An ad with a low CPA can look like a winner after one conversion. An ad with a high CPA can look like a failure before its conversions finish arriving. A table sorted by CPA puts both in order without telling you how much to trust that order.
A spend–CPA map gives the review more context. Put spend on the horizontal axis and attributed cost per acquisition (CPA) on the vertical axis. Then ask two questions: does this ad meet our economics, and do we have enough evidence to act?
A LinkedIn post attributed to Curtis Howland proposes this framework: review efficiency alongside spend, then classify ads into scale, test, and stop candidates. The operating rules below extend that idea with evidence checks and business constraints; they are recommendations, not demonstrated results from the post.
Start with comparable ads
Before drawing the chart, define the conversion being purchased. A lead, first purchase, and repeat purchase represent different outcomes. Their CPAs should not share a target simply because they appear in the same report.
Group ads by the differences that materially affect the decision: conversion event, customer type, offer, market, or audience. Keep the reporting window and attribution settings consistent within each group.
Avoid splitting the data so finely that every ad becomes its own category. Create a separate group when the economics or measurement conditions justify a different decision rule.
For each ad, collect:
- Spend and attributed conversion count.
- Attributed CPA, where conversions exist.
- Delivery dates and the reporting cutoff.
- The applicable target CPA and stopping threshold.
- Recent changes that could affect interpretation.
Ads with zero conversions need an explicit status. Their CPA is undefined. Keep them in a separate queue assessed by spend, elapsed time, and conversion delay; do not let them disappear from the review because the chart cannot plot them.
Separate economic thresholds from evidence requirements
The target CPA describes what the business can afford for the defined outcome. Set it using the economics relevant to that outcome, with assumptions about customer value made explicit. Platform-attributed CPA alone does not establish profitability or incremental acquisition.
The stopping threshold describes how much inefficiency the team is willing to tolerate during a test. It is a budget policy, not automatically a statistical conclusion.
For example, choosing a stopping line at a multiple of target CPA can limit exposure. It cannot, by itself, establish a particular confidence level that the ad will never work.
Write the evidence requirement separately. Specify how you will account for:
- Conversion volume and volatility.
- The time conversions typically take to appear.
- Maximum test spend or duration.
- What happens when the test reaches its limit without a clear result.
A minimum conversion count can help qualify a scale candidate. It cannot be the only stopping rule: an ad that produces no conversions would never qualify for a decision. Pair evidence requirements with a bounded testing budget.
Stopping an inconclusive test because it exhausted that budget is a valid operating choice. Record it as inconclusive, rather than claiming the creative was conclusively bad.
Turn the map into a decision queue
A logarithmic spend axis can make ads with very different budgets easier to see together. It does not make their CPAs more reliable. Keep conversion counts and evidence status visible beside each point.
Use three action categories.
Scale candidate: CPA meets the target, the evidence requirement is satisfied, and there is a plausible path to deploying more budget. Check downstream customer quality and delivery constraints before changing spend.
Continue testing: The result remains uncertain or falls within the permitted testing range. Every continuation needs a remaining budget, a next review point, and a reason more evidence could change the decision.
Stop candidate: Performance remains outside the permitted range after the relevant evidence and delay checks, or the test has reached its predefined exposure limit. Distinguish those two reasons in the decision record.
This prevents the test category from becoming permanent storage for ads nobody wants to decide on.
The map ranks candidates for review. Where budgets are controlled at a broader campaign or group level, an ad-level recommendation still needs to be translated into an action at the level the team can control.
Make each budget decision reviewable
For every action, record the data cutoff, evidence status, recommendation, actual decision, and next review date. Add a short explanation when the operator overrides the recommendation.
Useful explanations name a condition: conversions are still maturing, the offer changed, customer quality deteriorated, or a broader budget constraint prevents scaling. “Looks promising” is difficult to evaluate later.
For scale candidates, define the size of the next change and the condition that would reverse it. Do not treat a universal percentage increase as a guarantee of stable performance. The appropriate change depends on the campaign structure, available demand, and the cost of being wrong.
After a change, review the new period separately from the earlier result. A strong historical average can obscure weaker performance at higher spend. Even then, a before-and-after comparison is descriptive; it does not isolate the budget change from everything else that happened.
Validate the policy before relying on it
Start with historical decision snapshots. For each snapshot, use only the information that would have been available at that time. Apply the proposed rules, then compare their recommendations with later observations.
This can expose obvious problems: stopping before conversions mature, repeatedly extending weak tests, or labeling tiny samples as scale candidates.
Historical validation has a limit. Once an ad was stopped, you generally cannot observe what it would have done with continued spend. Later-attributed conversions can reveal reporting delays, but they do not fully reveal the opportunity cost of stopping. Treat false-stop estimates cautiously.
Next, run a bounded prospective pilot against the current review process. Where a credible comparison is feasible, keep creative supply, measurement windows, and targeting conditions as comparable as possible. Choose the business outcome and guardrails before starting.
A policy that lowers reported CPA while sharply reducing useful acquisition may be too conservative. Track acquisition volume and customer value alongside efficiency. If the design cannot estimate incremental outcomes, describe the result as attributed performance rather than causal lift.
Evidence and limitations
The LinkedIn post supports the spend-versus-CPA visualization and the proposed scale, test, and stop classification. It does not establish that this policy improves profitability, incremental acquisition, or decision accuracy.
Its suggested threshold multiples, fixed budget increments, revenue concentration benchmarks, and confidence claims lack sufficient supporting methodology in the available source material. They should not be adopted as universal rules.
The evidence requirements, decision log, bounded testing policy, and validation process here are operator recommendations derived from the framework. Their value needs to be tested in the account where they will be used. The chart can make decisions more explicit; it cannot resolve attribution uncertainty or prove that an ad caused additional demand.
Source basis
- A LinkedIn post attributed to Curtis Howland proposing an ad-level spend-versus-CPA decision map.
- Editorial analysis of the framework’s evidence requirements, attribution limitations, and unsupported benchmark claims.