Incrementality vs. attribution: what's the difference?
Attribution gives credit for a sale to the ads a customer saw or clicked before buying. Incrementality asks a harder question: would that sale have happened without the ads? Attribution tells you which ads were on the path. Incrementality tells you which sales the ads actually added.
Key takeaways
- Attribution splits the credit for each conversion across the ads someone touched. It's fast and detailed, but it can't tell which sales would have happened anyway.
- Incrementality measures the lift ads create over a baseline, usually by comparing a group that saw the ads with a similar group that didn't.
- Ad platforms' own conversion reports are attribution, scored by the company that sold you the ads.
- A media mix model estimates incrementality from history, for every channel at once, including offline sales.
- Use attribution to tune campaigns day to day. Use lift tests and a media mix model to decide budgets.
What attribution is
Attribution follows a customer's path from ad to purchase and decides how to split the credit. A last-click model gives all of it to the final ad clicked. A first-click model gives it to the first. Data-driven models, like the one Google Analytics uses, spread it across the touchpoints based on how paths that convert differ from paths that don't.
Attribution is good at detail and speed. You can see results by campaign, keyword and ad, often the same day. That makes it the right tool for running campaigns: which headline, which audience, which search term.
It has three blind spots:
- It only sees what can be tracked. Clicks and site visits, mostly. TV, billboards, radio and in-store sales sit outside it.
- It can't see the sales that would have happened anyway. Someone who searches your brand name and clicks your ad was probably already on the way. Attribution still credits the ad.
- Each platform grades its own homework. Every ad platform counts conversions from its own point of view, so added together they often claim more sales than you actually made.
What incrementality is
Incrementality is the difference the ads made. The cleanest way to measure it is to compare two similar groups: one that saw the ads and one that didn't. Whatever the first group did beyond the second is the lift.
There are three common ways to get there:
- Lift studies and holdouts. The platform or advertiser keeps a random control group from seeing the ads, then compares outcomes such as awareness, visits or sales.
- Geo tests. Ads run in some regions and stay off (or change) in others, and you compare sales between them.
- Media mix models. No live test. The model estimates incrementality from how sales moved with spend over time, for all channels at once. Read more in what is a media mix model.
Here's a real example. For a luxury resort launch, a YouTube Brand Lift study compared people who saw the lead spot with a control group who were eligible but didn't. Among 55 to 64 year olds, awareness was 6.73% for people who saw the spot against 4.13% for the control group, a 63% lift. That number is incremental by design, because the control group shows what would have happened anyway. See the study.
Side by side
| Attribution | Lift test | Media mix model | |
|---|---|---|---|
| Question | Which ads were on the path? | Did this channel add anything? | What did each channel add, and where should the next dollar go? |
| Data | Tracked clicks and visits | A test group and a control group | Sales and spend over time, by channel |
| Speed | Same day | Weeks per test | Months of history, then regular refreshes |
| Sees offline sales | Rarely | If the test measures them | Yes |
| Best for | Tuning campaigns and ads | Proving one channel | Setting budgets across channels |
When to use which
Use attribution for the daily work: bids, keywords, audiences and creative. Use a lift test when one big question needs a clean answer, like whether a new channel is worth keeping. Use a media mix model when the question is the whole budget, or when most of your sales happen where attribution can't see, like in a store. If you sell mostly offline, read how to measure marketing when most sales happen in stores.
When the methods disagree, that's useful too. A channel that looks great in attribution but adds little in a lift test or a model is often taking credit for sales that were coming anyway.
We use all three
We set up clean tracking first: Google Analytics, Google Tag Manager and tagged links, so attribution is as accurate as it can be. We run A/B and incrementality tests that prove what works before you scale spend. And we build media mix models with Google's open-source Meridian to measure every channel against all of your sales.
We report the results plainly, with the caveats. Our model figures are estimates, not holdout tests, and we say so. See how that looks on our Results page.
Common questions
Is incrementality better than attribution?
They answer different questions, so one doesn't replace the other. Attribution is the better tool for tuning campaigns, keywords and ads quickly. Incrementality is the better tool for deciding how much to spend and where.
Can I trust an ad platform's own lift study?
A lift study is a real experiment with a control group, so it measures lift far better than an attribution report does. It only covers that platform, and the platform runs it, so read the method and the fine print. The study in our resort case showed strong lift for one age group, and it also showed that not every age group or device moved.
What is a holdout test?
A holdout test keeps ads away from a group of people or places on purpose, then compares their results with the group that got the ads. The difference is the lift. It's the most direct way to measure incrementality, at the cost of not advertising to the holdout for a while.
