Growth economics and measurement architecture · Lesson 2

Unit economics and cohort quality

9 min read

Lesson objectives

  • Calculate cohort contribution, acquisition cost and observed payback on comparable horizons.
Reading tools

When two channels compete for budget, average customer acquisition cost alone is not enough. A customer acquired today may buy once and leave, or remain for years. Years that have not happened must not be recorded as fact. Unit economics examine revenue and variable cost associated with one decision unit, such as a customer or order, and how long it takes to recover the cost of acquiring it. A cohort is a group that entered under one criterion during one period, such as new subscribers in May. Do not mix a first-purchase cohort with a registration cohort, or compare a six-month-old cohort with a one-month-old cohort without aligning their observation horizons. If channel A is cheaper but attracts people who cancel quickly, dearer channel B may deliver more value. The task is to expose this difference without turning projected lifetime value into false certainty.

Start with contribution, not gross revenue. Suppose a shop sells an item for 100, with merchandise cost 55, shipping 10 and expected support 5. Contribution before advertising is 30. If 10% of orders are returned and each return costs another 20 to process in addition to a refund, the expected-cost model must be defined precisely: is 20 an incremental handling cost, and can merchandise be recovered? Do not count the refund and merchandise cost twice. In a simplified fictional model, if a completed order contributes 30, a returned order contributes -20, and 90% complete, expected contribution is 0.9×30 + 0.1×(-20) = 25. These numbers define mutually exclusive outcomes for teaching. An actual business must derive them from its finance records and return policy. Different margin definitions across teams change the allowable acquisition cost.

Customer acquisition cost, or CAC, is the acquisition spend attributed to a cohort divided by its new customers under an agreed definition. If a campaign spends 12,000 and records 400 new buyers, descriptive CAC is 30. Do not count 400 orders if some were placed by returning buyers or are duplicates. Agency fees, creative production or referral discounts might be included; document what was included and excluded before comparing an internal channel with an outsourced one. There is no universal CAC ceiling. If a customer contributes 12 in month one, 10 in month two and 8 in month three, total contribution is 30. Spending 30 to acquire that customer reaches nominal payback at the end of month three, before shared fixed cost and risk. If the company requires payback within two months, the same customer fails that constraint even if later contribution might be positive.

Calculate payback cumulatively for each cohort. Order each period’s contribution after variable costs, then subtract CAC once at the beginning. At month zero the balance is -30. After month one with contribution 12 it is -18; after month two with 10 it is -8; after month three with 8 it reaches zero. Dividing CAC by an average monthly contribution is misleading when retention changes sharply; the average can hide that many customers leave before break-even. If the observed cohort balance never reaches zero within available data, report “not yet recovered” rather than extending a straight line into an invented month. The decision also concerns cash and the ability to fund successive cohorts, not merely a return percentage.

Test how fragile the ranking is before turning it into a budget rule. Search would lose its current 1,500 advantage if it lost 60 monthly payments worth 25 in contribution. That could follow a rise in refunds or a correction to subscription records. Equally, projecting the 280 partnership payers into every later month without evidence would be unjustified. Build conservative and optimistic scenarios from mature data, labeling each assumption. Monthly contribution may also change with renewal discounts or heavier support use. Show one row per cohort and one column per month, and compute payment counts and margin, rather than displaying a single retention rate. The table exposes what drives the channel difference.

Watch for a blended average when the channel mix changes. Suppose month-one retention is 80% for search and 90% for partnerships. If the cheaper search channel takes a larger share of new customers, overall retention may fall even while neither channel changes its own performance. That does not establish that the product team harmed the experience; it is a composition effect. Display each channel and entry cohort separately, then calculate a weighted average if an overall description is needed. If subscription price or the definition of an active customer changes mid-period, do not interpret a changed ratio as a quality effect before re-segmenting cohorts. Budget decisions themselves alter the mix of customers described by the dashboard. The right conclusion may be that scaling one channel changes portfolio economics, not that every channel improved or deteriorated.

A future value estimate may help a decision, but label it as a scenario with explicit assumptions. For example, estimate month-four contribution if 60% of month-three payers remain, then ask whether older cohorts comparable to the current one support that rate. For partnerships in the case, the assumption yields 280×0.60×25=4,200. This is a projection, not collected cash. If support demand rises after month three, actual contribution may be lower. Do not use a lifetime-value point estimate alone to approve expensive acquisition. Show a range, payback period, available liquidity and the cohort size behind the projection. When the apparent channel gap is small relative to retention uncertainty, gathering more evidence or running a bounded spend test is more coherent than declaring a permanent winner.

Beware of incomplete observation. A report produced on the 22nd cannot classify a buyer acquired on the 20th as a 30-day non-returner; 30 days have not passed. Include only cohorts old enough for the chosen retention horizon in its denominator, and mark newer cohorts as immature. Mixing them in makes retention appear to fall when acquisition accelerates late in the period. Also distinguish a cancellation request from an actual missed payment if a grace period applies. Define the departure event for the financial question being asked.

Lesson summary

Calculate cohort acquisition payback on a mature common horizon.