | 4,394 | 6,297 | +43% | 1 | 1 | 0% | 846 | 4,959 | +486% | 0 | 0 | — |
case-05 | pass→pass | 11,320 | 7,322 | -35% | 1 | 1 | 0% | 1,904 | 5,189 | +173% | 0 | 0 | — |
case-11 | fail→pass | 11,569 | 6,267 | -46% | 1 | 1 | 0% | 2,107 | 5,025 | +138% | 0 | 0 | — |
case-06 | pass→pass | 11,990 | 9,832 | -18% | 1 | 1 | 0% | 2,146 | 5,544 | +158% | 0 | 0 | — |
case-07 | pass→pass | 16,474 | 15,643 | -5% | 1 | 1 | 0% | 2,673 | 6,408 | +140% | 0 | 0 | — |
case-08 | pass→pass | 14,501 | 19,926 | +37% | 1 | 1 | 0% | 2,453 | 7,295 | +197% | 0 | 0 | — |
case-09 | pass→pass | 11,929 | 9,857 | -17% | 1 | 1 | 0% | 2,079 | 5,500 | +165% | 0 | 0 | — |
case-12 | pass→pass | 9,549 | 10,793 | +13% | 1 | 1 | 0% | 1,533 | 5,601 | +265% | 0 | 0 | — |
case-13 | pass→pass | 4,220 | 4,812 | +14% | 1 | 1 | 0% | 912 | 4,825 | +429% | 0 | 0 | — |
case-14 | fail→fail | 8,134 | 7,321 | -10% | 1 | 1 | 0% | 1,259 | 5,019 | +299% | 0 | 0 | — |
case-15 | pass→pass | 13,679 | 14,155 | +3% | 1 | 1 | 0% | 2,243 | 6,046 | +170% | 0 | 0 | — |
case-16 | pass→fail | 13,808 | 16,117 | +17% | 1 | 1 | 0% | 2,133 | 6,505 | +205% | 0 | 0 | — |
case-17 | pass→pass | 13,779 | 17,423 | +26% | 1 | 1 | 0% | 2,242 | 6,746 | +201% | 0 | 0 | — |
case-18 | pass→pass | 9,044 | 8,295 | -8% | 1 | 1 | 0% | 1,618 | 5,245 | +224% | 0 | 0 | — |
case-19 | pass→pass | 15,121 | 13,919 | -8% | 1 | 1 | 0% | 2,390 | 6,079 | +154% | 0 | 0 | — |
case-20 | pass→pass | 13,132 | 12,536 | -5% | 1 | 1 | 0% | 2,317 | 6,184 | +167% | 0 | 0 | — |
case-21 | pass→pass | 16,098 | 15,876 | -1% | 1 | 1 | 0% | 3,220 | 7,217 | +124% | 0 | 0 | — |
case-22 | pass→pass | 17,658 | 19,185 | +9% | 1 | 1 | 0% | 3,005 | 7,218 | +140% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of 0 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.