Most attribution mistakes do not announce themselves. They show up as a slightly-too-confident dashboard, a quarterly review where the numbers almost reconcile, and a budget that keeps drifting toward the channels that look efficient on screen rather than the ones doing the work. The cost is rarely a single bad decision. It is a slow leak: a few percent of spend lost each month to over-credited channels, miscounted conversions, and tagging drift that nobody has time to clean up.
The good news is that the leaks tend to come from a small, repeatable list. Below are eight attribution mistakes we see most often when we audit a marketing team’s measurement setup, what each one actually costs, and the cleanest first fix. None of them require an attribution overhaul to address. Most can be fixed inside a sprint.
1. Treating Last-Click as the Only Truth
Last-click attribution credits the final touchpoint before a conversion and ignores every step that came before. It is fast, simple, and embedded in nearly every default ad-platform report. It is also the single most common reason upper-funnel channels get defunded.
A customer who reads three blog posts, watches a YouTube video, sees four retargeting ads, and finally clicks a branded search link before purchasing has eight touchpoints in their journey. Last-click hands all the credit to branded search and zero to anything that opened the door. Reallocate budget on that signal long enough and you starve the channels that fill the funnel, then wonder why branded search efficiency collapses six months later.
IAB Europe’s 2025 Sales Incrementality Measurement guidelines explicitly list last-click and last-touch platform-tracked methods among the approaches that should not be considered incrementality measurement, because they reveal correlation rather than causation. The fix is not to abolish last-click; it is to stop treating it as the answer. Run multi-touch attribution alongside it, compare the two, and make budget decisions where the models agree.
2. Adding Up Walled-Garden Conversions
Meta says it drove 1,200 conversions. Google Ads says 900. TikTok says 450. Add them up and you have 2,550 attributed conversions against an actual order count of 1,400. The math does not add up because each platform claims credit for conversions the others also claim, using its own attribution window and its own definition of an exposed user.
Industry observation puts the typical multi-platform overlap factor at 1.5 to 3 times actual conversions for consumer brands running across several walled gardens. Amazon, Meta, and Google each operate self-attributed measurement systems that were never designed to deduplicate against one another. AdExchanger has covered this dynamic at length: walled-garden platforms produce internally-consistent self-attributed numbers that, summed naively, will always exceed reality.
The fix is to stop summing platform-reported conversions across channels. Pick one source of truth that sees the whole journey, whether that is your warehouse, a dedicated attribution layer, or your CRM, and use it for cross-channel comparisons. Reserve the platform dashboards for in-platform optimization, where their internal attribution at least stays consistent with itself.
3. Confusing Attributed ROAS With Incremental ROAS
Attributed ROAS divides credit for sales that already happened. Incremental ROAS estimates how many of those sales would have happened anyway. They are not the same number, and the gap is rarely small.
Branded search is the textbook example. A retailer might see a 12x attributed ROAS on its branded campaign. Pause it for two weeks and revenue drops by 15 percent, not the 90 percent the dashboard implied. Most of the spend was paying to win clicks the brand was already going to capture organically. True incremental ROAS on that line was closer to 1.8x. The same dynamic shows up in retargeting, where the audience was already coming back, and in many forms of brand safety media that sit on top of demand the brand had already created.
This is the most expensive mistake on the list because it scales with budget. The bigger the line item, the bigger the gap between reported and incremental, and the more money quietly buys revenue you would have earned anyway. The fix is structural: read attributed ROAS for relative trends, and ground-truth the absolute level with periodic incrementality testing, as covered in our reported vs true incremental ROAS guide.
4. Letting UTM Tagging Drift
Attribution is downstream of data quality. If half your Google Ads URLs use utm_source=google and the other half use utm_source=Google, your reporting tool treats them as two separate sources, splits the metrics, and quietly under-reports both. The same fragmentation happens with facebook and fb, with handlers that drop UTM parameters on redirect, and with campaigns that use free-text values typed by whoever happened to be building the link that day.
GA4 treats UTM values as case-sensitive. A standardized naming convention, enforced through dropdowns or templates rather than free text, is what stops the split. Industry coverage of UTM data quality consistently puts the attribution-accuracy lift from a properly enforced naming convention in the high-twenty-percent range, simply by reuniting traffic that was being shattered into ghost line items.
The fix is unglamorous but quick: define a taxonomy (lowercase, kebab-case, finite source and medium values), audit your live links, and stop allowing free-text UTM entry for campaigns that touch revenue. This is exactly what the UTM Manager feature in Attriqs is built to enforce, and it is the cheapest measurement upgrade most teams can make.
5. Trusting Default Attribution Windows That Flatter the Platform
Every ad platform sets its default attribution window in the way that produces the largest claimable conversion count. Meta defaults to 7-day click and 1-day view-through, which means it can claim a conversion when someone scrolled past an ad in their feed yesterday and bought today. Google Ads has shifted to data-driven attribution as the default, which spreads credit across earlier touches but still favours its own platform’s surfaces.
The result is that platforms can report substantially more conversions than independent analytics tools see for the same time window. Industry analysis of cross-platform reconciliation routinely finds Meta reporting in the order of 25 percent higher conversion counts than warehouse-level analytics, with view-through credit accounting for a meaningful share of the gap. Apple’s App Tracking Transparency framework, which restricts cross-app tracking on iOS, has widened the gap further by pushing platforms toward modelled rather than observed conversions.
You do not have to throw out view-through entirely; it is a real signal in some categories. You do have to know how much of your reported ROAS depends on it. Tighten the windows to click-only for one reporting period, compare the result against your default view, and you will see exactly how much credit is structural rather than causal.
6. Ignoring Offline Conversions
Phone calls, showroom visits, in-store purchases, and buy-online-pick-up-in-store orders are still revenue. They just do not show up in a digital-only attribution stack. The result is that channels which drive offline behaviour, things like local search, paid social with a phone-call objective, radio, and out-of-home, get systematically under-credited and quietly defunded.
This is most acute in services businesses (home services, healthcare, legal, automotive) where a meaningful share of conversions still happen by phone. It also matters in retail, where store visits driven by digital ads close at a register that the ad platform never sees. If your attribution model cannot tie a phone call back to the ad that drove it, you are scoring a game where half the goals do not count.
Closing this gap requires two pieces. First, dynamic number insertion (DNI) so that every call carries the click and channel that produced it; our DNI call tracking guide covers the mechanics in depth. Second, a way to push qualified offline conversions back into your ad platforms (server-side conversion APIs in Meta, Google, and LinkedIn) so the bidding algorithms can optimize against the conversions that actually matter rather than the ones they happen to see.
7. Optimizing One Campaign in Isolation
A Meta prospecting campaign is showing a 6x ROAS. The team scales the budget two-fold the next week. Performance collapses. The diagnosis is rarely “the algorithm is broken.” It is usually that the campaign was riding warm audiences created by other channels, a recent email send, an organic post that went mildly viral, a paid-search push the previous month, and the new spend is being served to colder audiences that never had those priming touches.
In a multi-channel funnel, no campaign is independent. Channels prime each other. A channel that looks like the closer in attribution is often only doing the closing because something else did the opening. Scaling the closer in isolation does not multiply the openings; it just means more spend on the same finite warm pool, then on increasingly cold prospects, with predictably diminishing returns.
The fix is to evaluate channel performance against the full media mix rather than against itself. Marketing mix modelling and cohort-style analysis are both useful here; either one will tell you when a campaign’s apparent efficiency is borrowed rather than earned, and both will save you from confidently scaling a campaign into a bad week.
8. Picking One Attribution Model and Defending It
Every attribution model encodes assumptions. Last-click assumes the final touch deserves all the credit. Linear assumes every touch deserves equal credit. Time-decay assumes recent touches matter more. Position-based assumes the first and last touches are special. Data-driven assumes you have enough conversion volume for the model to converge on something stable. Each is right in some situations and wrong in others.
Picking one model on day one and defending it from then on is indistinguishable from guessing. It will produce internally-consistent numbers that drift further from reality the longer you use them. The minimum discipline is to run several models in parallel, read where they disagree, and use the disagreements as a signal about where your attribution is most uncertain. The IAB has held this position since its original Digital Attribution Primer: attribution is a portfolio problem, not a single-model problem.
The next layer up is to pair models with experiments. A multi-touch model tells you which channels move people through the journey. A holdout or geo test tells you which channels actually caused the revenue. Neither alone is sufficient; together they give you something you can defend to a finance team without flinching. Our incrementality testing guide covers the experimental side in detail.
Frequently Asked Questions
Which of these mistakes is most expensive?
Mistake number three (confusing attributed with incremental ROAS) is usually the most expensive in absolute terms because it scales with budget. The larger the line item, the larger the dollar gap between reported and true ROAS, and the more spend silently buys revenue you would have earned anyway. For a brand spending $500,000 a month on paid media, even a modest ten-percent gap between reported and incremental ROAS is $50,000 a month in misallocated spend.
Do I need an attribution platform to fix these?
For mistakes one, two, three, six, and eight, yes; you need something that sees the full journey rather than each platform’s self-reported view. For mistakes four, five, and seven, the fix is mostly process: a UTM standard, a tighter default window, and a habit of reading any campaign in the context of the channels priming it.
How do I know if my reported ROAS is inflated?
The fastest signal is a reconciliation: pick a thirty-day window, sum up every platform’s reported revenue, and compare against your actual booked revenue. If the platform sum is more than 1.3x your actual revenue, you are looking at material overlap and overcrediting. From there, run a holdout test on the largest line item to size the incremental gap.
Do these mistakes apply to B2B as much as e-commerce?
Yes, with different shapes. B2B journeys are longer (often 60 to 180 days) so attribution windows that flatter platforms (mistake five) are even more distorting. Offline conversions (mistake six) are especially under-counted in B2B because so many opportunities still close through sales conversations rather than online checkouts.
Where should a small team start?
If you can only fix one thing this quarter, fix UTM tagging (mistake four). Clean tracking is the foundation everything else depends on. Sophisticated models applied to fragmented data produce sophisticated nonsense. Once your tracking is clean, prioritize by spend: tackle the mistake that affects your largest line item first.
What This Looks Like Cleaned Up
A measurement setup without these eight mistakes does not look exotic. It looks like a single source of truth that sees the full journey across channels, a UTM taxonomy that everyone follows, attribution windows that are tight enough not to flatter any single platform, offline conversions stitched in via DNI and server-side APIs, and a habit of running multiple models side by side and ground-truthing them with periodic incrementality tests. Most of those pieces are policy decisions, not platform decisions. The platform just makes following them easier.
If you want to see what that stack looks like in one place, with multi-touch attribution, marketing mix modelling, DNI, UTM governance, and incrementality testing wired together rather than bolted on, get in touch and we will walk you through it.