Reported CAC can point budget in the wrong direction

A campaign reporting a $15 customer acquisition cost, or CAC, can deserve less budget than one reporting $60. Consider an illustrative comparison between branded search and paid social: people searching for a company by name may already intend to buy, while a social campaign may reach people who would otherwise never become customers. If advertising added only a small share of the branded-search customers, its cost per additional customer could be several times its reported CAC.

Giving more money to the campaign with lower reported CAC can therefore reward it for capturing existing demand while reducing investment in additional acquisition. For a marketing leader defending spending to finance, an easy-to-explain dashboard result can lead to a difficult budget decision. A more useful question is: “How many customers would we lose without this spend?”

Measure the customers advertising adds

Answering that question requires understanding how a campaign gets credit for a customer. CAC measures acquisition spending per customer, while return on ad spend, or ROAS, measures revenue relative to advertising spending. Conventional platform reports calculate these metrics using attribution: rules that assign conversions, completed actions such as purchases, to advertising after an ad interaction or exposure. A campaign can get that credit even when the customer would have bought anyway.

The branded-search example shows the problem because an ad can appear just before a purchase someone already intends to make. A customer may search for the company, click its paid result and buy, yet would have clicked its organic result if the ad had been absent. A high conversion rate and low reported CAC can therefore reflect access to customers already ready to purchase.

To estimate how much advertising changes behavior, incrementality testing compares observed outcomes with a counterfactual: what would have happened without the campaign. A holdout test creates that comparison by withholding advertising from a group and measuring the difference in outcomes. The estimated difference is lift, meaning the additional conversions caused by the advertising. Incremental CAC divides spending by the additional customers generated.

This measurement matters wherever campaigns reach people close to buying. Brand search, retargeting (advertising to people who have previously interacted with a business) and some affiliate programs can report efficient acquisition while capturing substantial existing demand. Campaigns reaching new audiences can appear more expensive while adding more customers. These tendencies suggest useful testing questions; paid social, paid search and affiliate activity each need campaign-specific evidence.

Campaign-specific evidence also has limits that matter for the budget decision. Attribution can credit purchases that would have happened through organic search or another route, while incrementality testing needs a sound comparison and enough activity to detect a meaningful effect. Even a useful estimate of a campaign’s current contribution leaves a further question: whether additional spending will acquire enough additional customers at an acceptable cost. Answering it requires another spending comparison and an agreed view of customer value.

Okoone experts
LET'S TALK!

A project in mind?
Schedule a 30-minute meeting with us.

Senior experts helping you move faster across product, engineering, cloud & AI.

Please enter a valid business email address.

Search tests can support opposite budget decisions

Two reported Google search examples show why the experiment must determine the direction of a budget change. In a Google brand search test, the findings supported reducing spending by roughly 90%, while customer acquisition did not change significantly. Paid ads had been getting credit for purchases that would otherwise have come through organic search. When paid spending fell, organic results absorbed that demand.

Organic substitution explains why paid conversions could fall while acquisition stayed largely unchanged. Customers still reached the business, but their purchases no longer passed through a paid click that received advertising credit. For the budget owner, the useful outcome was total customer acquisition across paid and organic routes. The finding supported a substantial reduction under the tested conditions, while “did not change significantly” leaves room for an effect the experiment could not clearly distinguish.

A separate non-brand search example supported increasing investment. A Google conversion lift study reported 55% conversion lift, giving the business a reason to accept a higher CAC and expand spending. New-customer growth subsequently increased in a way consistent with the finding.

Together, these outcomes show that the same broad channel can support different allocations under different conditions. Google sells advertising and benefits commercially when advertisers continue or expand spending, an interest relevant when interpreting a Google conversion lift study. Yet the brand and non-brand examples led to opposite decisions because the measured contribution differed. Neither result establishes how another search campaign will perform.

A separate account from an unidentified observer extends the problem beyond search: two channels had nearly identical conventional performance figures, yet their incremental results called for opposite budget decisions. Similar dashboard numbers can therefore conceal different contributions to acquisition. Measuring those contributions gives the business a basis for allocation, but the value of the additional customers determines how much it should pay.

Set the incremental-CAC decision rule before testing

Customer value turns a measured effect into a spending decision. First, compare outcomes with and without the campaign, estimate how many additional customers the spending caused, and divide spending by those customers to calculate incremental CAC. The team then compares that cost with its agreed customer-value threshold. Depending on the result, investment can increase, remain steady or decrease.

The experiment must match the spending decision under consideration. Testing whether an existing campaign adds customers answers a different question from testing whether a larger budget adds enough customers to justify expansion. Depending on the question, channel, available data and business objective, the team might use a platform conversion lift study, a geographic holdout comparing markets with different advertising exposure, or another controlled experiment. The comparison needs to represent the change the business is considering.

The choice of experiment determines the preparation needed for a useful result. Test duration and sample size affect the ability to detect a meaningful difference, while overlap between the advertising exposure of test and control groups can weaken the comparison. A short test or few conversions may leave substantial uncertainty about the campaign’s contribution. The team needs to account for these constraints when planning the experiment and interpreting its results.

Interpreting an effect also requires a separate judgment about its economics. Statistical significance indicates that a result meets an evidence threshold under the test’s assumptions; it does not establish profitability. A campaign can cause additional purchases while acquiring customers at an unacceptable cost. The spending rule must therefore specify what the business is willing to pay before the findings arrive.

For example, a hypothetical business that values a new customer at $200 over their lifetime could agree on three thresholds. An incremental CAC under $120 would support increasing investment; an incremental CAC between $120 and $180 would support maintaining it; and an incremental CAC above $180 would support reducing it. These are illustrative decision rules, rather than universal benchmarks or actual company results.

Agreeing on those rules in advance gives marketing and finance a shared basis for acting on the experiment. It also reduces the opportunity to reinterpret findings to protect a preferred campaign or budget. Once the thresholds are set, the practical challenge is to design and run a test that can inform the decision.

AI can help teams prepare and run experiments

Running that process can stretch a smaller company’s analytical capacity because design, data preparation, monitoring and interpretation all take time. Some teams therefore forgo testing, while large companies can draw on data-science teams and dedicated vendors. AI-enabled tools can reduce the work smaller teams must do themselves. Their usefulness depends on whether the team can inspect the work and judge whether the experiment answers its business question.

Preparation offers a substantial opportunity to save time. For a geographic holdout, general-purpose large language models, or LLMs, such as ChatGPT or Claude can help prepare data and examine historical regional sales patterns; these AI systems generate and analyze text and code. The aim is to identify markets whose past behavior makes them suitable test and control groups. Comparing those groups provides a basis for estimating what might have happened in the test markets without the advertising change.

Once candidate groups are available, the same tools can help estimate the group sizes and duration needed to detect meaningful lift, then write and explain analysis code. The explanation matters because someone needs to inspect how the calculations work. ChatGPT and Claude are commercial AI products whose providers benefit from their adoption, so their use should be judged against the analytical work they help the team complete.

Other tools support more specialized parts of the process. Teams seeking greater rigor can use Meta’s open-source GeoLift, while dedicated platforms such as Haus incorporate steps from the testing workflow. Meta has a commercial interest in advertising measurement through its advertising business, and Haus benefits from demand for measurement platforms. With any of these options, the team remains responsible for checking the comparison, assumptions and calculations.

After setup, monitoring helps identify events that could change what the comparison means. A regional promotion, for example, might increase purchases in a control market intended to represent outcomes without the tested campaign. AI monitoring may flag that change early enough for the team to assess its effect on the experiment. The conversion records could all be accurate while the promotion still weakens the control group’s usefulness.

A tracking failure creates a different problem because recorded conversions stop representing actual activity. If an error suppresses reporting for several days, the apparent decline may reflect missing records rather than changed customer behavior. AI-assisted review may detect the anomaly while the experiment is running. Early detection can give the team time to fix the fault and may prevent the need to repeat the test.

Once the experiment finishes, AI can prepare an initial readout of estimated lift and its confidence range, which expresses uncertainty around the effect estimate. It can calculate incremental CAC and compare the result with the spending thresholds agreed before testing. It can also examine markets, audiences or campaigns to identify segments worth studying in a later experiment. These breakdowns can guide further testing, while a new allocation requires evidence suited to that decision.

The readout must then explain the result in terms finance and leadership can use. AI can help describe the estimated number of additional customers, their acquisition cost and the spending rule the result meets. People still need to check the calculations, validate the generated analysis and judge whether the proposed allocation follows from the findings. Responsibility for the decision remains with the business.

Human review is especially important when an AI system offers a causal explanation for events that merely occurred together. If conversions rise during the week a campaign launches, the system may credit the campaign even when seasonality, a pricing change, competitor activity or a measurement error explains the increase. A well-designed experiment provides the causal evidence that AI can help prepare, examine and communicate. Minimum-volume requirements, risks from overlapping group exposure and calculation checks remain part of the work.

Test the spending level the business wants to approve

Even a well-run experiment estimates performance at a particular spending level and under particular conditions. A campaign with acceptable incremental CAC can become uneconomic as its budget grows, so a passing result supports a limited next allocation. The business then needs to measure whether the additional spending continues to acquire customers within its agreed cost threshold.

Market changes can also alter a campaign’s contribution, including the branded-search campaign considered at the outset. Brand search can add customers when competitors bid on the company’s name because the available alternatives have changed. An earlier finding that organic results absorbed demand cannot settle the campaign’s value under those conditions. The changed competitive environment creates a new testing question.

When findings remain inconclusive, the team should continue testing and accumulate evidence over time. As spending or market conditions change, it should repeat the measurement while carrying forward the agreed customer-value rules.

Key executive takeaways

  • Low CAC can misdirect budgets: Marketing budget owners need to assess how many customers advertising adds. A campaign capturing existing demand can report a low CAC while contributing little additional acquisition.
  • Measure incremental customers: Marketing analytics teams need a credible comparison of outcomes with and without advertising. Dividing spending by the estimated additional customers gives finance a more useful measure for allocation.
  • Let campaign evidence determine spending: A reported brand-search test supported cutting spending roughly 90%, while a separate non-brand study supported expansion. Budget owners need campaign-specific evidence before increasing or reducing investment.
  • Agree on spending rules before testing: Marketing and finance need shared incremental-CAC thresholds tied to customer value. Set those thresholds before results arrive, and design the experiment around the spending change under consideration.
  • Use AI with analytical oversight: AI can help analysts prepare data, design tests, flag anomalies and draft findings. The business remains responsible for validating comparisons, calculations and causal claims before approving budget changes.
  • Retest as spending and conditions change: An acceptable incremental CAC supports a limited next allocation. Marketing teams need fresh evidence as budgets grow, competitors act or market conditions change.

Alexander Procter

November 18, 2026

11 Min

Okoone experts
LET'S TALK!

A project in mind?
Schedule a 30-minute meeting with us.

Senior experts helping you move faster across product, engineering, cloud & AI.

Please enter a valid business email address.