A monthly report lands with good news. The company acquired 100 new paying accounts after spending 60,000onpaidmedia.MarketingreportsCACof60,000 on paid media. Marketing reports CAC of 600.

Finance looks at the same 100 accounts and gets $1,200. Its numerator also contains sales salaries, commissions, marketing tools and other acquisition costs. Nothing happened to the customer count; only the cost boundary changed.

Now add recovery time. If fully loaded CAC is 1,200andeachaccountcontributes1,200 and each account contributes 150 a month on a gross-profit basis, simplified payback is eight months. Once variable implementation, support and similar unit-linked costs reduce monthly contribution to 100,thesame100, the same 1,200 takes 12 months to recover.

Those numbers are useful for different operating questions. A channel owner may care about the 600paidCAC.Someonedecidingwhethertoputanother600 paid CAC. Someone deciding whether to put another 10 million behind the whole growth engine needs a wider view of acquisition cost, contribution and the time required to get the cash back.

Most unit-economics mistakes are not division errors. They happen when a familiar label survives while its denominator, cost boundary, cohort or observation window changes underneath it.

One cohort, two CACs

Start with acquisition cost alone. The figures below are hypothetical teaching numbers, not company benchmarks.

CAC viewNew paying accountsAcquisition costs includedCAC
Paid CAC100$60,000 paid media$600
Fully loaded CAC100$120,000 acquisition cost$1,200

Both calculations are internally valid. They do different jobs. Paid CAC can help an operator watch direct channel efficiency; fully loaded CAC can show how much resource the acquisition engine consumes more broadly.

That makes a trend such as “CAC fell from 1,200to1,200 to 600” impossible to interpret until the cost list is checked. If salaries, commissions or tools simply disappeared from the numerator, the business did not suddenly acquire customers at half the cost.

Payback needs its own comparison. Keep fully loaded CAC fixed at $1,200, then change only the contribution boundary:

Contribution basisMonthly unit contributionSimplified payback
Gross-profit basis$1501,200 / 150 = 8 months
Decision-relevant contribution basis$1001,200 / 100 = 12 months

The opening puzzle is therefore two-dimensional: what counted as acquisition cost, and what cash-generating layer is expected to recover it?

Write the measurement contract before the ratio

Before reviewing a unit-economics dashboard, I want the definitions that produced it. A simple measurement contract is enough.

Contract fieldWhat needs to be explicit?
Economic unitUser, payer, account, seat, location, order or transaction?
Eligibility / denominatorWhich units qualify, and how are trials, free users and cancellations treated?
CAC cost boundaryPaid media only, or also sales, commissions, tools, people and other acquisition costs?
Contribution boundaryHow far below revenue does the calculation go, and which variable costs rise with the unit?
Cohort start / ageWhen does a customer enter the cohort, and how mature is it now?
Observation windowThirty days, 12 months, realised lifetime or a forecast horizon?
LTV basisRevenue, gross profit, contribution or discounted cash flow?
Comparison sliceAre product, geography, channel and customer type genuinely comparable?

A dashboard will faithfully store CAC = 600 or NRR = 118%. It may not preserve the customer eligibility rule used at the time, a cost reclassification made two quarters later, or the fact that the current cohort is much younger than the comparison cohort.

Once those definitions drift, a chart can join several different measurements into what looks like one clean time series. The contract gives the team something stable to carry into CAC, LTV, retention and payback analysis.

The cost boundary determines what you are recovering

There is no single CAC numerator that serves every decision. Paid, blended and fully loaded CAC look at acquisition from different operating levels; incremental and marginal CAC become important when the question is what happens to the next block of spend.

The distinction becomes sharper when advertising attribution enters the calculation. A platform may attribute 100 conversions to a campaign, but that does not establish that all 100 were caused by the campaign. Google’s Conversion Lift uses treatment and control groups to estimate causal lift precisely because observed attribution and incrementality are different quantities. Research comparing observational advertising estimates with experimental benchmarks has also found material bias risk.

For capital allocation, an incremental view is therefore more informative when it can be estimated credibly. A randomised experiment will not be practical in every case; the report should still say whether it is using attribution or a causal estimate rather than letting the two blur together.

Contribution has a similar boundary problem from the other direction. Managerial accounting starts contribution margin at sales less variable costs. In an operating model, the useful question is which costs vary with the economic unit under review.

Suppose one B2B account contributes 150amonthonagross−profitbasis.Eachadditionalaccountalsorequiresanaverage150 a month on a gross-profit basis. Each additional account also requires an average 50 of variable implementation, support or usage-linked infrastructure cost. For the decision to acquire another such account, $100 may be the more relevant contribution layer.

This is where the two sides of the model meet: the acquisition-cost layer chosen for CAC has to be paired with a contribution layer that makes sense for the decision. Otherwise payback inherits the mismatch.

LTV is a model, not a stored fact

A customer can have realised revenue and realised costs. LTV goes further because it usually says something about the future.

That is why customer-lifetime-value research has to specify revenue, costs, retention, timing and discounting. Berger and Nasr compared CLV structures under different operating assumptions, while Gupta, Lehmann and Stuart placed acquisition, margin, retention and discounting in the same customer-value framework. The output depends on what the model includes.

A report labelled simply “LTV” might mean revenue LTV, gross-profit LTV, contribution LTV or discounted-cash-flow CLV. Changing that basis between periods can move the reported number even if customer behaviour is unchanged. Reclassifying service costs outside the contribution definition can do the same.

The familiar shortcut Expected lifetime ≈ 1 / churn rate also carries assumptions. It can be a useful approximation when the churn process is stable enough. Fader and Hardie’s work on retention shows why that condition matters: survivor behaviour and retention curves do not have to follow a constant churn process, and tenure or cohort composition can change the shape of the curve.

I therefore keep observed cohort economics separate from predicted LTV. The forecast can still be valuable, but its horizon, retention curve, margin basis, discount rate and cohort maturity belong beside the answer. A long-range LTV estimate contains more model than observation.

Retention needs a cohort, not just a percentage

A retention percentage is incomplete until you know what survived and whose revenue is being measured.

MetricWhat it asksBoundary to check
Logo retentionHow many original customer units remain?Customer survival alone does not describe retained revenue
Gross Revenue RetentionHow much starting-cohort revenue remains before expansion?Keep the same starting cohort
Net Revenue RetentionWhat happened to revenue from that starting cohort after expansion, contraction and churn?Do not add revenue from new customers

Public filings show how much detail can sit behind a familiar label. Snowflake explains a fixed customer cohort and compares the product revenue generated by that cohort across periods. Cloudflare uses Annualized Revenue from the same set of paying customers, includes expansion, subtracts contraction and attrition, and excludes new customers.

Those definitions are illustrations, not a basis for ranking the two companies by a headline retention percentage. A cross-company comparison would still need cohort entry rules, revenue basis, measurement period and customer unit.

The same caution applies inside one company during rapid growth. A business with mostly mature customers one year and a large influx of young customers the next can see aggregate ARPU, churn, margin or support cost move because the age mix changed. The direction depends on the underlying cohort curves, so the aggregate movement alone does not identify better or worse cohort quality.

A useful comparison holds eligibility, cohort start, cohort age, product and pricing context, cost basis and observation window as constant as the data allows.

Payback puts time back into unit economics

Now return to the 1,200fullyloadedCAC.With1,200 fully loaded CAC. With 150 of monthly gross-profit contribution, simplified payback is:

1,200 / 150 = 8 months

With $100 of monthly decision-relevant contribution:

1,200 / 100 = 12 months

This is where a static LTV/CAC ratio can hide something operationally important. Acquisition cash often leaves before contribution arrives. Free periods, upfront commissions, implementation cost, payment terms, usage ramp and customers churning before recovery can all push actual cash payback beyond a simplified margin-based calculation.

During fast growth, that timing matters. If CAC is paid on day zero while contribution arrives over a year or more, accelerating acquisition can increase working-capital or external-financing requirements. Shorter payback generally reduces that pressure, although it says nothing by itself about retention quality, ultimate LTV or market size.

In the hypothetical cohort, changing the contribution boundary alone moves simplified recovery from eight months to 12.

Measurement Contract recap with seven unit-economics definition fields and a hypothetical payback example: fully loaded CAC 1,200, 150 monthly contribution gives an 8-month simplified payback, and $100 gives 12 months; real cash payback may differ.

Run a fixed-definition audit before calling it improvement

When a dashboard says unit economics improved, recalculate the comparison periods under the same contract before looking for a business explanation.

AuditHold constantWhat can go wrong otherwise?
Unit / eligibilitySame customer type and inclusion ruleA smaller denominator improves ratios mechanically
CAC boundarySame acquisition-cost layerCosts move out and CAC appears to fall
Contribution policySame variable-cost treatmentMargin and LTV rise through reclassification
Cohort ageMatched or age-normalised cohortsMaturity mix creates false improvement or deterioration
Comparison sliceSame product, region and channelMix shift masquerades as efficiency
Retention basisSeparate observed from forecastA model change is mistaken for realised improvement
Payback basisSame contribution or cash basisRevenue or a shallower margin layer makes recovery look faster

If the fixed-definition version still shows lower CAC, stronger matched-cohort retention, higher contribution and shorter payback, there is an operating change worth investigating. Channel quality, sales productivity, onboarding cost or product value might explain it.

If much of the improvement disappears in the recomputation, the immediate problem is measurement consistency. Causal stories can wait until the periods are actually comparable.

Only then decide whether to spend more

The 600paidCACand600 paid CAC and 1,200 fully loaded CAC can both stay in the reporting system. They simply should not be given the same decision job.

For a decision to scale the whole go-to-market engine, I would want the economic unit and cost boundaries documented, improvement to survive matched-cohort and fixed-definition checks, contribution to be positive at the chosen decision layer, payback to fit the company’s financing constraints, and retention evidence to support the LTV assumptions being used.

A favourite LTV/CAC rule of thumb cannot substitute for those checks. No single LTV/CAC multiple, including 3:1, should be treated as a universal law. More importantly, one ratio cannot tell you how quickly cash returns, whether cohorts are comparable, whether acquisition is incremental, or whether costs have been moved elsewhere.

If the economics still look attractive after the contract is frozen and the periods are recomputed, the company has a much stronger basis for deciding whether to spend more.

References