Most marketing teams I work with trust their attribution reports the way pilots trust altimeters. The number says 47 conversions came from paid social, so paid social gets credit for 47 conversions. Clean, defensible, ready for the board deck.
Except it's wrong. And the gap between what your dashboard claims and what actually happened is large enough to blow your CAC payback model.
NP Digital's February 2026 study of 602 websites and 500 marketers quantified something operators have suspected for years: the delta between attributed conversions and incremental lift varies wildly by channel, and in most cases, attribution over-credits by double digits. Branded search, the darling of last-touch models everywhere, shows the widest gap. Display and retargeting aren't far behind.
This isn't a measurement curiosity. It's a capital allocation problem. If you're shifting budget based on attributed ROAS without adjusting for incrementality, you're systematically over-investing in channels that harvest demand you already created and under-investing in the ones that actually generate it.
The Branded Search Illusion
Branded search is the clearest example of attribution theater. Someone sees your LinkedIn ad, reads a case study, talks to a peer, and three weeks later types your company name into Google. Your attribution platform credits the branded search click. Your paid search manager celebrates. Your CFO approves more budget for the channel that "converts best."
But that conversion was going to happen anyway. The buyer already decided. Branded search didn't create demand; it intercepted it at the finish line.
The NP Digital data suggests branded search over-attribution runs somewhere between 30 and 50 percent depending on the vertical. That means for every 100 conversions your dashboard assigns to branded search, only 50 to 70 were actually incremental. The rest would have found you through direct navigation, organic search, or a bookmark.
This matters because branded search often looks like your most efficient channel on a cost-per-acquisition basis. It's cheap, it converts well, and it's easy to scale. But if half those conversions aren't incremental, your true CPA just doubled. Suddenly that "efficient" channel is mediocre, and the upper-funnel spend you've been cutting to fund it was actually doing the heavy lifting.
Retargeting's Diminishing Returns
Retargeting suffers from a similar problem, though the mechanics differ. The pitch is compelling: show ads to people who already visited your site, and they'll come back and convert. Attribution platforms love retargeting because the conversion window is short and the touchpoint is recent.
But retargeting also over-credits. Many of those users were already in an active buying process. They visited your pricing page, got distracted, and came back the next day. The retargeting ad they saw in between may have reminded them, or it may have been irrelevant noise they ignored. Attribution can't tell the difference.
The incrementality gap for retargeting typically runs 15 to 25 points. Not as severe as branded search, but enough to distort your channel mix if you're optimizing purely on attributed performance.
The practical implication: retargeting works, but it works best at modest frequency caps and tight audience definitions. The marginal return on the 15th impression to someone who already visited your demo page is close to zero. Your attribution model will still credit it, but your incrementality tests won't.
What Incrementality Testing Actually Requires
Knowing that attribution over-credits is useful. Knowing by how much, for your specific business, requires actual measurement. And incrementality testing is harder than most vendors admit.
The gold standard is a geo-holdout or audience-holdout experiment. You suppress ads to a randomly selected group and compare their conversion rate to the exposed group. The difference is your incremental lift. Simple in theory, painful in practice.
The pain comes from three places. First, you need enough volume to detect a statistically significant difference. If your baseline conversion rate is 2 percent and you're looking for a 10 percent lift, you need thousands of users in each group to hit 80 percent power. Most B2B companies don't have that kind of volume in a single channel.
Second, you need clean holdouts. If your suppressed group sees your brand through other channels, your test is contaminated. In a world of cross-device tracking gaps and walled gardens, true suppression is hard to achieve.
Third, you need patience. Incrementality tests take weeks to run and weeks to analyze. The marketing team that's used to daily dashboard updates finds this timeline excruciating.
None of these problems are unsolvable, but they require planning, statistical rigor, and executive air cover. You can't run incrementality tests on a whim, and you can't run them on every channel simultaneously.
A Practical Measurement Stack
For most B2B marketing teams, the answer isn't to abandon attribution or to run incrementality tests on everything. It's to build a layered measurement approach that uses each method where it's strongest.
Attribution stays useful for tactical optimization within a channel. Which ad creative converts better? Which landing page? Which audience segment? These are relative comparisons where the over-crediting bias cancels out. If Creative A gets 20 percent more attributed conversions than Creative B, it's probably actually better, even if the absolute numbers are inflated.
Incrementality testing is for strategic questions. Should we increase spend on this channel? Should we cut it entirely? Is our retargeting frequency too high? These are the decisions where the attribution gap matters, and where a well-designed holdout test pays for itself many times over.
Marketing mix modeling (MMM) fills the gap for channels that are hard to test directly. Brand campaigns, sponsorships, and offline media don't lend themselves to clean holdouts, but they do show up in aggregate demand curves. MMM won't tell you which podcast ad worked, but it will tell you whether podcasts as a category are contributing to pipeline.
High-growth companies tend to use all three methods, not because they have unlimited measurement budgets, but because each method answers different questions. The teams that struggle are the ones that pick a single source of truth and defend it past the point of usefulness.
The CFO Conversation
Here's where this gets practical. Your CFO doesn't care about attribution methodology. They care about whether marketing spend is generating returns that justify the investment.
If you walk into a budget review armed only with attributed ROAS, you're vulnerable. Any CFO who's done their homework (or talked to a board member who has) knows that attribution inflates performance. They'll discount your numbers, and you won't know by how much.
If you walk in with incrementality data, even partial data from a few key channels, you're having a different conversation. You can say: "Attributed ROAS on paid social is 4x, but our holdout test showed incremental ROAS of 2.8x. We're budgeting to the incremental number." That's a defensible position. It shows you understand the gap and you're not hiding behind vanity metrics.
The 24-point average gap between attributed and incremental performance isn't a scandal. It's a known limitation of a useful but imperfect tool. The scandal is pretending the gap doesn't exist and making million-dollar allocation decisions on numbers you know are wrong.
Run the tests. Adjust the models. Show the math. Your board will thank you, and your budget will survive the next efficiency review.
