In this guide
Your next steps
- Define the metric and denominator.
- Calculate all ratios from the same data window.
- Choose the business outcome before the review.
- Use the remaining metrics to diagnose differences.
Different metrics answer different questions
CTR can rise while purchase CPA gets worse. ROAS can jump after one large order. Start with the business outcome, then inspect auction cost, response, purchase rate and order value rather than choosing whichever metric looks best.
1. Use explicit definitions
Here CPA means cost per attributed purchase, CTR uses link clicks, and purchase rate uses impressions. Keep period, currency, event source and attribution settings consistent. An unavailable denominator is not zero; with zero purchases, a finite purchase CPA is undefined.
| Metric | Calculation | Question |
|---|---|---|
| Link CTR | Link clicks ÷ impressions × 100 | Did the ad produce a click response? |
| Purchase CPA | Spend ÷ attributed purchases | What did each reported purchase cost? |
| ROAS | Purchase value ÷ spend | How much reported revenue accompanied spend? |
| Purchase rate | Purchases ÷ impressions × 100 | How often did exposure accompany a purchase? |
| Post-click conversion rate | Purchases ÷ link clicks × 100 | How do purchases compare with clicks? |
2. Check the arithmetic with one example
For $1,000 spend, 100,000 impressions, 1,000 link clicks, 40 purchases and $3,200 purchase value: CTR is 1%, CPA is $25, ROAS is 3.2x and purchase rate is 0.04%. Purchases per link click is 4%, a different calculation.
CPM is $10 and average order value is $80. With rates as decimals: CPA = CPM ÷ (1,000 × purchase rate). ROAS = purchase rate × average order value ÷ cost per impression. These identities explain arithmetic drivers, not causation.
3. Match the metric to the decision
For an opening test, CTR may be a planned intermediate outcome with sales guardrails. For sales efficiency, prioritize purchases and business economics. For incremental impact, use an appropriate experiment rather than a platform ratio alone.
- CTR: diagnose response; curiosity and accidental clicks may not become customers.
- CPA: compare the cost with the value and margin of the acquired outcome. Scaling can change marginal cost.
- ROAS: inspect purchase count and order value. Attributed revenue is not profit or incremental revenue.
- Purchase rate: assess sales response per impression, while checking attribution, repeated exposure and sparse events.
4. Keep the economic and statistical limits visible
A higher-priced bundle can lift ROAS without improving purchase response. In a simplified model with 40% contribution margin before ads and relevant variable costs included, break-even ROAS is 2.5x. Your costs and customer-value assumptions may differ.
Allow conversion delay. View-through attribution can make purchases per click an imperfect click-only funnel measure. Cost and revenue totals alone do not supply all the variance inputs needed to call a difference statistically established.
Read complementary metrics in scale190
Overall and Creatives place performance measures together; comparisons add supporting evidence. Read each label and denominator: an interface measure labelled CAC is not automatically identical to purchase CPA.
The implementation distinguishes supported proportion tests from aggregate cost or return differences that cannot be tested from the available totals alone. Use the review to decide the next question, not to treat a score as a profit guarantee.
Sources & further reading
Worked examples are hypothetical. References provide further methodological background.
- Meta Blueprint: A/B tests and conversion lift studiesDifferent study designs answer different questions.
- NIST: inference for proportionsEvent rates, denominators and model assumptions.
