Case 02, Recurring Buy  ·  Appendix A
ROI MODELING + OUTCOMES

ROI-positive
by construction.
Not by scale.

The rule: each enrolled user clears ROI on transaction fees alone, within 12 months, with the incentive fully loaded as a cost. Built on the most conservative input set I could defend to finance. The output was 18.7% ROI per enrolled user, before counting LTV, MTU retention lift, or anything downstream.

PARENT CASE
RECURRING BUY · GEMINI
SCOPE
UNIT ECONOMICS · REWARD DESIGN
HORIZON
12 MONTHS · FEES ONLY
RESULT
18.7% ROI / ENROLLED USER
01 · FRAMEWORK

The rule that gated the program

Each enrolled user had to be profitable on transaction fees alone, within 12 months, with the full incentive loaded as a cost. Not LTV. Not MTU retention lift. Not strategic upside. Just one user, one year, fees in minus reward out.

That rule did two things. It forced the model to be defensible to finance on the weakest input set we could justify. And it made scale a non-question: ROI is a per-user margin, so if the unit economics work at one user, they work at thirty thousand.

DECISION RULE Ship if revenue per enrolled user over 12 months exceeds incentive spend per enrolled user. If profit per user is positive on transaction fees alone, the program is ROI-positive by construction, not by volume.
02 · INPUTS

The conservative input set

Every assumption was set at its defensible floor. Where a reasonable analyst could argue lower, we went lower. Where reasonable analysts could disagree on direction, we went neutral. The point was a number finance couldn’t haggle down.

Input Value Why this value
MTU base (post-initiative) 330,000 Actual achieved base. Pre-initiative was ~300K.
Average monthly buy $100 Conservative floor. Real recurring-buy users transact in this range and above.
Fee rate per side 2.5% Standard published rate.
Sell-side multiplier 1.0× Both sides of a trade are fee-bearing. Symmetric is conservative-to-neutral for accumulate products.
Churn per period 30% Drawn from the weakest historical cohort, not the average.
Incentive cap $150 / user Hard guardrail. Actual expected spend is far below.
03 · THE MATH

Per-user unit economics, churn-weighted

The model takes one enrolled user and projects 12 months. Each period applies the churn rate, so revenue and incentive both decline as the cohort thins. Three periods total (0–3, 4–6, 7–12 months).

// active fraction at each period Active0–3 = (1 – 0.30)1 = 0.70 Active4–6 = (1 – 0.30)2 = 0.49 Active7–12 = (1 – 0.30)3 = 0.343 // fee per user-month (both sides combined) $100 × 2.5% × 2 sides = $5.00 // revenue per enrolled user (12-mo, churn-weighted) Months 0–3: 0.70 × 3 mo × $5.00 = $10.50 Months 4–6: 0.49 × 3 mo × $5.00 = $7.35 Months 7–12: 0.343 × 6 mo × $5.00 = $10.29 Total revenue / enrolled user = $28.14 // 12-month expected payout per active user (tiered by engagement) 1% of active 12-mo users × $100 = $1.00 10% of active 12-mo users × $ 25 = $2.50 40% of active 12-mo users × $ 10 = $4.00 49% of active 12-mo users × $ 5 = $2.45 Expected 12-mo payout / active user = $9.95 // incentive per enrolled user (churn-weighted milestone payouts) 3-mo: 0.70 × $15.00 = $10.50 6-mo: 0.49 × $20.00 = $9.80 12-mo: 0.343 × $ 9.95 = $3.41 Total incentive / enrolled user = $23.71 // outputs Profit / enrolled user = $28.14 − $23.71 = $4.43 ROI on incentive spend = $4.43 / $23.71 = 18.7%

The model is intentionally agnostic to revenue per user, retention beyond month 12, and any downstream behavior. All three of those are real and material, and all three sit on top of the 18.7%.

PER-USER UNIT ECONOMICS · THE BUILDING BLOCK
One enrolled user, twelve months, transaction fees only.
Revenue / user
$28.14

Both sides of every trade, churn-weighted over 12 months.

Incentive / user
$23.71

Churn-weighted milestone payouts. Fully loaded as cost.

Profit / user
$4.43

Net of fully-loaded incentive. Positive at the floor.

ROI
18.7%

On incentive spend. Constant across scale.

$28.14 $23.71 = $4.43 / $23.71 = 18.7%
04 · STRUCTURE

The payout bands

Three milestones. The 3- and 6-month milestones are flat. The 12-month milestone is tiered by engagement: the heaviest users get the largest payout, with the long tail getting a small one. Payouts only go to active users at each milestone, so churned users don’t get paid the milestones they missed.

Milestone Per active user Active fraction Expected / enrolled
Month 3 $15.00 (flat) 70% $10.50
Month 6 $20.00 (flat) 49% $9.80
Month 12 $9.95 (tiered, see below) 34.3% $3.41
TOTAL $23.71

12-month tier structure

The 12-month payout is split across four engagement tiers. The blended expected value of $9.95 per active user comes out of this distribution:

Engagement tier % of active 12-mo users Payout Contribution to blended
Top 1% $100 $1.00
High 10% $25 $2.50
Mid 40% $10 $4.00
Long tail 49% $5 $2.45
Blended 100% $9.95
WHY TIER THE 12-MONTH Concentrating the largest payout on the top-engagement 1% does two things. It rewards the cohort most likely to refer, cross-sell, and produce LTV. And it keeps the blended cost down so the band economics still hold under conservative churn assumptions. The flat 3- and 6-month payouts establish the habit; the tiered 12-month payout rewards the depth.
MAX REALISTIC PER USER: $135. GUARDRAIL CAP: $150. A user who hits every milestone and lands in the top 12-month tier collects $15 + $20 + $100 = $135 over the year. The $150 incentive cap sits above that as a hard guardrail against design drift. The model expects ~$24 per enrolled user; the cap was never approached.
WHY MILESTONES, NOT CONTINUOUS Clean attribution. Clearer user mental model. Easier to model retention thresholds. Finance audits each band against actuals quarter-on-quarter, so deviations show up early.
05 · SCENARIOS

Three modeled cases. Plus what shipped.

The model carried three conversion-lift scenarios pre-launch, then we tracked what actually shipped against them. The per-user economics, and therefore the ROI, are identical across all four columns. That’s the point.

Scenario Conv. lift Enrolled Revenue Incentive Profit ROI
Base Case 6,000 $169K $142K $27K 18.7%
Upside Case 12,000 $338K $285K $53K 18.7%
Result Case 24,000 $675K $569K $106K 18.7%
Actual Shipped 33,000 $929K $782K $146K 18.7%
WHAT THE TABLE PROVES The ROI column is identical across every row. ROI is a per-user margin; volume changes profit, not ROI. The Base Case still clears. Even at the conservative 2× conversion-lift scenario, the program was profitable. That made it shippable on math alone, before any belief in upside.
ACTUAL EXCEEDED RESULT CASE Shipped result landed at 33,000 enrolled users, above the modeled Result Case of 24,000. The model used a 300K MTU base; actual base grew to 330K over the same window. Per-user economics held, profit scaled with the user count.
06 · IN-MARKET VALIDATION

What the experiment confirmed

The unit economics were the case for shipping. The A/B/n experiments were the case for the structure: which part of the experience moved the needle and how much.

Phase Cost Conversion lift Note
Education $0 CAC Pure narrative effect of the simulator. No payout. Vs. control.
Reward ~$24 / user Incremental lift on users on the fence. Vs. control.

The reward layer added incremental lift on top of education (8× total vs. 4× from education alone). That confirmed the band design: half the conversion was narrative, half was money applied surgically at the right milestones.

Aggregate adoption moved from 0.2% to 10% of MTUs, a 50× lift in adoption rate. Recurring-buy users moved from 600 to 33,000, a 55× lift in absolute users.

07 · THE FLOOR

What the 18.7% excludes (the upside)

The 18.7% ROI is the floor, by design. The model deliberately ignores everything that makes recurring buy genuinely valuable:

  • LTV beyond month 12. Recurring-buy users transact for years, not one. The model truncates at month 12.
  • MTU retention lift on the base. Recurring buy is a retention lever; the MTU base grew ~300K to ~330K over the window. The model attributes none of that to the program.
  • Downstream behavior. Recurring-buy users refer more, qualify for the credit card cross-sell at higher rates, and adopt other products. None of that is in the model.
  • The compounding behavior itself. Every adopter transacts automatically thereafter without further marketing. That spend efficiency isn’t reflected.
THE FRAMING 18.7% is the worst case. The real return is multiples higher. Building the floor model first was deliberate: it took the strategic argument off the table and made the program defensible on math alone.
08 · WHAT IT PROVED

The framework as a reusable artifact

ROI-positive incentive design isn’t a guess. It’s modeled, and the model has to clear on the most conservative input set you can defend. The per-user unit-economics framework here, fees only, 12-month horizon, fully-loaded incentive, became the template every retention and activation incentive the growth org has shipped on since. Same math, different products.

The version of incentive design worth scaling is the one where you can prove the floor before anyone has to believe in the upside.