How should marketers evaluate conflicting attribution results?
When Meta, MTA, and MMM disagree on the same campaign, here's why, how to compare them fairly, and how to decide which numbers to trust.
Linnea Zielinski · 12 min read
A family gets home from the drive to Grandma's house and starts arguing about how long the trip took. Dad says four hours, because he's counting from the moment they pulled out of the driveway. The GPS says three hours and twenty minutes, because it only counts time the car was moving. Mom says it took all day, because she's counting from when everyone started packing. Technically nobody's lying, and nobody did the math wrong. They're answering three different questions with the same words. Averaging their answers wouldn't settle anything, because the average doesn't describe any trip that actually happened.
Attribution tools disagree for the same reason. Meta, a multi-touch attribution (MTA) platform, and a marketing mix model (MMM) can all look at the same campaign and report very different revenue, and each one can be correct by its own definition. The trouble starts when marketers treat those numbers as competing answers to one question and then pick the one they like, average them, or stop trusting measurement altogether. Every budget decision rests on that choice, so knowing how to read conflicting results is what keeps a team from cutting a campaign that's working or scaling one that isn't.
Key takeaways
- Attribution tools disagree mainly because they use different methods, conversion definitions, attribution windows, and data coverage, and because they make different assumptions about whether a sale would've happened anyway.
- Before comparing numbers, confirm the tools are measuring the same thing, like revenue vs. orders, gross vs. net sales, attributed vs. added revenue, and website vs. total business revenue.
- Compare results on matching time periods, campaign groupings, and revenue coverage, and look at whether tools agree on direction and relative size before fixating on headline ROAS.
- The biggest gaps usually trace back to branded search, retargeting, halo effects across channels, and seasons when spend and demand rise together.
- Averaging conflicting results doesn't get you closer to the truth, because each tool's errors lean in a consistent direction and the tools are often measuring different things.
- Choose which evidence to trust based on the decision you're making, and put the most weight on findings that hold up across reasonable assumptions.
- Decide in advance which source governs which decision, so teams don't switch to whichever number supports the plan they already wanted.
Why attribution tools disagree
Attribution discrepancies aren't usually a sign that something's broken. They're mostly built into how each tool works. The three most common approaches start from very different places.
| Approach | What it measures | Its main blind spot |
|---|---|---|
| Platform attribution (Meta, Google, TikTok) | Conversions that followed an interaction with that platform's ads | Can't see other platforms, and can't tell whether the sale would've happened anyway |
| Multi-touch attribution (MTA) | How credit gets split across the touchpoints it can track | Misses journeys it can't connect and channels that don't generate clicks |
| Marketing mix modeling (MMM) | How much revenue each campaign added, estimated from spend and revenue over time | Depends on how well the model is built and how much variation is in the data |
Within those approaches, five specific differences drive most of the gaps.
Different measurement methods
Platform attribution and MTA rely on tracking people: a click, a view, or a matched ID connects an ad to a purchase. MMM works the other way around. It starts with the revenue you actually earned and works backward to estimate what drove it, without needing to follow anyone. Those methods can land on very different numbers even with perfect data.
Different definitions of a conversion
One tool might count purchases, another might count first-time customers, and a third might count total revenue. Some include tax and shipping, some subtract refunds, and some count a lead form as a conversion. Two tools can track the same campaign and report different totals simply because they aren't counting the same event.
Different attribution windows
Each platform sets its own rules for how long after an ad a sale can still count. Meta's default ad set setting is a 7-day click, 1-day engage-through, and 1-day view-through window, while most Google Ads reports show a 30-day lookback window by default. An MMM doesn't use a window at all. It estimates how long each campaign's effects last from the data itself (sometimes days, sometimes weeks), which can be longer or shorter than any platform's setting.
Different data coverage
Each tool only sees what it can see. Meta can't see your podcast ads or your Amazon sales, even when its own ads drove them. An MTA tool can't follow journeys it can't connect, which leaves out many CTV, audio, and cross-device paths. An MMM can include any channel and sales destination you feed it, but it can only measure what's in the data.
That limited view also causes double counting. If a shopper clicks a Google ad and later a Meta ad before buying, both platforms may claim the full sale, so adding up every platform's reported revenue can produce a total larger than what your business actually earned.
Different assumptions about what marketing added
This is the biggest difference of all. Platform attribution and most MTA tools give credit whenever an ad came before a sale. An MMM tries to separate revenue your marketing added from revenue you would've earned anyway through brand awareness, repeat customers, and seasonality. Those are different questions, so they'll almost always produce different answers.
First, make sure the tools are measuring the same thing
Before digging into methodology, check the definitions. A surprising share of attribution conflicts disappear once teams realize they're comparing two different metrics with the same label.
| Check | Why it matters |
|---|---|
| Revenue vs. orders | One tool may report dollars while another reports purchases, and order values vary |
| Gross vs. net sales | Discounts, refunds, tax, and shipping can swing totals by a lot |
| Platform-attributed vs. added revenue | Attributed revenue counts sales near an ad, while added revenue estimates what the ad actually changed |
| Website revenue vs. total business revenue | A tool that only sees your website will miss sales on Amazon and at retail |
That last row matters especially for omnichannel brands. If one tool measures only website sales and another measures website, Amazon, and retail revenue, the second one may report more revenue for the same campaign simply because it's counting more of your business.
An example: three tools, three answers, one campaign
Here's a hypothetical Meta prospecting campaign, measured three ways. These numbers are illustrative, and as you'll see, they aren't all measuring the same quantity.
| Source | Reported revenue | What it's actually counting |
|---|---|---|
| Meta Ads Manager | $200,000 | Website purchases within Meta's attribution window after a click, engagement, or view |
| MTA tool | $140,000 | Meta's share of credit across the tracked touchpoints in each journey |
| Marketing mix model | $250,000 | Total revenue the campaign added across sales destinations |
At first glance, the MMM looks like it's being generous and the MTA tool looks stingy. Breaking the MMM figure into its parts shows what's actually happening:
| Part of the MMM estimate | Revenue | Where it landed |
|---|---|---|
| Base revenue (direct contribution) | $120,000 | Website purchases the campaign drove directly |
| Halo effect: Amazon | $70,000 | Shoppers who saw the ad and bought on Amazon |
| Halo effect: branded search | $40,000 | Shoppers who searched your brand later |
| Halo effect: direct traffic | $20,000 | Shoppers who typed your URL straight in |
| Total contribution | $250,000 |
Now the three numbers make more sense together:
- Meta's $200,000 includes some website buyers who would've purchased anyway, which is why it's higher than the MMM's $120,000 in base revenue.
- The MTA tool's $140,000 is lower because it splits credit with other tracked touchpoints, like branded search clicks, that were partly created by the Meta campaign.
- The MMM's $250,000 is the highest because it counts revenue that landed outside Meta's view, on Amazon and through other channels.
They describe different slices of what the campaign did, and the useful question is which slice fits the decision you're making.
How to compare attribution outputs meaningfully
Once you know what each tool measures, you can line them up fairly. The goal isn't to force every tool to match. It's to remove the differences you can fix, so the ones left over tell you something real.
Start with your own business records, like your order system or finance-approved revenue. That's your reference point for how much revenue actually came in. It won't tell you what drove it, but it'll show you when a tool's totals don't add up.
Then work through these steps:
- Match the reporting periods. Use the same date range, time zone, and cutoff, and check whether a tool reports by the date of the ad interaction or the date of the sale.
- Align campaign groupings. Make sure "Meta prospecting" means the same set of campaigns in every tool, since one tool's campaign might be another's channel total.
- Check revenue coverage. Note which tools see website, Amazon, and retail sales, and which see only one of them.
- Identify missing channels. List any channels a tool can't see, like podcasts, CTV, direct mail, or affiliate, since their impact will show up somewhere else in that tool's numbers.
- Compare direction and size, not just headline ROAS. If every tool ranks campaigns in roughly the same order, that agreement is useful even when the dollar figures differ.
That last step is often the most practical. Here's a hypothetical example of what agreement and disagreement can look like across tools:
| Campaign | Meta-reported ROAS | MTA ROAS | Modeled ROAS | What it suggests |
|---|---|---|---|---|
| Prospecting A | 2.1 | 1.6 | 2.8 | Tools disagree on size, but all show it working |
| Prospecting B | 1.2 | 0.9 | 2.5 | Strong halo effects that click-based tools miss |
| Retargeting | 6.4 | 4.8 | 1.3 | Likely capturing demand created elsewhere |
When tools agree on direction, you can act with more confidence. When they point in opposite directions, like the retargeting row, that's where to dig in.
How to investigate major disagreements
Big gaps between tools tend to show up in the same few places. Knowing where to look turns a vague "the numbers don't match" into a specific question you can answer.
Branded search
Branded search almost always looks great in platform reporting and MTA, since it sits right before the purchase. The question is how much of that demand it created versus collected. If branded search revenue rises and falls with your Meta, YouTube, or CTV spend, those campaigns are likely creating much of the demand branded search is capturing.
What to check: whether branded search volume moves with upper-funnel spend, and how much of its reported revenue your MMM attributes to it versus to the campaigns feeding it.
Retargeting
Retargeting shows people ads after they've already visited your site or added something to their cart. Many of them were already close to buying, so platform and MTA reporting tend to give retargeting far more credit than an MMM does. Prescient often moves credit away from retargeting toward the prospecting campaigns that created the interest in the first place.
What to check: how retargeting's reported revenue compares with its Modeled ROAS, and whether prospecting campaigns look stronger once that credit is redistributed.
Halo effects across channels
A campaign can drive sales through channels its own platform can't see, like a Meta ad that leads to an Amazon purchase or a CTV spot that leads to a branded search. Tools that only see one channel will undercount those campaigns, while an MMM that measures halo effects will credit them more. Halo effects aren't only an upper-funnel story, either, since even a retargeting ad someone sees without clicking can lead to a later Amazon sale.
What to check: which sales destinations each tool can see, and how much of a campaign's modeled contribution comes from halo effects.
Seasonality and simultaneous spending changes
When you raise spend across several channels at once, like during Black Friday, every tool has a harder time telling campaigns apart. Platforms tend to claim the holiday lift for themselves, and some MMMs can confuse holiday demand with marketing impact or the reverse.
What to check: whether the disagreement is concentrated in peak periods or launches, and whether several channels changed spend at the same time.
Why taking the average isn't a reliable solution
When three tools give three different numbers, averaging them can feel like a fair compromise. In the example above, the average of $200,000, $140,000, and $250,000 is about $197,000. That number doesn't describe anything real, and it can lead you further from the truth for a few reasons:
- Errors don't cancel out. Platform attribution tends to lean high for campaigns that capture demand, like retargeting and branded search, and low for campaigns whose sales land elsewhere. Those errors point in consistent directions, so averaging carries them along instead of cancelling them.
- The tools are estimating different things. Averaging attributed website revenue with total added revenue across sales destinations is like averaging Dad's four hours with the GPS's three hours and twenty minutes.
- The average isn't closer to the truth. If one tool is right about a campaign and two are wrong in the same direction, the average lands closer to the wrong answers.
Other common mistakes when reconciling attribution
Averaging isn't the only shortcut that backfires. A few other habits come up often when teams try to make sense of conflicting numbers.
| Mistake | Why it's a problem |
|---|---|
| Assuming the lowest number is the safest | An MMM can report more revenue than platforms for campaigns whose sales land elsewhere, so lower isn't automatically more accurate |
| Treating any one tool as infallible | Every method has limits, including MMM, which depends on good data and a well-built model |
| Assuming every gap is a tracking error | Many gaps come from tools answering different questions, not from broken pixels |
| Comparing ROAS across platforms directly | A 6.0 on Meta and a 3.0 on a podcast platform use different rules, so they aren't comparable |
| Ignoring uncertainty | An estimate with a wide range of possible values isn't the same as a precise one |
| Switching sources to fit the plan | Trusting whichever tool supports the decision you already wanted undermines all of them |
How to decide which evidence to trust
Trust shouldn't go to one tool across the board. It should depend on the decision in front of you and on how well the evidence supports that specific call.
Match the method to the business question
Each tool is strongest for a different kind of decision. A simple way to divide the work:
| Decision | Best primary source | Useful supporting source |
|---|---|---|
| Which creative or audience to optimize inside Meta | Meta's own reporting | MTA |
| Which touchpoints show up in converting journeys | MTA | Platform reporting |
| How much revenue a campaign added across your business | MMM | Platform reporting for context |
| Whether to shift budget between channels | MMM with forecasting | Tests where they're practical |
| How much revenue the business actually earned | Your own order and finance records | None needed |
Evaluate data coverage and uncertainty
The more of your business a tool can see, the more weight its cross-channel conclusions deserve. A tool that only sees website sales shouldn't settle a question about Amazon. It also helps to know how confident each estimate is. If two campaigns' results fall within each other's range of likely values, the difference may not be big enough to act on.
Consider independent experiments, with their limits in mind
Geo tests and holdouts can add a helpful data point when tools disagree. They have real limits, though. Prescient's view is that incrementality tests are locally accurate and globally inaccurate: a test describes one campaign in one window under one set of conditions, so it can't settle every disagreement on its own. Tests also struggle with retargeting, since they can't separate shoppers who saw an ad from those who would've bought anyway. That's why Prescient shows brands their model's accuracy both with and without a test included, and the brand makes the final call.
Prioritize findings that hold up across reasonable assumptions
The most trustworthy conclusions are the ones that don't flip when you change the details. If a campaign looks strong across different time periods, different attribution windows, and both modeled and platform-reported views, you can act on it with confidence. If a conclusion only holds under one tool's assumptions, treat it as a question to investigate, not an answer.
Here's a quick checklist for any estimate before you act on it:
- Is it measuring the same outcome and scope as the decision you're making?
- Does the tool see the channels and sales destinations involved?
- Does the method separate added revenue from revenue you'd have earned anyway?
- Does it show how confident it is?
- Does the finding hold up across periods and reasonable assumptions?
Set the rules before the results arrive
Decide ahead of time which source governs which decision, how you'll handle disagreements, and what evidence it takes to override a number. When those rules are set in advance, attribution conflicts become routine questions to work through instead of debates about whose dashboard wins.
Where Prescient comes in
Prescient's marketing mix model uses a different method than identity-based attribution tools. It starts from the revenue you actually earned and works backward to estimate what each campaign added, including halo effects on Amazon, branded search, and direct traffic. Our MMM shows platform-reported numbers right alongside modeled ones, with base revenue and halo effects broken out, so you can see exactly where the tools agree and where they don't. No measurement approach is perfect, which is why Prescient also shows a model confidence score with its results.
That gives omnichannel brands a consistent way to read conflicting attribution and make budget calls on the revenue campaigns actually add, with our Media Forecasting feature to plan what happens when you scale or shift spend. If you want to see how your campaigns look once the numbers line up, book a demo.