Referral A/B Test: Single vs Double-Sided Incentive
A referral program experiment template that compares a single-sided incentive (only the referrer is rewarded) against a double-sided incentive (both referrer and referred user are rewarded). The test measures the full loop, rather than only invite-send rate, because cheaper-to-send invites can lower the LTV of the users they pull in.
Referral tests are almost always misread because teams measure the wrong step. The right unit is the full loop: send rate times accept rate times LTV of the referred user, against the cost of the incentive. A variant that wins on sends can lose on accepts, and a variant that wins on accepts can lose on the LTV of the resulting customers.
Copy the template
Use it in Notion, a Google Doc, or wherever your team already works.
# Referral A/B Test: Single vs Double-Sided Incentive ## Hypothesis Because [observation about referral volume or referred-user LTV], we will move from a single-sided to a double-sided incentive structure, and expect the loop ratio (send * accept * referred-cohort LTV / incentive cost) to lift for [audience] with no detectable increase in fraud. ## Variants - Control (A): Single-sided incentive (only the referrer rewarded). - Variant (B): Double-sided incentive (both referrer and referred user rewarded). ## Metrics - Primary: Loop ratio = invite-send rate * accept rate * 90-day LTV of referred cohort / incentive cost per referral. - Guardrails: fraud and self-referral rate, support-ticket volume on redemption, referred-cohort LTV vs baseline. ## Math - Sample size: ~2,000 users per variant for 20% relative effect on send rate. LTV guardrail needs the full 90-day window. - Duration: 6-10 weeks. ## Common failure to avoid Calling on invite-send rate. Double-sided incentives almost always lift sends; the question is whether the LTV justifies the cost.
The variants
Single-sided incentive: the existing user gets a reward when a referred user signs up or pays.
Example: Refer a friend, get a $30 credit when they upgrade.
Double-sided incentive: both the referrer and the referred user receive a reward.
Example: Refer a friend, you both get $30 in credit when they upgrade.
Metrics, math, and success criteria
Loop ratio: invite-send rate × accept rate × 90-day LTV of the referred cohort, divided by incentive cost per referral.
Fraud and self-referral rate, support-ticket volume on incentive redemption, LTV of referred-cohort users versus non-referred baseline.
Referral tests run on the existing user base, so size against your monthly active users. As a rule of thumb, you need at least 2,000 users per variant for a 20 percent relative effect on send rate, and the LTV guardrail needs the full retention window before it is readable.
Statistically significant lift in loop ratio with no detectable increase in fraud rate and equal or better referred-cohort LTV.
Six to ten weeks. Long enough to observe both the front of the loop (send and accept) and the back of the loop (referred-cohort retention and revenue).
Expected outcome range
Double-sided incentives commonly lift invite-send rate by 30 to 80 percent, with a smaller effect on accept rate. The LTV of referred users moves less and is the variable that decides whether the win is real; expect 0 to 15 percent variation either way.
Common failure mode
Calling the winner on invite-send rate. Double-sided incentives almost always lift send rate, because the asker has more to offer. The harder question is whether the resulting users are worth what they cost, and that only shows up at the back of the loop.
What this unlocks next
- If double-sided wins on loop ratio, test the size of the incentive next; the optimal amount is rarely the round number.
- If single-sided wins, the deeper question is whether the issue is incentive structure at all or whether the referral surface (in-product placement, timing) is the real bottleneck.
- Either result lets you size the upper bound of how much referral can compound your acquisition, which informs the paid-versus-referral budget split.
Running this template manually vs in GrowthLab
| Step | Manual (spreadsheet) | In GrowthLab |
|---|---|---|
| Design the incentive | Pick an amount that feels right, ship and hope. | AI drafts both incentive structures with the loop ratio framing pre-applied, so send rate alone cannot mislead the team. |
| Prioritize | Referral test ranks below CTA tests because the timeline is longer. | ICE pre-scored. ROTI accounts for the longer LTV window so the test is judged on the right horizon, not the wrong one. |
| Run and track | Track sends and signups, lose sight of fraud and referred-cohort LTV. | Loop ratio is the primary metric. Fraud rate and referred-cohort LTV vs baseline are wired as guardrails, with alerts on spikes. |
| Capture the learning | Next incentive-amount test starts from a hunch, not a baseline. | The result, the loop math, and the incentive cost are stored, so the next incentive-amount test starts from your own data. |
Inside GrowthLab
Inside GrowthLab the referral template ships with the loop ratio as the primary metric (not send rate), the LTV-of-referred-cohort guardrail wired against your baseline, and a fraud-rate tracker on incentive redemption. ROTI is computed once the LTV window closes, and the learning is stored so the next incentive-amount test inherits the baseline.
Frequently asked questions
Should I run a single-sided or double-sided referral program?
It depends on your unit economics. Double-sided incentives almost always lift invite-send rate because the referrer has more to offer, but the test that matters is whether the resulting users are worth the higher incentive cost. Run the experiment and let the loop ratio decide, not the send rate.
What metric should I optimize for in a referral test?
The full loop: invite-send rate times accept rate times the 90-day LTV of the referred cohort, divided by the incentive cost per referral. Send rate alone overweights the front of the loop; LTV alone ignores volume. The loop ratio captures both.
How do I prevent fraud in a referral A/B test?
Track referral-to-account fingerprints (device, IP, payment instrument) and gate the incentive payout on a meaningful product action by the referred user, rather than only signup. A fraud-rate guardrail catches the case where a variant attracts gaming behavior at scale.
Go deeper
- What is a growth loop
- What is a North Star Metric
- Growth loops guide
- Growth experimentation: the 2026 ultimate guide
About GrowthLab
GrowthLab is an experiment management tool where AI drafts the hypotheses, ICE and ROTI prioritize them, and every learning compounds into the next batch.
