Tracking paid ads and measuring their impact are different jobs. Tracking tells you which ads touched the people who bought. Impact tells you how many of those sales would not have happened without the ads. A brand can have excellent tracking and still waste a third of its budget on ads that reach people who were buying anyway. I covered the tracking side in best ad tracking tools for e-commerce. This post is about impact, which the industry calls incrementality.
There are three ways to measure it. Experiments compare people or regions that saw ads with ones that did not. Media mix models estimate each channel’s contribution from spend and sales patterns over time. And attribution, done carefully, gives you a daily approximation that you calibrate against the other two. The tools below cover all three. If the terms are new, start with what is incrementality testing, and when is it worth paying for?.
Platform lift studies
The cheapest real experiment you can run. Meta’s Conversion Lift holds out a randomized group from seeing your ads and compares purchase rates, and Google offers lift measurement for some campaign types. There is no extra software cost for eligible advertisers. The limits are that each platform measures only itself, you need enough conversions for a readable result, and the platform is still grading its own work, even if the method is sound. I use them as a calibration check twice a year.
A do-it-yourself geo holdout
For a channel you can target by region, pause it in some markets and keep it running in comparable ones for four to six weeks, then compare total sales growth between the two. The tooling is a spreadsheet and location targeting in the ad platform. It is the most underrated method on this list, because it measures total business impact rather than attributed sales, and it costs nothing but discipline. My walkthrough is in running a cheap geo holdout to sanity-check a channel.
Haus
Haus runs geo experiments as a service, handling market selection, statistical design, and analysis. It is also one of the partners OpenAI named this month for early geo-based tests of ChatGPT ads. A managed platform makes sense once a test result could move a large share of budget and you want a design that leadership will trust. Ask what minimum spend and duration it needs for a conclusive result at your size. Alternatives are in Haus alternatives for incrementality testing.
Recast and Meridian
Media mix modeling estimates every channel’s contribution at once, including ones that are hard to experiment on, like podcasts or brand campaigns. Recast builds and runs models as a service, and Google’s Meridian is a free open-source option if you have someone comfortable with statistical modeling. Both need roughly two years of clean weekly spend and sales data and enough variation in spend to learn from. They answer budget allocation questions well and day-to-day campaign questions poorly. More in Recast alternatives for marketing mix modeling.
ThoughtMetric
ThoughtMetric is an attribution tool, not an incrementality platform, but it covers the daily layer that experiments and models cannot. (Disclosure: ThoughtMetric sponsors this site.) It reports new-customer ROAS and cost per new-customer order by channel, campaign, and ad, which is a closer proxy for incremental impact than blended ROAS, since returning customers are the sales most likely to happen without ads. It also folds post-purchase survey answers into attribution. Pricing starts at $99 a month for 50,000 pageviews with every feature included and a 14-day trial. Use it daily, and calibrate it against a lift test or holdout once or twice a year.
How the pieces fit
At most brands under $20M, the practical setup is attribution for daily decisions, one or two experiments a year on your biggest channels, and a media mix model only once spend is large and spread across many channels. When the experiment and the attribution disagree, trust the experiment and adjust how you read the attribution. I described when each method is worth it in MTA, MMM, or incrementality, depending on what you spend.
Picking tools
- Want a real experiment at no software cost? Run a platform lift study.
- Want total business impact for a regional channel? A do-it-yourself geo holdout.
- High-stakes test that leadership needs to trust? Haus.
- Large budget across many channels and two years of data? Recast or Meridian.
- Need a daily read on new-customer impact by campaign? ThoughtMetric (our sponsor), calibrated by tests.
Leave a Reply