Performance marketing: the metric chain, measurement and budget decisions

Performance marketing is paid marketing that you buy, optimise and judge by a measurable result: a purchase, a sign-up, a lead. Every channel is held to two numbers, what one result costs and what it brings back, and budget moves toward the channels where that trade is best. Doing it well comes down to three skills: reading the chain of metrics from impression to lifetime value, measuring what your ads actually caused rather than what they were near, and moving money by marginal return. This guide covers all three and links to a deeper guide for each step.

Performance marketing is paid advertising that is bought, optimised and evaluated against a measurable outcome, such as a sale or a lead, so that each channel can be judged by its cost per outcome and its return on spend.

What is performance marketing, and how is it different from brand marketing?

Performance marketing is the part of your marketing that has to pay for itself in countable outcomes. Brand marketing builds awareness and preference that pay off later and are hard to tie to a single click. The difference is the measurement contract, not the channel: an Instagram campaign can be either, depending on what you hold it to.

Three things mark a campaign as performance marketing:

1. A defined outcome. One key event, such as a purchase or a qualified lead, is written down before the campaign starts.
2. A cost ceiling. You know the most you can pay for that outcome and still make money, usually from your margin.
3. Closed-loop tracking. Every paid link is tagged and every outcome is recorded, so cost and result can be joined per campaign.

The trap is treating everything as performance marketing because it is easier to report. A video campaign that lifts searches for your brand two weeks later will look like a loss in a click-based report. That is a measurement problem, and the sections below on lift tests and marketing mix modelling are how you solve it.

Which metrics make up the performance marketing chain?

Every paid campaign runs on one chain of numbers, and each link multiplies the next. Money becomes impressions (CPM), impressions become clicks (CTR), clicks become customers (conversion rate), customers become revenue (order value) and, over time, lifetime value. Return on ad spend is the product of the whole chain, which is why a weak link anywhere shows up as a bad ROAS.

A worked example of the chain

Take a hypothetical online store that spends $10,000 in a month on one social campaign.

• CPM is $8, so the campaign buys 10,000 ÷ 8 × 1,000 = 1,250,000 impressions.
• CTR is 1.2%, so 1,250,000 × 0.012 = 15,000 clicks, and CPC is 10,000 ÷ 15,000 = $0.67.
• Conversion rate is 2%, so 15,000 × 0.02 = 300 orders, and CPA is 10,000 ÷ 300 = $33.33.
• Average order value is $60, so revenue is 300 × 60 = $18,000, and ROAS is 18,000 ÷ 10,000 = 1.8.

Now find the lever. If the landing page lifts conversion rate from 2% to 2.5%, orders rise to 375, revenue to $22,500 and ROAS to 2.25, with no change to the ad. Raising order value from $60 to $70 with a bundle gives $21,000 and a ROAS of 2.1. A 10% cheaper CPM would add only 10% more of everything. The chain tells you where to spend your effort, and it is often not in the ad account.

How do you know if a performance campaign is profitable?

A campaign makes money on the first order when its ROAS is above your break-even ROAS, which is 1 divided by your gross margin. Below that line, every sale loses money until repeat purchases pay it back. So you need two checks: break-even ROAS for the first order, and LTV to CAC for the customer.

Continue the example. The store’s gross margin is 40%, so:

• Break-even ROAS = 1 ÷ 0.40 = 2.5
• Contribution from the month = $18,000 × 0.40 − $10,000 = $7,200 − $10,000 = −$2,800

At a ROAS of 1.8 the campaign loses $2,800 on first orders. That is not automatically a reason to stop. Suppose these customers buy 2.5 times in their first year at the same $60 order value. Their margin-based lifetime value is 2.5 × 60 × 0.40 = $60, against a CAC of $33.33, so LTV to CAC is 60 ÷ 33.33 = 1.8. The campaign recovers its cost within the year but leaves little room for overheads, and it only holds if the repeat rate is real. The customer lifetime value guide shows how to estimate that repeat rate from cohorts instead of assuming it.

Two rules keep this honest. Use contribution margin, after shipping, payment fees and discounts, not list-price margin. And use the revenue your store recorded, not the revenue the ad platform claims, because each platform counts the conversions its own ads touched and the totals across platforms usually add up to more orders than you shipped. The ROAS formula guide covers why platform ROAS runs high.

Which channels does performance marketing use?

Any channel that sells on a measurable action can be run as performance marketing. What changes between channels is how close they sit to purchase intent and how much of their effect a click can capture.

Search ads and marketplace ads usually look best in attribution reports because they sit right before the purchase. That makes them the place where some of the measured sales would have happened anyway, and the place where a lift test most often surprises people. Social and creator spend looks worst in click reports and is where marketing mix modelling earns its keep.

Tagging decides whether any of this is measurable. Every paid link needs consistent UTM tags; the UTM parameters guide gives a naming convention, and the website tracking guide covers the rest of the setup that turns clicks into recorded conversions.

How do you measure what your ads actually caused?

Use three methods, each for its own question. Attribution splits credit for conversions among the touchpoints on the path, which is good for daily optimisation. Lift tests compare an outcome against a holdout or a forecast without the ad, which proves whether one change worked. Marketing mix modelling estimates each channel’s contribution from spend and sales over time, which guides the split of the whole budget.

Attribution is where most teams start, and it is useful as long as you remember what it is. GA4 now offers three attribution models: data-driven, paid and organic last click, and Google paid channels last click; first click, linear, time decay and position-based were removed in November 2023 (Google). The guide to attribution models compares six models on one customer journey and shows where each one misleads.

A lift test is the step from correlation to cause. In Meta’s lift studies, people are randomised into a test group that sees your ads and a control group that does not, and the difference in conversions between the groups is the effect of the campaign (Meta). The incrementality testing guide covers holdout, geo and time-based designs and how to read the confidence interval.

Marketing mix modelling works at the level of the whole budget. Google’s open-source model, Meridian, can use experiment results as priors and produces response curves that show how return per unit of spend diminishes as spend grows (Google). The practical order is the same in most tools: attribution every day, a lift test for every large change, and a mix model once you have the history.

How should you split a performance marketing budget?

Move budget by marginal return, not average return. The average ROAS of a channel tells you how its whole budget did; the marginal ROAS tells you what the last increase earned, and that is the only number that matters for the next one. Keep adding to a channel while its marginal ROAS is above break-even, and stop when it falls below.

A hypothetical example with a 2.5 break-even ROAS:

Ranked by average ROAS, channel A looks like the winner and B looks marginal. Ranked by marginal ROAS, the last $5,000 on A lost money (1.5 is below 2.5) while the last $5,000 on B made money (2.8 is above it). Moving $5,000 from A to B is the profitable move, even though it lowers the blended ROAS headline.

This is exactly what a response curve shows, and Meridian’s documentation describes its budget optimiser in these terms: it shifts budget so that some channels lose incremental revenue and others gain more, and the optimised ROI is at least as high as the current one (Google). Without a mix model you can still estimate marginal return. Change one channel’s budget by 20% or more, hold it for two weeks, and measure the change in conversions with a lift test.

How do you test and improve campaigns without fooling yourself?

Change one thing at a time and size the test before you run it. Creative, landing page and offer tests are the cheapest way to move the chain from the worked example, but a test that stops early or runs on too little traffic produces confident noise.

Four habits prevent most bad calls:

1. Pick the metric first. Judge a landing page test on conversion rate or revenue per visitor, not on CTR, which the landing page cannot change.
2. Calculate the sample size. Decide the smallest effect worth detecting, then compute how many visitors each variant needs. The A/B test sample size guide walks through baseline, minimum detectable effect and power.
3. Do not peek and stop. Checking daily and stopping at the first significant result inflates false positives.
4. Let bidding settle. Automated bidding learns from conversion data; Google’s Target ROAS strategy, for example, predicts the value of each auction and sets bids to hit the average return you set (Google). Large target changes reset that learning, so give a new target time before you judge it.

What are the most common performance marketing mistakes?

Most losses in performance marketing come from measurement, not media buying. The same mistakes show up in nearly every account review:

• Optimising to the wrong event. Bidding on add-to-cart or page views because they have more volume teaches the platform to find people who browse.
• Trusting the sum of platform reports. Two platforms claiming the same order makes ROAS look higher than the bank account does.
• Ignoring margin. A ROAS target copied from someone else’s industry means nothing; break-even depends on your own margin.
• Judging brand and prospecting spend on last click. It will always look worse than retargeting and search, which harvest the demand it creates.
• Scaling on averages. Doubling the budget of the channel with the best average ROAS often doubles spend on its saturated part.
• Untagged links. Creator posts, channel posts and SMS links without UTM tags land in direct or referral and make the best channels invisible.

How do you measure performance marketing in AdPix?

AdPix is a web analytics platform that replaces Google Analytics 4: realtime, acquisition, engagement, conversion and e-commerce reports, plus attribution, lift testing and marketing mix modelling, with no data sampling and a free plan. For performance marketing, that puts the whole loop in one place: where each visit came from, which visits converted, how much revenue followed, and whether the spend caused it.

Where each part sits, per AdPix pricing:

• Free: realtime, acquisition and UTM campaign reports, user journeys, conversion funnels, order analysis and the AI assistant, up to 100,000 events a month.
• Starter (499,000 toman a month): adds online sales and revenue analysis, segments and audiences, invalid-traffic detection, marketing mix modelling and campaign lift measurement.
• Pro (2,990,000 toman a month): adds advanced campaign and channel analysis, true campaign impact measurement, and MMM for each channel’s share of sales.

Two details matter for Iranian media buyers. AdPix recognises local ad networks such as Yektanet, Tapsell and Sabavision, marketplaces such as Torob and Digikala, and messengers such as Telegram, and puts their paid traffic in a paid channel under the network’s name; the AdPix docs note that GA4 typically shows this traffic as referral or unassigned (AdPix docs). And AdPix does not pull spend from ad platforms: you import weekly spend per channel, which the mix model requires, and without it a lift test still gives a verdict but no incremental ROAS or CPA (AdPix docs). The same page recommends running a lift test first and feeding its result into the model.

AdPix runs alongside Google Analytics, so you can keep GA4 while you compare. The Google Analytics alternative guide compares the two criterion by criterion, and the AdPix features overview explains how the lift and MMM screens work.

Which performance marketing guides go deeper?

Each guide below takes one link of the chain further:

• Return on spend: the ROAS formula guide covers break-even ROAS, ROAS against ROI and why platform ROAS over-reports.
• Customer value: the customer lifetime value guide gives three CLV formulas and the LTV to CAC ratio.
• Credit across channels: the attribution models guide compares six models on one journey.
• Proof of cause: the incrementality testing guide explains lift tests, holdouts and where MMM fits.
• What to report: the marketing KPIs guide shows how to pick three to five KPIs per funnel stage.
• Testing: the A/B test sample size guide tells you how many visitors a test needs and how long to run it.

Start with your own break-even ROAS. Take your contribution margin, divide 1 by it, and put that number next to every channel’s ROAS in this month’s report; the channels below the line are where the next decision is.

ROAS calculator

Return on ad spend and your break-even point. The free plan covers 100,000 events a month, no credit card. Setup takes under five minutes.

Start freeTry it free