Unit economics: the four checks that decide whether growth helps or hurts

Why the sign matters more than the size

Unit economics is the per-customer profit arithmetic of a business: what it costs to acquire one customer, what that customer contributes over their lifetime, and how long the payback takes — the numbers that determine whether growth compounds value or destroys it.

The reason this sits ahead of almost everything else in an assessment is that it decides the sign of growth. If each customer eventually contributes more than they cost, growth compounds and the only real question is how fast you can fund it. If each customer costs more than they contribute, every improvement in the growth chart brings insolvency closer, and the chart looks identical either way.

That is the whole reason investors check these four numbers before they get excited about a curve.

The four checks

1. The identity

Does annual revenue equal customers × ARPU?

This sounds too trivial to state, and it is the check that fails most often. A model showing 25 customers at $120 monthly ARPU and $3.75M annual revenue has an error of three orders of magnitude — the real figure is $36,000. It happens because the revenue line and the customer line get maintained in different places by different people at different times.

Run this before anything else. INITE Studio's validator treats a drift above 15% between stated revenue and customers × ARPU as a blocking error, with the slack absorbing mid-year cohorts and churn timing rather than genuine mistakes.

2. LTV over CAC

CAC is all sales and marketing spend in a period — salaries and tooling included, not just ad budget — divided by customers acquired in that period. Excluding salaries is the single most common way CAC gets reported at a fraction of its real value.

LTV is gross profit per customer over their expected lifetime. Gross profit, not revenue: if it costs you 30% of the subscription to deliver the service, only 70% is available to pay back acquisition.

The ratio:

• Below ~1.5 — payback is arithmetically impossible once margin and churn are applied. This is a blocking condition, not a warning.
• 1.5 to 3.0 — sub-prime. Workable at very early stage while you are still finding the channel, not workable as a plan.
• 3.0 and above — the common definition of healthy.
• Above 5.0 — usually read as underinvestment rather than excellence. If each customer returns five times their acquisition cost, the obvious question is why you are not buying more of them.

3. Cohort survival

Take a cohort of customers acquired in one month. Apply your actual monthly churn. What fraction of that revenue is still there twelve months later?

At 5% monthly churn, roughly 54% survives the year. At 8%, closer to 37%. This is compounding working against you, and it is why an LTV computed from an annual churn assumption tends to be far more generous than one computed from the monthly rate you actually observe.

The check that matters: does the cohort still generate positive contribution by the time payback should have happened? If churn eats the revenue before CAC is recovered, the LTV/CAC ratio was describing a customer who does not exist.

4. CAC payback

How many months until a customer has repaid what it cost to acquire them?

CAC ÷ (gross margin × monthly ARPU). Dividing by revenue instead of gross profit is the standard error, and it understates the payback period by exactly your cost of delivery.

• Under 12 months — strong.
• Around 18 months — the standard early-stage benchmark.
• Beyond ~36 months — the model has stopped being a business. At that horizon you are financing customers, and the financing cost is not in the model.

Payback matters independently of LTV/CAC because it determines how much cash you need to grow. Two companies with identical ratios and payback periods of 6 and 30 months have completely different funding requirements, and the second one runs out of money first.

A worked example

A team reports: 400 customers, $95 monthly ARPU, $520K annual revenue, CAC $780, monthly churn 4%, gross margin 72%.

Identity. 400 × $95 × 12 = $456,000. Reported $520,000. Drift is 12% — inside the 15% tolerance, plausibly mid-year cohorts. Passes.

Lifetime. At 4% monthly churn, average customer lifetime is 1 ÷ 0.04 = 25 months. Gross profit per month is $95 × 0.72 = $68.40. LTV is 25 × $68.40 = $1,710.

Ratio. $1,710 ÷ $780 = 2.19. Below the 3.0 healthy line, above the 1.5 floor. Sub-prime: workable now, not a plan.

Payback. $780 ÷ $68.40 = 11.4 months. Comfortably inside 18.

Cohort. After twelve months at 4% monthly churn, about 61% of the cohort's revenue remains — and payback lands at month 11, so the cohort clears its acquisition cost before the erosion becomes serious. Passes.

The diagnosis is specific: payback is healthy, the ratio is not, and the gap between those two facts is churn. A 25-month lifetime is the constraint. Halve churn to 2% and lifetime doubles to 50 months, LTV goes to $3,420, and the ratio moves to 4.4 — without touching CAC or price.

That is the value of running all four checks rather than one. A single ratio would have said "below three, needs work". The four together say "the acquisition engine is fine, fix retention", which is a different quarter.

The four levers, and which actually move

The ordering surprises people. Founders reach for CAC first because it feels controllable, and it is the lever most likely to move against you over time: the cheapest channel is cheapest because it is small, and scaling means buying more expensive customers.

Price is the most underused. A 20% price increase raises LTV by 20% and shortens payback by roughly 17%, with no operational change — and in B2B the churn cost of a modest increase is frequently smaller than founders fear.

Churn is the hardest and worth the most, because it is the only lever that compounds. It also cannot be fixed by scale, which is why "we'll fix retention later" is the assumption most likely to end a company.

Cohorts, and why the monthly average lies

Every number above is computed over a population. The moment that population is a mix of vintages, the average stops describing anybody.

A company acquiring customers every month has cohorts at different ages. Blended churn averages a healthy two-year cohort against a leaky one acquired last quarter through a new channel, and reports something in between that matches neither. The blended number goes up and to the right while the recent cohorts — the ones that represent the business you are actually building — get worse.

The check is mechanical: take three cohorts six months apart and plot retention by month-since-signup on the same axes. Three shapes are possible.

Curves flattening at a similar level. Healthy. The product holds a stable fraction of who it acquires.

Later cohorts retaining worse. Something changed — usually a channel. Growth is being bought from a worse-fitting population, and the blended average will hide it for two or three quarters.

Later cohorts retaining better. The product is improving. Blended numbers understate you, and an LTV built from them is too conservative.

The same split applies to CAC. A blended figure across organic and paid hides that the paid channel — the one you would scale — may be unprofitable standing alone. Compute LTV/CAC per channel or the ratio describes a business nobody is running.

The B2B and consumer split

The thresholds in this guide are SaaS conventions, and two adjacent models break them.

Long sales cycles. When the cycle runs six to nine months, the CAC incurred in Q1 produces revenue in Q3, and a naive period calculation — this quarter's spend over this quarter's customers — attributes cost to the wrong cohort. It overstates CAC while you are growing and understates it while you are flat. Match spend to the cohort it acquired, not to the calendar.

Usage-based pricing. ARPU is not a constant, so "monthly ARPU × lifetime" is the wrong shape. Expansion within an account frequently exceeds churn out of it, which is what net revenue retention above 100% means, and at that point a lifetime-value calculation built on a fixed monthly figure understates the business badly. Model the expansion curve or model nothing.

Marketplaces and consumer have their own problem: the customer may be free and the revenue may come from the other side, so a per-user LTV is meaningless and contribution per transaction is what matters.

The general principle survives all three: cost to acquire, value delivered, time to recover. The specific formula does not.

The order to fix things in

When several of the four checks fail, the sequence matters, because two of them are cheap and two are quarters of work.

1. Fix the identity first. If revenue does not equal customers times price, every other number is computed on sand. This is an afternoon.
2. Recompute with gross profit, not revenue. Half of unhealthy-looking unit economics are healthy ones measured wrongly, and half of healthy-looking ones are the reverse. Also an afternoon.
3. Segment by channel and cohort. Frequently the business is fine and one channel is not. Days, not weeks.
4. Then act. Price first — it is the fastest real lever and the most underused. Then retention, which is the largest and slowest. CAC last, because it is the one most likely to move against you as you scale.

Founders reliably start at step four and reach for CAC, because it feels like the controllable one. It is the one that gets harder with size.

What breaks these numbers

Blended CAC across channels. Organic and paid customers cost wildly different amounts. A blended figure hides that the paid channel — the one you would scale — is unprofitable on its own.

LTV from revenue instead of gross profit. Inflates every downstream number by your cost of delivery.

Annual churn applied to a monthly reality. Compounding makes these very different, and the annual figure always flatters.

Excluding salaries from CAC. Sales and marketing headcount is acquisition cost. Leaving it out is how a $780 CAC gets reported as $180.

Survivor-biased LTV. Computing lifetime from customers who are still around measures the ones who did not churn, which is the wrong population.

Each of these produces a number that looks fine. That is what makes them worth checking mechanically rather than by eye — and why the validator treats identity drift as blocking rather than advisory. A model that fails arithmetic does not need a judgement call.