Performance marketing metrics: which metric drives which decision

Performance marketing metrics are not good or bad on their own. Each one answers a specific decision: raise the budget, replace a creative, fix the landing page, move a bid target or cut a channel.

Cost per thousand impressions (CPM), click-through rate (CTR) and cost per click (CPC) start one chain. Conversion rate, cost per acquisition (CPA) and return on ad spend (ROAS) finish it. A break in one link shows up in every link after it.

So pick the decision first, then read the metric that answers it. Set the thresholds from your own margin, not from an industry average.

This guide orders the metrics by the decisions they drive rather than as a glossary. The formulas live in the site's calculators, so they are not repeated here. Instead, one hypothetical ecommerce example shows how the metrics read together.

Most performance marketing decisions fall into a handful of types. The table below shows which metric each decision rests on and when to act. The thresholds are left blank on purpose, because your margin sets them in the next section.

Some of these metrics come from the ad platform and some from your own sales data. CPM, CTR and CPC are what the platform measures. Customer acquisition cost (CAC) and lifetime value (LTV) come from your order and customer records. The bigger the decision, the more it should rest on your own data.

The answer to "what is a good ROAS?" depends on product margin. A ROAS that is profitable for a brand selling at a 60% gross margin is a loss for a brand selling at 25%. So the first job is to find what one order contributes before advertising. You can pin down gross margin with the gross margin calculator and average order value (AOV) with the AOV calculator.

Hypothetical example: an ecommerce brand starts from its order value after sales tax and returns. Taking shipping, payment fees and packaging out of gross profit leaves the contribution per order. The table below holds the values every example in this guide uses.

Contribution per order is the most the brand can spend on ads per order. In other words, break-even CPA equals contribution. If the brand wants to keep part of it as profit, target CPA falls and target ROAS rises. The ROAS calculator page covers the calculation in detail.

Taxes and returns are often missed. An ad platform usually sees the gross order amount at checkout as the conversion value. If refunded or cancelled orders and tax stay in that value, reported ROAS looks better than real profitability. The order value behind your threshold and the conversion value you send to the platform should use the same definition.

CPA looks like one number, but it is the result of three metrics. CPM sets the cost of reaching people. CTR sets how many impressions become clicks, and conversion rate sets how many clicks become orders.

Google defines CTR as clicks divided by impressions. It defines conversion rate as the average number of conversions per ad interaction. Finding the broken link when CPA rises also tells you which team should act.

The same hypothetical brand pays a $6 CPM, which gives a starting cost per click of $0.50. The campaign is above break-even but misses the target. The table below shows how two different failures produce a similar CPA increase.

In both cases CPA passes the $25 break-even point, but the fix differs. In the first case CTR fell while conversion rate held. The problem sits in the ad itself, meaning the creative or the targeting. Lowering the bid will not help here, but a new creative will.

In the second case CTR held and conversion rate fell. The ad still earns clicks, but visitors do not buy on the site. Look at the landing page, stock, price or checkout.

Working through that step with the method in the conversion rate optimization guide is faster than changing ad settings. If CPA rises while CTR and conversion rate hold, the cause is usually a higher CPM, meaning more competition in the auction.

For budget decisions, read marginal return, not average ROAS

Average ROAS shows the return on all spend. A budget decision depends on the return of the last dollar added. As budget grows, the platform reaches the easiest buyers first and harder ones later. That is why average ROAS usually falls as budget rises.

In the hypothetical example, the brand tests three budget levels on the same campaign. The profit column is orders times the $25 contribution, minus ad spend.

At the highest budget, average ROAS is still above the 3.2 break-even value. Someone reading the average would keep raising the budget. Yet profit falls after $4,000.

The reason is marginal ROAS, the return on the extra spend alone. The first $2,000 step brings $6,800 in extra revenue, a marginal ROAS of 3.4. The second step brings only $4,800, so marginal ROAS drops to 2.4.

If marginal ROAS on extra spend is above break-even, more budget grows total profit. If it is below, more budget shrinks it.

To measure this, change budget in steps and watch each step long enough. When seasonality or a sale period changes at the same time, it gets hard to tell whether the change came from budget or the calendar.

Set the bid target to match your data volume

Target CPA and target ROAS are where the thresholds from your margin enter the platform. Google's Target CPA page says it will try to keep cost per conversion equal to the target you set. That is an average, not a ceiling. Some conversions cost more than the target and some cost less.

Target ROAS needs values assigned to conversions first. Google's Target ROAS page lists at least 15 conversions in the past 30 days for Search and Shopping campaigns. To evaluate Smart Bidding results, the Smart Bidding page recommends periods of a month or longer with at least 30 conversions (50 for Target ROAS).

Meta's learning phase raises the same question. Meta's learning phase page says the phase usually ends after about 50 results in the week after the last significant edit. It adds that CPA is usually higher during this time. So cutting a campaign on CPA while it is still learning means deciding on a number that has not settled.

In practice, this means not moving targets often. Each significant edit can restart learning. In an account with few weekly conversions, merging campaigns can supply the volume a target-based bid needs. So can choosing a more frequent upstream conversion, such as add to cart.

Impression share and frequency show the ceiling on scale

Before raising budget, check whether the extra spend has anywhere to go. In search campaigns, impression share tells you. Google's impression share data page reports share lost to budget and share lost to ad rank separately.

If share lost to budget is high, there are searches the extra budget can reach. If share lost to rank is high, the problem is the bid or ad quality, not the budget. Google also notes that the lost-to-budget column is not compatible with the Maximize conversions strategy.

In social channels, frequency shows the ceiling. Meta's frequency page defines frequency as impressions divided by reach. It adds that if performance drops as frequency rises, the audience may be experiencing ad fatigue. If CTR falls while frequency rises, you need a new creative or a new audience, not more money for the same one.

CPA measures the cost of a conversion in a campaign. Customer acquisition cost (CAC) measures the full cost of winning a new customer. The platform cannot always tell whether an order came from a returning customer or a first-time buyer. That is why CAC should come from your own customer records, for example with the CAC calculator.

A CAC above first-order break-even can still be profitable if customers buy again. First-order ROAS does not answer that question, but the ratio of LTV to CAC does.

The LTV:CAC ratio shows whether acquisition pays off over time. The CAC payback period shows how many months it takes to earn the spend back.

The two metrics answer different questions. Even with a high LTV:CAC ratio, a long payback period can strain cash flow. Use payback for short-term budget decisions and LTV:CAC for channel and product investment.

Ad platforms count conversions with their own attribution rules. Meta's Results page gives the default attribution setting as 7-day click, 1-day view and 1-day engagement conversion. In Google, attribution models set how much credit each ad interaction gets, and the default is data-driven attribution.

The same order can be counted by both Meta and Google. That is why the sum of channel-reported conversions often exceeds what your order system shows. Platform ROAS is useful for comparing campaigns inside one channel. To judge whether total spend is working, track total revenue divided by total marketing spend (MER) from your own data.

Attribution does not tell you whether a sale would have happened without the ad. An incrementality test answers that.

Google's Conversion Lift page describes incrementality experiments as tools to understand how well ads drive the action you care about at a point in time. It also notes that the feature is not available in every account. The test rests on a comparison group that does not see the ads. The control group guide explains how that group is set up.

Reading the metrics in the same order every week stops you from reacting to a random number. The order below runs from the widest decision to the narrowest.

1. Check total revenue and total marketing spend from your own data. If the ratio is below target, look for the problem in the channels.
2. Compare each channel's CPA and ROAS with the break-even and target values.
3. For a campaign off target, read CPM, CTR and conversion rate separately and find the broken link.
4. If you are considering a budget change, check marginal ROAS and impression share or frequency.
5. Hold off on decisions for campaigns still in learning or without enough conversion volume.
6. Once a month, update CAC, LTV:CAC and payback period from customer records.

The point of this order is not to track every metric. It is to make each decision with the right one. When a number changes, first ask which decision it touches. If there is no answer, that number is a note for the week, not an alarm.