What a merchant is worth, what it costs to win one, and how long it takes to earn the money back, by acquisition channel. Built from 28,000 merchants acquired between January 2023 and September 2026.
Unit economics by segment and channel (merchants acquired in the last 12 months)
LTV = discounted contribution margin over 60 months, weighted by the chance the merchant is still active (10% annual discount). Contribution = net revenue after interchange, less processing, fraud losses, support and partner revenue share. CAC = acquisition spend ÷ new merchants, plus onboarding. Payback = months until cumulative expected contribution covers CAC.
What if?
Reallocate paid search budget
Moved spend goes to SMB partnerships and organic in proportion to their current spend. Each extra dollar buys merchants at a rising cost: CAC × (1 + added spend ÷ current spend)0.5.
Merchants still active, % of cohort
Paid search is getting worse
Since 2025, paid search costs about 50% more per merchant, and the merchants it brings churn faster. 2026 cohorts are too young to judge yet. The 2025 LTV uses its observed first-year churn, scaled onto the older curve.
Contribution per merchant-month (last 12 months)
Cohort retention
How it works
Retention: monthly churn hazard by months since acquisition, Kaplan-Meier style, so merchants still active count as censored rather than retained forever. Beyond the observed tenure, the latest year's hazard is carried forward.
Margin curve: average contribution per active merchant at each tenure. Payment volume ramps over the first quarter, then grows (mid-market expands, SMB is flat). The partner revenue share drops off after month 12.
CAC: the last 12 months of spend for each channel (paid media, partner bounties, content, sales salaries and commissions) ÷ new merchants, plus onboarding and KYC cost.
Reallocation: uses each channel's LTV and a square-root cost curve, so moving money is not treated as free, unlimited growth.
All data is synthetic, generated with channel-specific volume, churn and cost profiles. Built with AI-assisted coding (Claude); the analysis and review are my own.