Meta's Own Study Found Rules-Based Attribution Undervalues Meta by 31%

Meta's own study found that, compared with incrementality measurement, the median advertiser using rules-based attribution undervalues Meta by 31% (PPC Land). On November 2, Smartly began selling a tool that puts a second estimate right next to the platform's own number (PPC Land). Only customers Smartly calls eligible can get it, and no price was published (PPC Land). Here is what I would do with that if I owned the ad account.

What Smartly put on the screen

Smartly says its platform serves more than 800 brands and oversees over $7 billion in media spend (PPC Land). On November 2, 2026, it started selling a premium add-on named Intelligence Pro (PPC Land). The add-on lives inside the same tool its customers use to plan and launch campaigns across Meta, TikTok, Amazon and other networks (PPC Land).

The piece that matters for anyone who sells for a living is INCRMNTAL Connect. Smartly says it adds "an incrementality-based view of media contribution alongside platform-reported performance" (PPC Land). In plain terms, a second estimate of what the ads caused sits next to the count the platform reports about itself (PPC Land).

The package also includes three planning tools. Reach Planner lets a team line up media scenarios and forecast reach and frequency across channels before launch (PPC Land). Headroom Analysis aims to show where extra spend may still bring growth and where returns may start to level off (PPC Land). Benchmarks shows how a campaign compares with similar work in the same industry and region, and for the same platform and goal, using pooled and anonymized data (PPC Land).

A sample panel on Smartly's product page shows an average day of $1,223 in spend and a $4.89 cost per acquisition. It puts CPA ranges beside three confidence levels, from $4.41 to $5.21 at 95% down to $4.79 to $5.04 at 50%. The conversion figure stays at 235 to 277 no matter which level you pick. PPC Land points out that intervals of different widths should not produce the same conversion range, so those figures look like placeholders (PPC Land).

What Smartly left out may matter as much as what it announced. It did not say which customers qualify as eligible, did not publish a price, did not report how many advertisers were in the testing before launch, and did not release checks on the incrementality estimates (PPC Land).

Hearst Newspapers is the only customer named in the release. Sierra Smith, a senior paid social marketing manager there, said her team used to pull data from several platforms and analyze it by hand. She said Headroom lets them weigh budget options from past results and cut the manual digging. Look closely at the benefit she describes: faster analysis, not proof of better sales. The release attaches no numbers to what Hearst got from it (PPC Land).

So why does a second number need to exist at all?

The reason a second number exists

Ad networks grade their own numbers. That is the gap a second opinion sells into (PPC Land).

Meta's own research points the same way. Its Suite of Truth white paper covered 307 studies from 54 advertisers, run from March 2022 to November 2024 (PPC Land). At the median, the paper found, rules-based attribution gives Meta 31% less credit than incrementality measurement does (PPC Land).

The error can run both ways. Meta's default setting counts a click in the past seven days and a view in the past day. In 2025, practitioners argued that a marketer claiming a 10 ROAS might be doing worse than one claiming a 2, since credit for a view can attach to orders those buyers would have placed anyway (PPC Land).

Measurement models can fail in their own ways, too. A study published in Marketing Science in 2023 by Gordon, Moakler and Zettelmeyer, discussed on PPC Land, looked at 663 Facebook experiments. Model-based estimates of upper-funnel lift put the median at 83%, while the randomized tests put it at 29% (PPC Land). That paper tested particular observational methods, not the one INCRMNTAL uses, so it says nothing about how Smartly's model performs (PPC Land).

Many advertisers are already sorting through answers that disagree. May 2025 research from Kantar found the typical advertiser running 3.8 measurement tools, and 55% of them had seen those tools give conflicting answers (PPC Land).

Smartly's chief product officer, Maxwell Tang, framed the launch around decisions rather than data. As AI speeds up campaign work, he said, "the quality of each decision matters more" (PPC Land).

Two other facts set the stage. In May 2025, at Google Marketing Live, Google cut the minimum budget for an incrementality test in Google Ads to $5,000 (PPC Land). And Smartly's March 16, 2026 letter of intent named INCRMNTAL as a purchase target, a Tel Aviv firm with 11 to 50 employees; Smartly's press page later listed the deal as completed on May 7, 2026 (PPC Land).

Which leaves the owner's question: whose number sets your budget?

What this means for an owner who sells

My read: ignore the enterprise packaging. The real story is that the number on your budget screen now has a competitor. And you may have nobody assigned to referee the fight.

Picture a fitness studio putting a few thousand a month into Instagram to fill trial classes. The platform can take credit for bookings that follow a click or a view. Some of those people were coming anyway. They drove past the sign. A member dragged them in. They searched the studio name. Under the default window, an order that follows a view can still be credited to the ad (PPC Land). The owner sees a cost per lead and adds budget. Incrementality asks the blunter question: how many of these would have happened without the ad at all? That is the number now sitting beside the platform's (PPC Land).

Picture a B2B team using Meta and LinkedIn to book demos, with deals closing weeks later after a rep works the account twice. Measurement tools can disagree badly in that setup. A preprint from August 2026 tied to a Zalando researcher ran the numbers on simulated data, and a standard marketing mix model put paid search at 10.61x ROAS when the real figure was 4.20x (PPC Land). The estimate only came close to the truth once geo-experiments were part of the model (PPC Land). My take: when your team argues over a number that far off, they are not debating marketing. They are guessing out loud.

Confidence is not the same as measurement. An Incubeta survey of marketing leaders in the UK and the US, published in May 2026, found 70.4% felt confident their budgets were being spent well, 41.6% admitted that part of the spend was not delivering full value, and only 34.4% used one unified measurement approach. The survey did not disclose its sample size (PPC Land).

Here is the loop I would leave open for you. If the second estimate comes from inside the tool you buy ads with, who checks it? PPC Land raises that point directly: when the model lives inside the buying tool and belongs to the company selling that tool, it can steer budgets just by being the default on the screen (PPC Land). That is why my next steps are about tests, not tools.

How to use this week

Seven steps. None of them require buying anything.

1. Write the platform's number down somewhere else. Screenshot the campaign report on the first of the month and paste it into a doc your sales manager can open, so the dashboard is not the only copy of the truth you keep.

1. Ask what the number was compared to. A platform-reported number and an incremental number answer different questions, and Meta's own study found rules-based attribution gives Meta 31% less credit than incrementality measurement does (PPC Land). If your agency or your rep cannot say which one they are quoting, you have found your problem.

1. Run one real holdout instead of arguing. A holdout keeps a randomly selected group away from the ads, and that group becomes the baseline (PPC Land). Since May 2025, one test in Google Ads can start at a $5,000 budget (PPC Land). If that is too much in one month, save for it. One clean test beats a year of dashboards.

1. Split new from repeat. Go through last month's leads or booked demos by hand and mark the ones who were already in your system. Repeat buyers are the cases where the platform's credit can go wrong (PPC Land).

1. Check the attribution window before you judge a campaign. Meta's default counts a click in the past seven days and a view in the past day (PPC Land). Shorten the window for a week and see which campaigns still look good. My read: the ones that fall apart were taking credit, not making sales.

1. Decide in advance which number moves money. Write one line in your budget doc: if the platform and the test disagree, the test wins. PPC Land asks a blunter version: when platform figures, a model and an in-house model disagree, which number gets to move the budget (PPC Land)? Answer that before the meeting, not during it.

1. Make the numbers show the trail. Use your AI tool of choice to reconcile the two lists. Feed it your last 90 days of leads plus your front desk's log of who actually showed up for a trial, and ask it to list every lead the platform took credit for that never walked in. For a B2B team, ask for a one-page sheet that lines up reported leads against closed deals by campaign and flags any campaign credited with deals that were already in the pipeline. Then ask for three questions a rep should raise before trusting a campaign's cost per lead, and what a bad answer sounds like.

Mistakes to avoid

• Don't move budget the day a second number appears. One modelled estimate is an argument, not a verdict. Wait until it repeats.
• Don't treat a model like a test. A holdout is a randomized comparison. A model like INCRMNTAL's infers a baseline from how spend and results move over time (PPC Land). Those are not the same evidence.
• Don't add a fourth tool to settle a fight between three. The average advertiser already runs 3.8 measurement tools, and 55% have seen them disagree (PPC Land). More dashboards do not end arguments.
• Don't let the company selling you the ads be the only one grading them. That gap is why the second number exists, and it only helps if someone with authority actually reads it (PPC Land).