by Om RathodOpen Meta Ads Manager and Google Ads on the same day, for the same $50k in sales, and you'll get two different stories about who made it happen. Both platforms will claim credit. Add up their numbers and you'll blow past your actual revenue for the month, sometimes by a wide margin.
That gap is why more DTC brands are turning to incrementality testing for ecommerce brands as a real practice, not just a buzzword from a Twitter thread. In plain terms: incrementality testing measures the sales that would not have happened without a specific ad, channel, or campaign. Nothing more mystical than that.
The core problem it solves is simple. Platform-reported ROAS is inflated because it counts conversions that would've happened anyway, branded search from people already Googling you, retargeting shown to shoppers who were already headed to checkout. Meta and Google aren't lying exactly, they're just measuring correlation and calling it causation.
This post walks through what incrementality testing actually measures, why last-click and multi-touch attribution fall apart for smaller ecommerce accounts, the main test types worth knowing, and how to run your first test without hiring a data science team.
The concept is a counterfactual. You compare a group exposed to an ad against a matched group that wasn't exposed, then see what actually differs in sales. That's it.
This is where incrementality testing splits from what your dashboards show you. A dashboard sees a conversion happen after someone viewed an ad and credits the ad. That's correlation. Incrementality testing asks whether the ad caused the conversion, or whether that customer would've bought anyway. That's causation, and it's a much harder, much more honest question.
Here's a scenario that plays out constantly: a brand spending $20k a month on Meta prospecting runs a proper test and finds only $8k of that spend actually drove sales that wouldn't have happened organically. The other $12k was buying credit for demand that already existed.
That gap shows up in a metric worth knowing: incremental ROAS (iROAS) versus platform-reported ROAS. For mature ad accounts, the difference between the two often runs 30-50% [VERIFY]. If you've never run a test, assume your real number is lower than what's on the dashboard, not higher.
Last-click attribution has a structural flaw: it hands full credit to whatever touchpoint happened right before checkout. Usually that's a retargeting ad or a branded search click, the easiest, cheapest conversions to "win." Meanwhile, the TikTok ad that actually introduced the customer to your brand three weeks earlier gets nothing.
Multi-touch attribution was supposed to fix this by spreading credit across the journey. But iOS 14.5+ and cookie deprecation gutted the data volume MTA needs to model accurately, especially for smaller ecommerce accounts that don't have millions of events flowing through their pixel. The models are guessing with less information than they used to have, which makes the outputs feel precise while being anything but.
Here's the ecommerce-specific version of the problem: a customer sees a TikTok ad, searches your brand on Google, then buys off a retargeting email. Three platforms, three dashboards, three claims of credit for one sale. Add it up across your whole funnel and you're triple-counting a meaningful chunk of revenue.
Incrementality testing sidesteps this mess entirely. It doesn't need to track an individual user across platforms or stitch together an identity graph. It just needs two comparable groups, one exposed, one not, and a way to measure the difference in outcomes. No cookies, no device matching, no guesswork about whether that iPhone user is the same person on desktop.
Geo holdout tests
Conversion lift studies
PSA/ghost ad tests
Matched market testing
Budget size should decide which one you pick. Geo holdouts make sense once you're spending $30k or more a month in a single channel, because you need enough volume in each region to see a real signal. Below that, the platform-native lift tests inside Meta and Google are the more realistic starting point since you're not building anything new, just switching on a feature you already have access to.
Step 1: Pick one channel and one specific question. Not "does advertising work" but something answerable, like "is our Meta prospecting budget incremental above $15k a month?"
Step 2: Choose a test structure. For a brand without a data team, a geo holdout is the most accessible option. It doesn't require platform API access or a statistician, just matched regions and a spreadsheet.
Step 3: Set a duration. Two to four weeks minimum. Ecommerce has weekly purchase cycles, paydays, and weekend spikes, and a shorter window will have you reading noise as if it were signal.
Step 4: Hold everything else constant. No new promotions, no surprise email blasts, no inventory swings during the test window. One variable at a time, or the results are junk.
Step 5: Calculate lift correctly. Compare revenue per capita in test versus control regions, not raw revenue totals. A bigger city will always out-earn a smaller one in raw dollars, that's not lift, that's population.
Testing during a skewed period. Running a test through a site-wide sale, a product launch, or Black Friday week guarantees your baseline demand is nothing like normal demand. Wait for a quiet stretch.
Too small, too short. A $5k test budget over 10 days won't produce a result you can trust. If you can't commit real spend and real time to it, you'll get a number that looks decisive and means nothing.
Ignoring cross-channel spillover. Pause Meta and Google Search often just picks up the slack from the same underlying demand. If you only watch the channel you paused, you'll wrongly conclude it had zero impact.
Treating one result as permanent. CAC shifts, creative fatigues, seasonality changes. A test that told you Meta prospecting was incremental in March doesn't guarantee the same answer in September. Re-test quarterly, especially on your biggest spend channels.
A test result only matters if it changes where money goes next. If Meta prospecting comes back with a $8k iROAS out of $20k spend, that's your signal to shift budget toward the channel or campaign that's actually driving new demand, not just harvesting existing intent.
This is a lot easier when you can see blended performance across every channel in one place, rather than toggling between five ad platform dashboards and a Shopify admin tab. Pulling Amazon, Shopify, Meta, and Google data into a unified insights view makes the gap between platform-reported ROAS and actual blended revenue obvious, and that gap is usually the first clue about where to run a test next.
Trivas's AI Wingman layer is built for exactly that moment. It flags anomalies in blended performance data, like a channel that looks great on its own dashboard but isn't moving total revenue, which is often the tell that you need an incrementality test before you cut or scale a budget. If you're a performance marketer juggling five attribution sources already, that's the starting point worth exploring before you build a formal test.
Platform attribution tells you a partial story dressed up as a complete one. Incrementality testing fills in the rest with actual causal evidence, not just a conversion that happened to occur after an impression.
You don't need a data science team to start. You need one channel, one test type, and a control group you didn't contaminate with a sale or a promo. That's a real test, and it'll tell you more than another month of trusting platform ROAS at face value.
The stakes only go up from here. CAC keeps climbing, and ad platforms keep getting better at claiming credit for demand you would've captured anyway. Testing once and calling it done won't hold up.
If you want a clearer view of where your attribution gaps actually are before you commit budget to a formal test, book a walkthrough with Trivas and see how the numbers line up.
explore more insights
3 min read
3 min read
3 min read
.d53b12e5.png&w=3840&q=75 "Trivas.ai")