Incentive Structure Tuning

Tune the reward size, shape, and timing to maximize k without breaking unit economics or attracting fraud.

Why

The incentive size has a non-linear effect on k: below a threshold, the reward isn't worth the social friction of inviting, so k stays flat; above it, k jumps, then plateaus again as the reward exceeds the social cost. Finding the threshold (usually $5-15 for consumer, $50-200 for B2B) requires testing 3-4 sizes, not guessing. The reward shape (flat vs tiered vs lottery) changes who participates: flat rewards get broad participation, tiered rewards motivate the top 10% of inviters who drive 60% of volume. Timing matters too: instant rewards convert 2x delayed rewards because the inviter sees the payoff and invites again. Tuning all three levers (size, shape, timing) is what separates a k of 0.3 from a k of 0.8 on the same product.

When

When the program is live and k has plateaued. Tune one lever at a time, measure for 2 weeks before the next.

How

1. Baseline the current k, cost/acq, and fraud rate over 4 weeks so you have a stable comparison
2. Lever 1 — size: test 3 reward sizes (current, 1.5x, 2x) in parallel cohorts; plot k vs cost/acq for each
3. Find the knee: the size where k jumps most per dollar of reward; below it you're underpaying, above it you're overpaying
4. Lever 2 — shape: A/B flat ($10 always) vs tiered ($5 for 1-2 invites, $15 for 3+, $30 for 10+) vs lottery (chance at $100)
5. Pick the shape that maximizes k at your fraud tolerance: tiered drives volume but attracts professional inviters
6. Lever 3 — timing: A/B instant credit (on recipient signup) vs delayed (on recipient activation or first payment)
7. Instant wins for consumer (the inviter re-invites immediately); delayed may win for B2B (filters out low-quality invites)
8. Lock the winning combination, re-measure for 4 weeks, then revisit quarterly as the audience saturates

Example

# Lever 1 — size test (3 parallel cohorts, 2 weeks each)
Cohort A: $5 reward   → k=0.28, cost/acq=$10, fraud=0.8%
Cohort B: $10 reward  → k=0.41, cost/acq=$20, fraud=1.1%  ← knee
Cohort C: $20 reward  → k=0.44, cost/acq=$40, fraud=2.9%
# $10 is the knee: k jumps +46% from $5, only +7% more from $20 at 2x cost

# Lever 2 — shape test (at $10 size)
Flat $10:        k=0.41, 80% of users invite 1, top 10% invite 3
Tiered 5/15/30:  k=0.52, 60% invite 1, top 10% invite 8  ← higher k, more concentration
Lottery $100:    k=0.31, broad but low effort per invite
# Tiered wins on k; accept concentration risk (fraud watch on top inviters)

# Lever 3 — timing (at $10 tiered)
Instant (on signup):    k=0.52, inviter re-invites within 2d
Delayed (on activation): k=0.38, filters low-quality but loses re-invite loop
# Instant wins for consumer; lock: $10 tiered, instant

Common Problems

⚠Guessing the reward size instead of testing 3-4 — you land below the knee (k flat) or above it (overpaying for the same k).
⚠Using a flat reward when a tiered one would motivate the top inviters who drive 60% of referral volume.
⚠Delaying the reward 'to prevent fraud' — the delay kills the re-invite loop and costs more k than the fraud it prevents.
⚠Changing all three levers at once, so you can't tell which change caused the k movement.

Best Practices

✓Test 3 reward sizes in parallel and pick the knee where k jumps most per dollar; below it you underpay, above it you overpay.
✓Use tiered rewards to motivate the top 10% of inviters who drive the majority of volume; flat rewards cap your k.
✓Give rewards instantly (on recipient signup) for consumer products; the re-invite loop is worth more than the fraud it risks.
✓Change one lever at a time and measure 2 weeks before the next; confounded changes can't be attributed.