CAC Optimisation: The Four Frameworks That Fix Broken Metrics

Customer acquisition cost is not one metric serving one purpose. An operational team, a finance function and a board can all ask about CAC while requiring different costs, different customer definitions and different standards of evidence. Problems arise when a blended or channel-attributed figure is applied to a decision it was never designed to answer.

This guide explains four CAC views, which are working, fully loaded, incremental and channel-attributed, and shows which decisions each can and cannot support. For practical methods to lower acquisition cost while protecting customer contribution and retention, see the guide to reducing CAC without sacrificing customer quality.

A note on evidence. Published claims are numbered and referenced below. Anonymous client examples are first-hand observations, not representative benchmarks.

Why your CAC number is probably wrong

Most CAC disputes are not disagreements about performance. They are two people using the same three letters to mean different things and not realising it.

The number changes, sometimes by a factor of two or more, depending on four choices nobody writes down:

Which costs go in the numerator. Media only, or media plus salaries, agencies, tools, creative production and sales labour.
Which customers go in the denominator. All new accounts, or first-time paying customers, or customers net of refunds and cancellations, or accounts excluding reactivations and duplicates.
Which period each side covers. Spend in the month it was incurred, against customers who converted in that month, even where the sales cycle is ninety days.
What counts as caused by the spend. Every conversion a platform claims, or only the customers who would not otherwise have arrived.

Any of those four can be defended. What cannot be defended is changing one of them and reporting the movement as a performance result. In my experience the most common version is a business that improves its cost allocation, watches CAC rise, and spends a quarter investigating a marketing problem that never existed.

Fix the definition before you interpret the number.

The four CAC frameworks

Working CAC for daily operations

Working CAC includes the costs an operating team can change this week. Typically that means media spend, promotional discounts funded by marketing, and directly variable campaign costs.

Use it for: in-flight campaign decisions, bid and budget adjustments, creative and channel iteration, weekly operational review.

Do not use it for: pricing decisions, investor reporting, or any judgement about whether the business is profitable at the customer level. Working CAC deliberately excludes real costs, so it will always look better than the economics.

Fully loaded CAC for strategic planning

Fully loaded CAC adds the costs required to run the acquisition function: marketing salaries, sales compensation, agency retainers, technology licences, creative production, data and analytics support, and any operational cost incurred specifically to convert or onboard a customer.

Use it for: annual planning, pricing, unit-economic assessment, board and investor reporting, and any comparison against customer contribution.

Do not use it for: day-to-day channel optimisation. Allocating fixed overhead across a weekly view produces movements that operators cannot act on.

The gap between working and fully loaded CAC is business-specific. Some organisations find it small. Sales-assisted businesses with long cycles usually find it large. Calculate it rather than assuming a ratio, and publish both figures with their definitions attached.

Incremental CAC as the standard for budget allocation

Incremental CAC is the additional cost of acquiring customers who would not otherwise have been acquired. It is the only one of the four that answers the question budget decisions actually pose, which is what the next pound will produce.

It is also the hardest to obtain, because it requires a credible counterfactual: a holdout, a geographic or matched-market test, a controlled spend change, a marketing-mix model, or another design that estimates what would have happened anyway.

Use it for: channel reallocation, scaling decisions, deciding whether to enter or exit a channel, and any claim that a change created a saving.

Do not use it for: routine weekly reporting, where the measurement cost usually exceeds the value.

Controlled experiments remain the strongest general design for establishing that a change caused an outcome, though they require adequate sample size, valid randomisation and evaluation criteria agreed before the test starts. (1)

Channel-attributed CAC and why it should not drive allocation

Channel-attributed CAC divides channel spend by the conversions a model assigns to that channel. It is useful for diagnosis and misleading as a budget rule, because attribution distributes credit according to a rule somebody chose rather than measuring what caused anything.

The research on this gap is consistent and uncomfortable. In a large eBay field experiment, the people clicking advertisements often already intended to purchase, so non-experimental estimates materially overstated the causal return for parts of the programme. (3) A comparison of fifteen large Facebook experiments found that observational methods frequently failed to reproduce randomised effects even after extensive adjustment for demographic and behavioural variables. (4) A later study analysing 663 large-scale experiments concluded that the observational approaches examined could not reliably estimate a campaign’s causal effect, even with access to far richer user-level data than most advertisers hold. (5)

That last point deserves emphasis, because it removes the usual escape route. The problem is not that your data is insufficiently detailed. Researchers with more than five thousand user-level features could not close the gap either.

Use it for: noticing that something changed, and deciding what to test next.

Do not use it for: declaring a channel efficient, cutting a channel, or claiming a saving.

Choosing the numerator: which costs belong in CAC

Decide the cost boundary by the decision you are making, then hold it constant.

Cost	Working CAC	Fully loaded CAC
Paid media	Include	Include
Marketing-funded discounts and offers	Include	Include
Agency fees	Usually exclude	Include
Marketing salaries	Exclude	Include
Sales compensation and commission	Exclude	Include
Marketing technology licences	Exclude	Include
Creative and content production	Case by case	Include
Analytics and data support	Exclude	Include
Onboarding and implementation cost	Exclude	Include where required to convert
Brand and category-creation spend	Case by case	Include, with a note on its delayed effect

Two recurring arguments are worth settling in advance.

Brand spend. Excluding it flatters CAC and makes demand creation look free. Including it inside a monthly figure attributes long-dated effects to a short window. The workable answer is to include it in fully loaded CAC while stating that its effects are delayed and distributed, and to evaluate it through longer-term demand measures rather than monthly CAC.

Founder and executive time. In early-stage companies this is often the largest genuine acquisition cost and the one most consistently omitted. If a founder spends most of their week selling, a CAC that excludes them is not measuring the business.

Whatever you decide, contribution on the other side of the comparison has to be built consistently. The Association of Chartered Certified Accountants defines contribution as total revenue less total variable costs. (2) A fully loaded CAC compared against revenue-based lifetime value is not a unit-economic assessment. It is two unrelated numbers in the same sentence.

Choosing the denominator: which customers count

The denominator is where most quiet inflation happens.

Decide explicitly whether you count:

new customers only, or all customers including reactivations;
accounts or individuals, since shared and household accounts distort both sides;
customers gross of refunds and cancellations, or net;
trial starts, activations or first paying conversions;
deduplicated records, or whatever the CRM contains.

That last one is not a technicality. One of the largest cosmetics groups I worked with had never reconciled its offline database against its several online-store databases, so every purchase a customer made on a new channel was stored as a new user. Their customer count was inflated and their CAC correspondingly understated, and nobody had noticed because each system was internally consistent. Before you interpret a CAC trend, confirm that the denominator counts people rather than records.

Aligning the time period

Spend and customers must cover the same acquisition event, not the same calendar box.

Where the sales cycle is short, monthly spend against monthly customers is adequate. Where it is long, that comparison comes apart. Current-period cost gets divided by customers produced by earlier-period activity, so CAC looks better while spend is growing and worse when spend is cut, in both cases for reasons that have nothing to do with efficiency.

Two workable approaches:

Cohort matching. Attribute spend to the period in which the customers it produced converted, using the observed lag.
Lagged comparison. Compare spend in period one against customers in period two, where the lag reflects the median cycle.

Pick one, document it, and do not change it in a quarter when performance is being reviewed.

The hidden cost in every lost deal

Acquisition spending is consumed by every opportunity, not only the successful ones.

If a business spends to generate a hundred opportunities and converts ten, the cost of the ninety unsuccessful ones sits inside the CAC of the ten that closed. This is why efficiency and effectiveness are not the same thing. A cheaper lead that converts less often can raise the cost per acquired customer while every upstream metric improves.

It also means CAC is partly a measure of qualification and sales execution rather than media buying, which is uncomfortable for whoever owns the CAC target.

The methodology does not tell you which intervention to use. Poor-fit demand, weak qualification, funnel leakage and sales execution can all increase the cost absorbed by each successful customer, and they call for different responses. Diagnosis and prioritisation are covered separately, in the reduction guide linked at the end of this article.

Why external CAC benchmarks are hard to use

External CAC benchmarks are difficult to interpret because published figures frequently use different cost scopes, customer definitions, sales cycles, margins and attribution methods. A comparison is useful only when the numerator, denominator, period, business model and customer maturity are sufficiently similar, and published benchmarks almost never disclose enough to establish that.

There is also a structural reason to expect wide variation between companies that look comparable. Research modelling acquisition and retention spending across wireless telecommunications markets in 41 countries found that acquisition cost per customer was more sensitive to market position and competitive conditions than retention cost was. (9) Two competitors in one category can face materially different acquisition economics without either managing the function differently.

The same caution applies to customer-value concentration, which is often imported as a benchmark. Research revisiting the Pareto rule across a large sample of businesses found concentration ratios differing materially by product, service, subscription and non-subscription model, with an overall average well below the familiar formulation. (10)

Treat benchmarking as a reason to investigate, not as a method for setting targets. The strongest comparison is normally your own CAC over time, recalculated on a stable definition and segmented by acquisition motion and customer type.

If you do cite an external figure, require four things before publishing it: the population measured, the metric definition, the observation period, and who paid for the research. Most widely circulated CAC statistics fail at least two of those tests, and several fail all four.

Data foundations and reported CAC

Fragmented systems create inconsistent cost allocation, duplicate customer records and conflicting attribution outputs.

These are measurement-control problems rather than evidence that consolidating software will reduce acquisition cost. Any change to the marketing technology stack should be assessed on total ownership cost, data quality, operational adoption and its measured effect on customer acquisition, using the same evidential standard you would demand of a media test.

I would add one observation from a decade of selling and servicing this category. The reported benefit of a platform change is almost always bundled with several simultaneous changes to strategy, process, targeting and staffing, which makes the platform’s individual contribution unrecoverable after the fact. That is not an argument against consolidating. It is an argument for not believing the number afterwards.

Attribution, incrementality and marketing-mix modelling

Three different tools answer three different questions.

Method	Question it answers	Main limitation
Attribution	How should observed conversions be allocated across touchpoints?	Allocates credit by rule; does not establish causation
Experiment	Did this specific change cause a difference in outcome?	Requires scale, valid randomisation and time
Marketing-mix modelling	How have spend levels related to outcomes across a long period?	Correlational, sensitive to specification, poor at short-run detail

None of the three is a substitute for the others, and each fails differently.

Advertising experiments in particular often need very large samples, because the customer outcome is variable and the advertising effect is small relative to that variance. Lewis and Rao found wide uncertainty around return estimates even in large field experiments. (6) The practical consequence is that an inconclusive test should be reported as inconclusive rather than converted into a point estimate.

Experiment design also has to reflect how platforms actually select and deliver advertising. Methods such as ghost ads were developed specifically to make advertising control groups more efficient and more relevant, which is a reminder that a naive holdout can be biased by the delivery system itself. (7)

For selecting among these methods and designing the reporting around them, use the marketing measurement framework.

CAC payback methodology

Payback measures how long the business waits to recover the acquisition outlay. It is a cash question, not a profitability question, and it frequently binds before profitability does.

Calculate it as acquisition cost divided by periodic contribution per customer, not periodic revenue. Using revenue produces a payback period that is wrong by exactly the size of your variable costs, always in the flattering direction.

State four things whenever a payback figure is published:

which CAC definition was used;
whether the denominator is contribution or revenue;
whether the figure is a mean or a median, since acquisition cohorts are usually skewed;
the cohort maturity, because a payback estimate on a young cohort is a forecast.

Acceptable payback depends on cost of capital, funding position, growth rate, contract structure and how confident you are in the retention assumption. There is no cross-industry payback threshold that survives contact with those variables, and quoted medians generally describe whoever happened to be in someone’s sample.

The LTV:CAC ratio

The ratio compares the value a customer is expected to produce against what it cost to acquire them. It is a useful diagnostic and a poor objective.

Academic customer-valuation work defines customer value as expected discounted future earnings rather than revenue. (8) A ratio built on revenue-based lifetime value and fully loaded CAC is comparing incompatible quantities.

Four principles hold regardless of business model.

A ratio cannot be interpreted without its inputs. Margin, growth strategy, payback, forecast horizon and the confidence behind the lifetime estimate all change what a given ratio means. Two businesses reporting the same figure can be in entirely different positions.

Maximising the ratio is not the objective. A high ratio is consistent with underinvestment. It can mean the business is refusing to fund growth that would create value at a lower ratio. The formal version of this point is that the spending level maximising return on investment is lower than the level maximising total profitability. (11) If your incentive is tied to the ratio and your board wants profit growth, you are being paid to underspend.

Ratio and cash recovery are different constraints. Two investments can share an identical ratio while having radically different payback periods and working-capital demands. A company can pass the ratio test and still run out of money.

Thresholds are internal decision rules, not empirical findings. If your business uses a minimum ratio to approve spend, present it as a policy your finance team set, with the assumptions behind it, and run sensitivity analysis on the customer-lifetime input. The commonly quoted cross-industry targets do not have the evidence base their repetition implies.

For the full treatment of how ratio, payback and marginal contribution should drive spending decisions between acquisition and retention, see how to allocate acquisition and retention investment.

Which metric for which decision

Decision	Use	Do not use
Weekly campaign optimisation	Working CAC	Fully loaded CAC
Channel reallocation	Incremental CAC	Channel-attributed CAC
Annual budget setting	Fully loaded CAC with incremental evidence	Blended historical average
Pricing	Fully loaded CAC against contribution	Working CAC
Board and investor reporting	Fully loaded CAC, with the definition stated	Working CAC without disclosure
Assessing customer quality	Cohort contribution and realised payback	First-order revenue
Entering a new channel	Test design plus incremental CAC	Benchmarks from other companies
Claiming a saving	Incremental CAC with a control	Any attributed figure

Reporting CAC to boards, investors and operating teams

Different audiences need different figures, and the failure mode is publishing one number to all three.

Operating teams need working CAC at a frequency they can act on, segmented by campaign and channel, with the understanding that it excludes real costs.

Finance and executive teams need fully loaded CAC by acquisition motion and customer segment, matched to the right period, alongside contribution and payback.

Boards and investors need fully loaded CAC with the definition disclosed, a stable methodology across periods, and explicit notice whenever that methodology changes. A definitional change presented without a note is the single most damaging thing you can do to your own credibility, because it will eventually be discovered and every prior figure becomes suspect.

Publish a short definitions page alongside the reporting. It takes an afternoon and it ends most recurring arguments.

Common CAC calculation errors

Comparing periods after changing the definition. Record definitional changes separately from performance movement.
Mixing blended and new-customer CAC. Reactivations in the denominator make acquisition look cheaper than it is.
Using revenue-based LTV against fully loaded CAC. The comparison is incoherent.
Ignoring the sales-cycle lag. Growing spend flatters CAC; cutting spend inflates it.
Treating attributed conversions as caused conversions. This is the error the research keeps documenting. (3) (4) (5)
Counting records rather than people. Unreconciled identity understates CAC.
Excluding sales labour in sales-assisted businesses. The largest cost is then missing.
Averaging across segments with different economics. A blended figure can hide one channel funding another’s losses.
Reporting a payback estimate on an immature cohort as a result. It is a forecast.
Importing an external benchmark without its definitions. See the section above.

From measurement to intervention

Selecting the right CAC measure tells you whether an apparent problem is operational, fully loaded or incremental. It does not tell you which action will fix it.

Practical interventions may involve customer qualification, message and offer continuity, conversion friction, the sales hand-off, channel experiments or the development of new acquisition routes. Their effect should be evaluated using the CAC definition appropriate to the decision, with contribution and retention guardrails attached. See the practical guide to diagnosing and reducing unnecessarily high CAC.

Where the question is not how to reduce cost but which customers are worth paying more for, see when a higher CAC can be rational.

Conclusion

CAC is a family of related measurements, not a single number, and most of the damage it does comes from using one member of that family to answer another one’s question.

Decide the cost boundary, the customer definition and the time alignment before you interpret anything. Use working CAC to operate, fully loaded CAC to plan and report, incremental CAC to allocate, and attributed CAC only to decide what to investigate next.

Then hold the definitions still long enough for the trend to mean something. Most organisations do not need a better metric. They need the same metric, calculated the same way, for four consecutive quarters.

Frequently Asked Questions

What is the difference between blended CAC and new-customer CAC?

Blended CAC divides total acquisition spend by all customers recorded in the period, including reactivations, renewals and in some systems duplicate records. New-customer CAC divides the same spend by first-time customers only. Blended CAC is almost always the lower and more flattering figure, and the gap between the two widens as a business matures and its reactivation volume grows. Publish both, state which is which, and never compare one period's blended figure against another period's new-customer figure.

When should working CAC be used rather than fully loaded CAC?

Use working CAC for decisions an operating team can act on within days, such as bid changes, budget shifts between campaigns and creative iteration, because it moves quickly and reflects costs that team controls. Use fully loaded CAC for annual planning, pricing, unit-economic assessment and any external reporting, because it includes the salaries, tools, agencies and sales compensation that working CAC deliberately omits. The error to avoid is using working CAC to argue that customer acquisition is profitable, since it excludes real costs by design.

Should founder time be included in startup CAC calculations?

Yes, where founders are doing meaningful acquisition or sales work, which in early-stage companies is often most of it. A CAC that excludes the most expensive person in the business selling four days a week is not measuring acquisition cost, and it produces a number that cannot survive the transition to a hired sales team. Value the time at a defensible replacement rate, disclose the assumption, and keep the treatment consistent so that the figure remains comparable as the business scales.

How does incremental CAC differ from attributed CAC in practice?

Attributed CAC divides channel spend by the conversions an attribution model assigns to that channel, which allocates credit by a rule rather than measuring cause. Incremental CAC divides spend by the customers who would not have been acquired without it, which requires a holdout, a geographic test or another credible counterfactual. The gap between the two is largest in conversion-adjacent channels such as branded search and retargeting, where the audience has already demonstrated intent, and it is precisely those channels that look most efficient on attributed reporting.

What are the most common CAC calculation errors?

The most damaging are changing the definition between periods and reporting the movement as performance, mixing blended and new-customer figures, comparing revenue-based lifetime value against fully loaded CAC, ignoring the sales-cycle lag so that spend and customers cover different acquisition events, and counting database records rather than deduplicated people. Each of these can move the reported number substantially without anything changing in the business, which is why the definition should be settled and documented before anyone interprets a trend.

Can a lower channel CAC be used to reallocate budget?

Not by itself. A channel-attributed figure describes assigned conversion credit, whereas budget allocation requires evidence about incremental customers or contribution. Published research analysing hundreds of large-scale advertising experiments has found that observational methods often fail to recover the causal effects that randomised tests measure, even with unusually rich data. Use an appropriate experiment, a marketing-mix model or another causal design first, then apply the practical reallocation process in our guide to reducing CAC without reducing customer quality.

Is there a CAC payback period every business should target?

No. Acceptable payback depends on your cost of capital, funding position, growth rate, contract structure, gross margin and how confident you are in the retention assumption behind the forecast. Widely quoted medians generally describe whoever happened to be in a particular vendor or consultancy sample, and rarely disclose the population, definitions or period behind the figure. Set your own threshold from your cash position, calculate payback on contribution rather than revenue, and state the cohort maturity whenever you publish it.

References

Ron Kohavi, Roger Longbotham, Dan Sommerfield and Randal M. Henne. Springer Nature. “Controlled experiments on the web: survey and practical guide.” Data Mining and Knowledge Discovery, Volume 18, 140-181. Published online 30 July 2008; issue date February 2009. DOI: 10.1007/s10618-008-0114-1. link.springer.com/article/10.1007/s10618-008-0114-1 Source classification: independent research; all authors were affiliated with Microsoft when published.
Association of Chartered Certified Accountants. “Cost-volume-profit analysis.” ACCA technical article, exam support resources. No publication date shown; accessed 27 July 2026. accaglobal.com cost-volume-profit analysis Source classification: official professional accountancy guidance; independent of any marketing vendor.
Thomas Blake, Chris Nosko and Steven Tadelis. The Econometric Society / Wiley. “Consumer Heterogeneity and Paid Search Effectiveness: A Large-Scale Field Experiment.” Econometrica, Volume 83, Issue 1, 2015, 155-174. DOI: 10.3982/ECTA12423. onlinelibrary.wiley.com/doi/10.3982/ECTA12423 Source classification: independent peer-reviewed research; eBay Research Labs affiliations and the eBay experimental setting are disclosed.
Brett R. Gordon, Florian Zettelmeyer, Neha Bhargava and Dan Chapsky. INFORMS. “A Comparison of Approaches to Advertising Measurement: Evidence from Big Field Experiments at Facebook.” Marketing Science, Volume 38, Issue 2, 2019, 193-225. Published online 4 April 2019. DOI: 10.1287/mksc.2018.1135. pubsonline.informs.org/doi/10.1287/mksc.2018.1135 Source classification: independent peer-reviewed research with two Facebook-affiliated co-authors.
Brett R. Gordon, Robert Moakler and Florian Zettelmeyer. INFORMS. “Close Enough? A Large-Scale Exploration of Non-Experimental Approaches to Advertising Measurement.” Marketing Science, Volume 42, Issue 4, 2023, 768-793. Published online 7 November 2022. DOI: 10.1287/mksc.2022.1413. pubsonline.informs.org/doi/10.1287/mksc.2022.1413 Source classification: independent peer-reviewed research. Interested-party disclosure: the study analyses experiments run on Meta advertising infrastructure, and one co-author was affiliated with Meta Ads Research.
Randall A. Lewis and Justin M. Rao. Oxford University Press. “The Unfavorable Economics of Measuring the Returns to Advertising.” The Quarterly Journal of Economics, Volume 130, Issue 4, November 2015, 1941-1973. Published 6 July 2015. DOI: 10.1093/qje/qjv023. academic.oup.com/qje/article/130/4/1941/1914592 Source classification: independent peer-reviewed research.
Garrett A. Johnson, Randall A. Lewis and Elmar I. Nubbemeyer. American Marketing Association / SAGE. “Ghost Ads: Improving the Economics of Measuring Online Ad Effectiveness.” Journal of Marketing Research, Volume 54, Issue 6, 2017, 867-884. First published online 1 December 2017. DOI: 10.1509/jmr.15.0297. journals.sagepub.com/doi/10.1509/jmr.15.0297 Source classification: independent peer-reviewed research; the method was developed around online advertising platforms and demonstrated using a retailer campaign.
Sunil Gupta, Donald R. Lehmann and Jennifer Ames Stuart. American Marketing Association. “Valuing Customers.” Journal of Marketing Research, Volume 41, Issue 1, February 2004, 7-18. DOI: 10.1509/jmkr.41.1.7.25084. doi.org/10.1509/jmkr.41.1.7.25084 Source classification: independent peer-reviewed research; empirical application covering five companies. The authors state that the cost of improving retention was not included in the model.
Sungwook Min, Xubing Zhang, Namwoon Kim and Rajendra K. Srivastava. American Marketing Association. “Customer Acquisition and Retention Spending: An Analytical Model and Empirical Investigation in Wireless Telecommunications Markets.” Journal of Marketing Research, Volume 53, Issue 5, October 2016, 728-744. DOI: 10.1509/jmr.14.0170. journals.sagepub.com/doi/10.1509/jmr.14.0170 Source classification: independent peer-reviewed research; dataset covers 41 countries, 1999 to 2007.
Daniel M. McCarthy and Russell S. Winer. Springer Nature. “The Pareto Rule in Marketing Revisited: Is It 80/20 or 70/20?” Marketing Letters, Volume 30, Issue 2, June 2019, 139-150. DOI: 10.1007/s11002-019-09490-y. doi.org/10.1007/s11002-019-09490-y Source classification: independent peer-reviewed empirical research.
Tim Ambler. American Marketing Association. “Maximizing Profitability and Return on Investment: A Short Clarification on Reinartz, Thomas, and Kumar.” Journal of Marketing, Volume 69, Issue 4, October 2005, 153-154. DOI: 10.1509/jmkg.2005.69.4.153. journals.sagepub.com/doi/abs/10.1509/jmkg.2005.69.4.153 Source classification: independent peer-reviewed academic clarification.