Lift Analysis: What It Is and How to Run One
. By Ads Daddy.
# Lift Analysis: What It Is and How to Run One  Lift analysis answers one blunt question: **what happened because of your ads that would not have happened without them?** You compare a group that saw your advertising with a similar group that didn't. The gap between the two is the lift. That gap is the only part of your ad spend that is truly paying for itself. Everything else is people who were going to buy anyway, and your ad just stood next to them taking credit. ## TL;DR - Lift analysis compares an exposed (test) group with an unexposed (control) group. The difference in outcomes is what your ads caused. - Relative lift = (test rate − control rate) ÷ control rate. Absolute lift is the plain gap in percentage points. - It needs random assignment, enough volume and a clear question. Without those, the number is noise with a nice font. - It tells you what attribution can't: whether a sale would have happened anyway. - Run it on your biggest budget lines first, then repeat it. One test is a snapshot, not a verdict. ## Table of Contents - [What lift analysis actually means](#what-lift-analysis-actually-means) - [Lift analysis vs attribution: different questions](#lift-analysis-vs-attribution-different-questions) - [The formulas, in plain English](#the-formulas-in-plain-english) - [A worked example, with dollars](#a-worked-example-with-dollars) - [What lift can measure](#what-lift-can-measure) - [How to design a lift test](#how-to-design-a-lift-test) - [How to read the results without fooling yourself](#how-to-read-the-results-without-fooling-yourself) - [Where lift analysis goes wrong](#where-lift-analysis-goes-wrong) - [Our perspective: lift is a budget tool, not a science project](#our-perspective-lift-is-a-budget-tool-not-a-science-project) - [Want someone else to run the tests?](#want-someone-else-to-run-the-tests) - [A quick plan for your first lift test](#a-quick-plan-for-your-first-lift-test) - [FAQ](#faq) ## What lift analysis actually means Think of it as a taste test. You hand one table a sample and the other table nothing. Then you count who ends up buying the sandwich. If both tables buy at the same rate, your sample wasn't doing much, however good it looked. In advertising, the "sample" is your ad. The two tables are: - **Test group:** people who could be shown your ads. - **Control group:** people who look the same but are held back from seeing them. Then you measure an outcome. It might be purchases, leads, brand searches, sign-ups or revenue. The difference between the groups is the lift. Analysts also call this incrementality, because it's the *incremental* result on top of what would have happened anyway. If you want the full backstory on that idea, we've written about [what ad incrementality is](https://adsdaddy.com/au/blog/what-is-ad-incrementality). Two things matter in that definition: 1. **It's causal.** You aren't asking who clicked. You're asking what changed because of the ad. 2. **It needs a baseline.** No control group, no lift analysis. Comparing this month to last month is not a control. Seasonality, a sale or a competitor's campaign can swamp everything. ## Lift analysis vs attribution: different questions Attribution says: "This customer touched these ads before buying, so credit goes here." Lift says: "Would this customer have bought without those ads?" Those are not the same question. And they can give wildly different answers. Here's the classic trap. Someone searches for your brand name, clicks your brand ad, and buys. Attribution happily gives the ad 100% of the credit. Lift analysis may find that most of those people would have clicked your free organic listing instead. The ad was a toll booth on a road they were already driving down. Neither tool is the villain. Attribution is a fast, daily view of which touchpoints show up in journeys. Lift is a slower, sturdier reality check. Use attribution to steer day to day. Use lift to check that the steering is pointing somewhere real. Our guide to [attribution modelling and ROI](https://adsdaddy.com/au/blog/what-is-ad-attribution-a-2026-guide-for-marketers) covers the first half, and [why assisted conversions matter](https://adsdaddy.com/au/blog/why-track-assisted-conversions) covers the stuff last-click quietly hides. ## The formulas, in plain English There are two ways to express lift. Mixing them up is the most common mistake in marketing reports, so let's be clear. **Absolute lift** is the simple gap, in percentage points: > Absolute lift = test conversion rate − control conversion rate **Relative lift** (often called the lift rate) shows the gap as a share of the control: > Relative lift = (test rate − control rate) ÷ control rate Say the test group converts at 4% and the control at 3%. - Absolute lift: **1 percentage point** - Relative lift: 1 ÷ 3 ≈ **33%** Same result. Two very different headlines. "Ads lifted conversions by 33%" sounds huge. "Ads added one percentage point" sounds modest. Both are true. Always say which one you mean, and always show the underlying rates next to it. **Sales lift** is the same thinking applied to revenue: the sales in the exposed group minus the sales you'd expect from that group without ads (the baseline). Where the comparison is between two groups of different sizes, work from rates or per-person averages, not raw totals. ## A worked example, with dollars This is an illustration with made-up numbers, so you can follow the maths. A Shopify store spends A$5,000 on a four-week prospecting campaign. The audience is randomly split: | | Test group | Control group | |---|---|---| | People | 10,000 | 10,000 | | Purchases | 400 | 300 | | Conversion rate | 4% | 3% | - Absolute lift: 1 percentage point - Relative lift: about 33% - Incremental purchases: 400 − 300 = **100** Now the bit that makes finance people sit up. The ad platform reports 400 purchases, so it says each cost A$12.50. But only 100 of those were caused by the ads. Your **cost per incremental purchase** is A$5,000 ÷ 100 = **A$50**. If your average order gives A$35 in margin, the platform's story says "profit machine". The lift story says "this campaign loses money on first purchase". Maybe lifetime value rescues it. Maybe it doesn't. Now you know which question to ask. That's the whole point. Lift analysis tends to be a cold shower for some campaigns and a pleasant surprise for others. Upper-funnel video that looked weak in last-click reports can show real lift. Retargeting that looked like a hero can show almost none. ## What lift can measure Lift isn't just about sales. You pick the outcome that matches the job the campaign has: - **Conversion lift:** extra purchases, leads or sign-ups. - **Sales or revenue lift:** extra dollars, including basket size and repeat purchase. - **Brand lift:** changes in awareness, recall or consideration, usually measured by survey. See our piece on [brand lift and e-commerce growth](https://adsdaddy.com/au/blog/brand-lift-ecommerce-growth-au). - **Search lift:** more people searching for your brand after exposure. - **Store or offline lift:** extra in-person visits or phone enquiries. Pick one primary outcome before the test starts. If you pick the winner after the results come in, you've gone fishing, and fish always bite eventually. It also helps to be clear about the numbers you track alongside lift. If you're sorting out which metrics belong on the dashboard, this walkthrough on [tracking sales KPIs](https://www.solution4guru.com/how-to-track-sales-kpis-in-monday-crm-what-should-you-measure-and-how-do-you-set-it-up) is a handy companion. ## How to design a lift test You don't need a data science team. You need discipline. Here's the order we'd follow. 1. **Write a specific hypothesis.** "Meta prospecting ads will increase first purchases from new customers over four weeks." Not "ads help." 2. **Choose one primary KPI.** Purchases, qualified leads, whatever the business actually cares about. 3. **Randomly split the audience.** Random assignment is what separates a real experiment from a before-and-after guess. It spreads out all the things you can't see, like income, intent and mood. 4. **Hold the control back.** Ideally the control group is eligible to be shown the ad but isn't (often done with a "ghost ad" or holdout). That keeps the groups comparable. 5. **Use a geo split when you can't randomise people.** Match regions that behave alike, run ads in some and not in others. Matched markets are especially handy for local and regional businesses. 6. **Size it properly.** Small groups mean wobbly results. If your conversion rate is low, you need more people or more time to see a difference. 7. **Leave everything else alone.** Don't launch a sale, swap the landing page or change the budget mid-test. 8. **Run it long enough.** Cover at least a full buying cycle. A product people consider for three weeks can't be judged in four days. ### Platform tools or DIY? Major ad platforms offer built-in lift studies that run this split for you. These are the easiest start when you have enough spend and volume. Where you don't, a geo holdout or a simple audience holdout you manage yourself can do the job. Either way, the logic is the same.  ## How to read the results without fooling yourself A result is only as good as its confidence. Here's a quick sanity checklist. - **Is the difference statistically meaningful?** A lift of 0.2 points on small groups is probably noise. Treat results inside the noise floor as "no clear answer", not as "it worked". - **Look at the range, not just the point.** A confidence interval that runs from −5% to +40% tells you almost nothing. - **Check percentage vs percentage points.** Go back to the formulas above. Say both. - **Repeat it.** One test can be a fluke. A lift that shows up again is something you can plan budgets around. - **Compare cost per incremental result,** not cost per reported result. No shame in an inconclusive test. It's a cheaper lesson than burning another quarter of budget on a hunch. ## Where lift analysis goes wrong It's powerful, not magic. These are the usual ways it misleads. **Too little data.** Online conversion rates are often low and the effects small. With sparse conversions you can need very large groups to detect a real difference. **Contaminated controls.** If your control group sees the same campaign elsewhere, say on another platform or through a different ad format, the gap shrinks and you underrate the ads. **Other things changing at once.** A public holiday, a price drop, a competitor promotion or a second campaign can all move results during the test. Keep a log of everything that changes. **A vague hypothesis.** "Let's see what happens" produces a report, not a decision. **Testing too many things.** Every extra channel or combination multiplies the volume you need. Test one major question at a time. **Treating it as one-and-done.** Audiences, creative and competition shift. Think of lift measurement as a routine, a bit like a dentist check-up, not a once-off miracle cure.  ## Our perspective: lift is a budget tool, not a science project We see the same pattern with new clients. They arrive with a platform dashboard glowing green, and revenue that doesn't quite match. Everything "converts", yet the bank balance shrugs. That's usually a lift problem in disguise. The ads are claiming credit for customers who were already on their way. Our approach is simple: - **Start with the money lines.** Test where the spend is biggest, usually brand search, retargeting and prospecting on Meta. - **Pair lift with attribution.** Use attribution for daily decisions and lift as the audit that keeps it honest. - **Keep tests small and repeatable.** A clean geo split or audience holdout beats an elaborate study nobody finishes. - **Judge by cost per incremental result,** and tie it back to margin. ROI is the scoreboard. Everything else is commentary. We also think most small and mid-sized businesses overcomplicate this. You don't need a PhD. You need a hypothesis, a holdout and the patience to let the test finish. If you want a wider view of where this fits in your reporting, our [2026 guide to measurement in campaigns](https://adsdaddy.com/au/blog/practical-guide-understanding-ad-metrics) maps it out. > *— Adrian* ## Want someone else to run the tests? Lift analysis is only useful if the campaigns around it are well built. That's the job we do. Adsdaddy builds and manages paid campaigns across [Meta, Google, YouTube, LinkedIn and Microsoft Ads](https://adsdaddy.com/au), with tracking set up so you can see what's driving results, and you never have to learn the ad platforms yourself. If what you really need is a steady flow of enquiries, look at our [lead generation service](https://adsdaddy.com/au/lead-generation). For finance businesses, our [pay-per-lead options](https://adsdaddy.com/au/finance-brokers) start from A$70 per lead for consumer car finance leads and from A$170 per lead for commercial ABN finance leads. A custom-built lead gen system starts from A$17,500 as a one-off setup. Want to know how much of your current ad spend is genuinely incremental? **[Book a call](https://adsdaddy.com/au)** and we'll look at it with you. ## A quick plan for your first lift test If you want to start this week, keep it brutally simple: 1. Pick the one campaign that takes the most budget. 2. Define one outcome: purchases, or qualified leads. 3. Split the audience or the regions into test and control. 4. Run for a full buying cycle without changing anything else. 5. Calculate absolute lift, relative lift and cost per incremental result. 6. Decide: scale, fix or cut. Then test again. That's it. No jargon, no expensive software. Just a clean answer to a question most advertisers never ask. ## FAQ ### What is lift analysis in simple terms? It's a controlled comparison between people who saw your ads and similar people who didn't. The difference in outcomes, such as purchases or leads, is the lift. It tells you what the advertising actually caused, not just what it was present for. ### How do you calculate lift? Subtract the control group's conversion rate from the test group's, then divide by the control rate. With 4% in the test group and 3% in the control, relative lift is about 33% and absolute lift is 1 percentage point. Always report both, plus the raw rates. ### What does "lift" mean in marketing? In marketing, lift is the extra result you get because of a campaign, compared with what would have happened without it. That result might be sales, sign-ups, searches or brand awareness. It's the part of performance you can honestly credit to the ads. ### How is lift different from attribution? Attribution assigns credit to touchpoints along a customer's path. Lift tests whether the sale would have happened anyway. Attribution is useful for daily optimisation, while lift is better for checking that your spending is really driving new business. ### Do small businesses need lift analysis? If you spend a meaningful amount on ads, yes, in a simple form. A geo holdout or audience split can show whether a campaign is adding customers or just claiming ones you'd already won. Very small budgets may not have enough volume for a clear result, so start with your biggest line. ## Recommended - [Ad Quality Score Explained: What It Is and How to Lift It](https://adsdaddy.com/au/blog/explain-ad-quality-score) - [Use ad analytics to boost your ROI and campaign results](https://adsdaddy.com/au/blog/use-ad-analytics-to-boost-your-roi-and-campaign-results) - [Why conduct ad audits? Unlock better ROI today](https://adsdaddy.com/au/blog/why-conduct-ad-audits-unlock-better-roi-today) - [Attribution Modelling: Boosting ROI for Marketers](https://adsdaddy.com/au/blog/attribution-modelling-marketing-roi)