How Do You Choose a Marketing Measurement Platform for Ecommerce?

Start with the decisions you need to make. Then pick a marketing measurement platform that can actually answer them. Before you sign, have it measure a decision you've already made, and ask what it does when its own numbers disagree.

Meta, Google, and TikTok each claim revenue, and when you add it up, it's more than your store actually made. Then finance wants to cut $100K from paid media, and nobody can say which ads are actually bringing in new sales.

If you only need to tune campaigns inside one small channel, basic attribution might be all you need. Once more money is on the line, or you're asking "what's actually driving growth?", you need more.

What should a measurement platform help you decide?

Think about what you need to decide in the next 90 days. Tuning ads day to day? That's attribution. If you're splitting budget across channels, you want a media mix model (MMM). To find out whether a campaign added any sales at all, a well-run test is usually your clearest answer. Surveys tell you where people first heard of you.

An MMM looks at your spend, your sales, and things like promos and seasonality, then estimates how much each channel adds.

An incrementality test changes your ads for some regions or people and compares their sales to a group you left alone. Always-on fills the gaps between tests with modeled estimates. Just know a daily estimate isn't a daily experiment.

What happens when your measurement methods disagree?

Most of the time they disagree because they're counting different things. So first, figure out what each number actually measures. If two methods really do clash on the same question, go with the one that was set up better. If the question is whether your ads caused sales, that's usually a well-run test.

Here's a made-up example, but you've probably seen one like it. Meta Ads Manager shows 3.5x ROAS, your MMM says 1.2x, and your survey says 40% of new customers found you on Meta. All three can be right:

A test can settle it if two things are true: it's big enough to catch a real change, and the regions you compare against are a fair stand-in for what would have happened otherwise. Once you have a result, use it to tune your MMM, as long as the test and the model cover the same channel, customers, time period, and spend level. And models do miss. When researchers checked 15 Facebook ad experiments against methods that only use observed data, those methods often didn't match the experiment results (Gordon et al., Marketing Science, 2019).

Tests can prove you wrong, too. Rylee + Cru paused Google Brand Shopping, the classic "they'd have bought anyway" campaign, in 15 markets for 20 days. They lost about $139K in revenue on about $21K in spend, a 6.57x incremental ROAS (iROAS) with a range of 4.73x to 8.30x (Rylee + Cru case study). That's the return on the spend they paused. The next dollar they add could earn less. And it's one brand, so test yours before you cut it.

No method should get more say than the question it can actually answer.

Ask every vendor: when your numbers disagree, how do you handle it, and when do you tell a client the data can't answer the question?

Can it measure sales that happen off your website?

Check this before you sign. TV, podcasts, YouTube, and CTV drive sales without clicks, and a lot of those sales happen on Amazon or in stores where your pixel can't see. If a platform only reads site clicks, it will undercount those channels and you might cut spend that's working.

Most marketers can't do this yet. Nielsen's 2025 Annual Marketing Report found only 32% measure their media spend across digital and traditional channels together (Nielsen).

TUSHY hit this exact problem. Their site pixel couldn't connect a TV ad to an Amazon order, so they turned off linear TV in some regions for 28 days.

In those regions, Amazon revenue dropped an estimated 9.34%, about $47K, and the whole 90% range was still a drop (TUSHY case study). Shopify dropped by a similar amount, but that result was less certain.

A click-based tool would likely have rated that TV low. The sales it drove were on Amazon, where the pixel can't see, and the geo test is what caught them.

Is your data ready for a measurement platform?

Probably not as ready as you think. An MMM needs enough history, and enough ups and downs in your spend, to help separate your ads from seasonality and everything else that moves sales. A little over two years of weekly data is a common starting point. Geo tests need sales by region from everywhere you sell. Attribution works better with server-side tracking. And someone on your team has to act on the results.

More history only helps if it still looks like your business today. And better tracking helps attribution see more sales, but it still can't tell you which sales the ads caused.

Make sure you can export your data if you leave. Not sure if you need a team to run it or just the software? See how to vet a marketing measurement consultancy.

How do you test a measurement platform before buying?

Give every finalist the same past decision and some of your data they haven't seen, and have them predict what happened. A bad prediction is a red flag. Of the ones that hold up, pick the one that can explain where each channel's number came from and tells you what it doesn't know.

A demo runs on the vendor's best data. A proof of concept runs on yours.

1. Pick a real decision from last year, like "we cut brand search 30% in Q2."
2. Hold back your most recent 8 to 12 weeks. The model doesn't get to see them. It predicts them from the spend you actually ran. We call this a forward test.
3. Compare the prediction to what really happened. A big miss is a warning. A close match is good, but a model can nail total revenue and still give credit to the wrong channel.
4. Ask what evidence supports each channel's number. Which tests back it up?
5. Ask for the range, and what it doesn't cover. "2.1x, between 1.6x and 2.6x" is a good start. Then ask whether that range covers bad inputs and wrong model assumptions, or just normal noise.

Shopping for testing only? Have the vendor design a test for your decision and tell you the smallest lift it could catch before anything runs. Our guide to picking an incrementality testing platform goes deeper, and the nine vendor checks cover what any vendor should show you.

Full disclosure, we sell this. Stella puts attribution, surveys, tests, and MMM in one place, with modeled estimates between tests and a range on every number.

Test us the same way. If you only need to tune one small channel, you probably don't need all of this yet.

Frequently asked questions

What's the difference between an attribution tool and a measurement platform?

An attribution tool splits credit for sales across the ads someone clicked or saw. A measurement platform adds tests and MMM, which try to estimate what your ads actually added.

Why do Shopify, Meta, and Google report different revenue?

Each one uses its own rules and time windows. Several platforms can claim the same sale, so their totals often add up to more than your store made.

Is attribution enough to measure ecommerce growth?

Not if you need to know what your ads caused. It's useful for running campaigns, but it gives credit instead of showing what would have happened without the ad.

Do I need MMM if I'm already running incrementality tests?

It helps once you're splitting budget across several channels and have enough history. Tests answer one question at a time. An MMM fills in the rest of your mix, with more assumptions.

Can one platform measure Amazon and retail sales too?

Some can, if they can see those sales by region. A geo test that includes Amazon and retail by state or DMA can catch sales your pixel misses.

Bring one decision you made last quarter. We'll show you how we'd have measured it, range included. Book a demo.