Most digital advertising budgets rest on a comfortable lie. Not a malicious one—more like a story we’ve repeated so many times we’ve forgotten to check if it still holds. The lie is that we can trace a customer’s path from first impression to final purchase and assign credit to each touchpoint with any real precision. We call these attribution models. And they’re broken in ways that waste money and warp decisions.
I’ve spent years deep in digital advertising, watching teams celebrate conversions their reports say came from a particular channel, a specific campaign, a single ad. And I’ve watched those same teams ignore the quiet truth: the number is wrong. Not slightly off. Wrong in ways that are built into how these models are constructed.
This isn’t some abstract gripe. When attribution misfires, money goes to the wrong places. Channels that look efficient on a dashboard get fed. Channels that actually move behavior get starved. The people making these calls aren’t dumb—they’re working with the tools they’ve got. But those tools still rely on assumptions that collapse under any real scrutiny.
What Attribution Models Claim to Do
At the simplest level, attribution models answer one question: Which marketing touchpoint gets credit for a conversion? A user sees a display ad, clicks a search result, opens an email, and then buys something. Should the credit go to the first touch? The last? Split evenly? The model picks a rule and runs with it.
The usual suspects are straightforward:
- Last-click attribution: The final touchpoint before conversion gets 100% of the credit. Still the default in most platforms because it’s easy to measure—and because platforms that tend to sit at the end of the chain (like branded search) have zero incentive to change it.
- First-click attribution: The initial touchpoint takes all the credit. Handy for understanding discovery, but blind to everything that comes after.
- Linear attribution: Every touchpoint gets an equal slice. Feels fair but pretends all interactions carry the same weight, which they don’t.
- Time-decay attribution: Touchpoints closer to the conversion get more credit. Acknowledges recency but still applies a formula that might have nothing to do with actual influence.
- Position-based (U-shaped) attribution: The first and last touchpoints get the bulk, with the remainder split across the middle. Tries to balance discovery and closing, but the percentages are just someone’s best guess.
These are all rules-based models—a human decided the math ahead of time. Then there are data-driven models that use statistical methods to assign credit based on observed patterns. They sound fancier, and in some ways they are, but they inherit most of the same weaknesses.
The Core Problem: We Don’t See What We Think We See
The fundamental issue with attribution is that we’re trying to reconstruct a user’s journey from scraps of data that were never meant to tell a complete story. Every step of that reconstruction adds error.
Tracking Is Incomplete by Design
A user might interact with your brand across six devices, three browsers, two physical locations, and a dozen sessions spread over weeks. Your tracking infrastructure catches fragments. Cookies get blocked or expire. Cross-device matching leans on probabilistic guesses. iOS updates clamp down on what you can track. Android privacy settings do the same. A desktop session at work and a mobile session at home might look like two separate people, or the same person might look like a new user every time they clear their cookies.
What attribution models see isn’t the actual journey. It’s whatever survived the gauntlet. When a model assigns 40% credit to paid search and 30% to email, it’s running math on a dataset that’s missing huge pieces of reality. The arithmetic can be internally sound and still produce a garbage answer because the inputs are garbage.
View-Through Conversions Are a Guess Dressed as a Metric
Most platforms report view-through conversions—when a user sees an ad, doesn’t click, but later converts, and the platform grabs credit. The logic is that the impression influenced the user even without a click. In theory, this captures real influence. In practice, it’s often noise parading as signal.
Serve a million impressions to people who were already going to buy your product, and you’ll record thousands of view-through conversions that had nothing to do with your ads. The platform’s attribution window—often 30 days—makes sure nearly any subsequent conversion gets roped back to the impression. It’s correlation without causation, wrapped in a metric that looks legitimate. Teams see a high view-through conversion rate and pump more spend, reinforcing a loop that inflates the apparent value of display advertising.
Walled Gardens Don’t Share Data
Google, Meta, Amazon—they all run their own attribution systems. They report conversions inside their ecosystems, using their own definitions and attribution windows. A single purchase might get claimed by Google Ads, Facebook Ads, and an email platform all at once, each taking full or partial credit under different rules. When you try to reconcile these numbers, they don’t add up. They can’t, because each platform works with only the data it can see, and each has a built-in incentive to show rosy results.
There’s no neutral referee. No single source of truth that spans every channel. A user’s actual path might include organic social, a review site, a display ad they ignored, a search ad they clicked, and an email that finally pushed them to purchase. Google sees the search click. Facebook sees the social impression and maybe the display ad. The email platform sees the open. None of them see the whole picture, but each will report a conversion with a confidence the data doesn’t support.
Why Last-Click Still Dominates (and Why That’s a Problem)
Despite years of industry chatter about better models, last-click attribution remains the default in most reporting tools and most decision-making. The reasons are practical:
- It’s unambiguous. Every conversion has exactly one last click, so there’s no squabbling over how to split credit.
- It’s easy to measure across platforms because it doesn’t require stitching together multiple touchpoints from different sources.
- It aligns with how performance marketers are incentivized—if your job is to drive conversions, you want credit for the click that immediately preceded the sale.
The problem is that last-click attribution systematically undervalues everything that happens before the final click. Brand awareness campaigns, content marketing, mid-funnel nurturing—all get zero credit unless they happen to be the last touch. A display campaign that introduced a user to your brand six weeks ago gets nothing, while the branded search click right before purchase gets 100%. The search team looks like heroes. The brand team looks like a cost center. Budget shifts accordingly, and the top of the funnel slowly starves, reducing the flow of users who eventually become those last-click conversions.
You can watch the cycle unfold: cut brand spend, branded search volume drops, last-click conversions decline, performance marketers scramble to optimize a shrinking pool, and nobody connects the dots because the attribution model says brand wasn’t contributing anyway. It’s a self-reinforcing error.
Data-Driven Models Aren’t the Fix They Seem
The industry’s answer to rules-based models is data-driven attribution—using algorithms to analyze conversion paths and assign credit based on statistical patterns rather than predetermined formulas. Google Ads offers one version. Analytics platforms offer others. The pitch is that these models learn from actual data, so they’re more accurate.
They are better in some narrow ways. They can spot that certain sequences of touchpoints correlate with higher conversion rates and assign credit accordingly. But they suffer from the same data gaps as every other model. If the underlying data is missing interactions across devices, platforms, and offline touchpoints, the algorithm is learning from an incomplete picture. It might be fancier about distributing credit within the data it has, but it can’t assign credit to touchpoints it never saw.
There’s also a subtler problem: data-driven models optimize for what they can measure, not for what matters. If your tracking is strongest for paid channels and weakest for organic, the model will naturally assign more credit to paid because that’s where the data is cleanest. It’s not that paid is necessarily more influential—it’s that paid is more measurable. The model confuses data availability with causal importance.
Selection Bias in Every Conversion
Attribution models assume that the users who convert are representative of all users exposed to your advertising. They’re not. The people who click on your ads are already more interested than those who don’t. The people who see an ad and later search for your brand were probably already aware of you. When you compare conversion rates between people who saw an ad and people who didn’t, you’re not comparing apples to apples—you’re comparing people with different baseline levels of intent.
This gets especially distorting for retargeting campaigns. Retargeting shows ads to people who’ve already visited your site, so they’ve already raised their hand. Their conversion rate will naturally be higher than cold audiences, but attribution models will credit the retargeting ad without accounting for the fact that these users were more likely to convert anyway. The retargeting campaign looks wildly efficient, so budgets shift there. Meanwhile, the campaigns that brought users to the site in the first place get less credit because their conversions happen later, after retargeting has wedged itself into the path.
The Incrementality Gap
What we actually want to know is not “which touchpoint was present” but “which touchpoint caused the conversion that wouldn’t have happened otherwise.” That’s incrementality—the additional conversions generated by a specific marketing activity, above and beyond what would have occurred organically. Attribution models don’t measure incrementality. They measure correlation, not causation.
Measuring incrementality requires controlled experiments: holdout groups, geo-split testing, randomized exposure. These are harder to run than simply looking at an attribution report, so most organizations don’t do them. Instead, they trust that the attribution model’s credit assignment reflects real causal impact. It doesn’t. A channel can have high attributed conversions and near-zero incrementality if it’s mostly intercepting users who would have converted through another path. A channel can have low attributed conversions and high incrementality if it’s reaching new audiences that other channels miss.
This is the central disconnect in digital advertising measurement. We’ve built an entire optimization ecosystem around a metric that doesn’t answer the question we think it answers. Attribution tells us what happened before a conversion. It doesn’t tell us what caused the conversion.
What to Do Instead
The solution isn’t to abandon measurement—it’s to stop treating attribution as the single source of truth and start treating it as one signal among many, with known limitations. There are practical approaches that produce better decisions even if they don’t produce perfect numbers.
Run Incrementality Tests Where They Matter Most
For your largest channels and campaigns, run controlled experiments. A simple geo-split test—showing ads in one region and not in another, then comparing conversion trends—can reveal whether a channel is actually driving incremental results. These tests aren’t perfect either; regional differences can muddy the results. But they’re orders of magnitude closer to the truth than raw attribution numbers.
Use Media Mix Modeling for Strategic Allocation
Media mix modeling (MMM) takes a top-down approach: it looks at total sales over time and uses statistical methods to estimate how different marketing channels contributed, controlling for seasonality, pricing changes, and other external factors. MMM doesn’t rely on user-level tracking, so it’s not tripped up by cookie blocking or cross-device fragmentation. It’s slower and less granular than attribution, but for strategic budget decisions—how much to spend on TV versus search versus display—it’s often more reliable.
Triangulate Multiple Data Sources
Don’t lean on a single attribution model. Compare last-click, first-click, and data-driven models to understand the range of possible interpretations. If all models agree that a channel is performing well, that’s stronger evidence than any single model’s output. If they disagree wildly—which they often do—that disagreement itself is information. It tells you that the data isn’t conclusive and that decisions should be made with more caution.
Build Internal Benchmarks, Not Just Platform Reports
Platform-reported conversions are designed to make the platform look effective. Build your own conversion tracking that uses consistent definitions across channels, ideally with server-side tracking that’s less vulnerable to browser restrictions. Set a standard attribution window that you control, rather than accepting each platform’s default. This won’t solve the fragmentation problem, but it reduces the self-serving bias baked into platform numbers.
The Bottom Line
Attribution models are comfortable fictions. They give us tidy numbers we can drop into spreadsheets and present in meetings. They let us feel like we understand what’s happening. But the tidy numbers are wrong, and the confidence they create leads to worse decisions than honest uncertainty would.
The fix isn’t a better model—the model-building approach itself is the problem. As long as we’re trying to reconstruct individual user journeys from fragmented tracking data, we’ll be solving a puzzle with half the pieces missing and guessing at what the picture should look like. The better path is to combine multiple measurement approaches, acknowledge the uncertainty, and make allocation decisions that don’t depend on false precision.
Stop asking “Which channel gets credit?” and start asking “What would happen if we turned this off?” The second question is harder to answer, but it’s the one that leads to better spending.
Frequently Asked Questions
Why doesn’t Google Ads attribution match Google Analytics attribution?
They use different attribution models by default, different attribution windows, and different methods for counting conversions. Google Ads typically defaults to last-click within its own ecosystem, while Google Analytics might use a different model and can include touchpoints from other channels. They also handle cross-device and view-through conversions differently. The discrepancy isn’t a bug—it’s two systems measuring different things and calling them both “conversions.”
Is there an attribution model that actually works?
No single model works on its own. Every model has blind spots, and those blind spots are large enough that treating any one model as truth leads to misallocation. The most useful approach is to run incrementality tests for major channels, use media mix modeling for strategic planning, and treat attribution reports as directional signals rather than precise numbers. If you need a single model to report on, data-driven attribution tends to be less wrong than rules-based models, but “less wrong” isn’t the same as “right.”
How do privacy changes affect attribution?
Privacy changes—iOS tracking restrictions, third-party cookie deprecation, browser fingerprinting prevention—all reduce the amount of user-level data available for attribution. This makes cross-device and cross-session tracking less reliable, which means attribution models are working with even less complete data than before. The trend is toward aggregate measurement methods like media mix modeling and conversion lift testing, which don’t require stitching together individual user journeys. This is a structural shift, not a temporary disruption.
Should small businesses even bother with attribution?
Yes, but with realistic expectations. For a small business with limited budget and simple marketing, last-click attribution combined with asking customers how they heard about you can be sufficient. The key is not to over-optimize based on attribution data alone. If you’re spending $5,000 a month on ads, you probably don’t need a complex attribution stack. Focus on tracking revenue by channel at a high level and running simple tests—like pausing a channel for two weeks and watching what happens to overall sales.
