The metrics that connect spend to results

Ad dashboards offer dozens of numbers. In practice, campaigns are steered by a short list: a few cost metrics, a few rate metrics, and one or two that measure business outcomes. Here is each with its exact formula and the mistake people make reading it.

Published by AdFire Digital · Updated July 26, 2026 · Educational guide, independently written

Cost metrics

Rate metrics

Outcome metrics

How the metrics relate

The chain runs: impressions → clicks → conversions → revenue, and each arrow is one of the rates above. That means the outcome metrics decompose into the others:

• CPA = CPC ÷ conversion rate — you can lower CPA by buying cheaper clicks or by improving the page.
• CPC = CPM ÷ (CTR × 1000) — on CPM-bought media, better creative literally lowers your click cost.
• ROAS = (conversions × average order value) ÷ spend — raising order value improves ROAS without touching the ads.

A worked example

You spend $1,000, serve 200,000 impressions, get 3,000 clicks and 60 sales worth $2,400:

• CPM = 1000 ÷ 200,000 × 1000 = $5.00
• CTR = 3,000 ÷ 200,000 = 1.5%
• CPC = 1000 ÷ 3,000 = $0.33
• Conversion rate = 60 ÷ 3,000 = 2%
• CPA = 1000 ÷ 60 = $16.67
• ROAS = 2,400 ÷ 1,000 = 2.4

Whether 2.4 is good depends entirely on margin: with a 40% margin, break-even ROAS is 2.5 — this campaign is just below water and needs a cheaper click, a better page, or a higher order value.

Turn unit economics into decision limits

A dashboard cannot tell you whether a CPA is acceptable until the business defines the value of a conversion. For ecommerce, start with contribution margin after product cost, payment fees, fulfillment, expected returns and variable service cost. For leads, multiply the value of a closed customer by the observed lead-to-sale rate, then subtract the variable cost of handling the lead. The result is a ceiling, not a target: spending exactly the full expected value leaves no room for overhead, uncertainty or profit.

Suppose an accepted lead closes at 20% and each closed job contributes $500 before advertising. The expected contribution per accepted lead is $100. A $70 cost per accepted lead leaves $30 before overhead and uncertainty; a $120 cost loses money under those assumptions. If the campaign dashboard reports $40 per form but only half the forms are accepted, the operational cost is $80 per accepted lead. The qualification step changes the decision.

Attribution changes credit, not the underlying sale

Attribution models distribute conversion credit across touchpoints. Last-click gives the final eligible click the credit; data-driven attribution estimates the contribution of interactions from the advertiser's path data. Changing the model can change reported channel conversions without changing the number of orders in the business system. Therefore, compare models to understand journeys, but reconcile total outcomes against a source of truth such as completed orders or accepted leads.

Keep model names and windows beside every report. “Paid social generated 40 conversions” is incomplete if one report uses a seven-day click window and another includes view-through conversions over a different period. A clean comparison states the conversion definition, attribution model, window, timezone and whether modeled outcomes are included.

Account for delay, noise and sample size

Recent results are often incomplete because conversions happen after the click and platforms may update modeled or attributed outcomes later. Daily CPA also swings when conversion counts are small. A campaign with one conversion yesterday and three today did not necessarily become three times better; random timing can produce the difference.

Use a consistent review window that matches the buying cycle, compare like-for-like weekdays when seasonality matters, and annotate material changes. For tests, preselect the primary metric and avoid stopping the moment a preferred variant moves ahead. Statistical confidence is not created by a platform label alone: instrumentation quality, representativeness and practical business effect still matter.

Compare marginal performance before scaling

Average CPA or ROAS describes the spend already purchased. A budget decision needs the likely result of the next unit of spend. As a campaign expands, it may move beyond the most responsive queries, audiences or placements, so marginal CPA can rise even while the historical average remains acceptable.

Scale in measured steps and compare the additional spend with the additional qualified outcomes over a window that accounts for delay. If spend rises by $1,000 and contribution rises by only $700, the increment failed even if the blended lifetime report still looks profitable. This does not prove advertising caused the change—seasonality and other channels may move too—but it is a stronger budget diagnostic than applying the old average ROAS to a new spend level.

Further reading on this site

See the terms behind these metrics in the glossary, or go back to the channel guides: search, social, programmatic.