A marketing director runs the same campaign data through three attribution models and gets three different answers about which channel drove revenue. Paid search looks like the hero in one report, organic content in another, and email in the third. Nothing changed except the math.

This isn't a data quality problem—it's a modeling problem. Attribution models aren't measuring reality directly. They're applying logical frameworks to assign credit for outcomes that involve multiple touchpoints over time. Each framework encodes different assumptions about how influence actually works.

Understanding why these models disagree matters because attribution shapes budget decisions worth millions. Teams that treat attribution as a single source of truth tend to over-invest in whatever the chosen model favors. Teams that understand the underlying mechanics use multiple models strategically, matching the analytical approach to the business question being asked.

Last-touch attribution assigns 100% of conversion credit to the final interaction before purchase. If a customer clicks a Google ad and converts, that ad gets full credit—regardless of the twelve previous touchpoints. It's computationally simple and dominates because most analytics platforms default to it.

First-touch flips the logic, crediting the initial channel that introduced the customer. Linear attribution distributes credit equally across all touchpoints, while time-decay models weight recent interactions more heavily than older ones. Position-based models (often called U-shaped) give 40% credit each to first and last touches, splitting the remaining 20% across middle interactions.

Algorithmic attribution—including Markov chains, Shapley values, and machine learning approaches—takes a fundamentally different path. Rather than imposing fixed rules, these models analyze conversion patterns across thousands of customer journeys to estimate each channel's incremental contribution. A Markov model, for instance, calculates how conversion probability changes when a specific channel is removed from the path.

The mathematical differences cascade into operational differences. Rule-based models can be calculated in a spreadsheet and explained in a meeting. Algorithmic models require more data, more compute, and more trust in opaque calculations—but they capture interaction effects that rule-based approaches miss entirely.

Takeaway

Every attribution model is a theory of influence dressed up as math. The numbers don't lie, but the assumptions underneath them might not match how your customers actually behave.

Last-touch attribution carries a hidden assumption: that the closing channel deserves all credit because it sealed the deal. This systematically rewards bottom-funnel activities like branded search and retargeting while starving the awareness channels that created demand in the first place. Teams optimizing for last-touch often watch their funnels mysteriously dry up after months of cutting upper-funnel investment.

First-touch inverts the bias, overvaluing initial discovery channels while ignoring the nurture activities that actually convert interest into action. Linear attribution assumes every touchpoint contributes equally—a tidy idea that ignores the obvious reality that some interactions are pivotal while others are forgettable.

Time-decay models embed the assumption that influence fades predictably, which works for impulse purchases but distorts long consideration cycles common in B2B. Algorithmic models reduce these specific biases but introduce new ones: they're only as good as the data they observe, meaning they systematically undervalue offline channels, dark social, and brand effects that don't generate trackable clicks.

Every model has a worldview, and that worldview shapes which channels look efficient. A CMO who switches from last-touch to algorithmic attribution isn't getting closer to truth so much as adopting a different bias profile. The honest question isn't 'which model is correct' but 'which biases am I willing to live with given my strategic priorities.'

Takeaway

Choosing an attribution model is choosing which channels you'll systematically undervalue. There is no neutral choice—only informed ones.

Match the model to the decision. If you're optimizing direct-response campaigns with short consideration windows, last-touch or time-decay attribution often suffices—the signal-to-noise ratio in longer journeys isn't worth the modeling complexity. If you're allocating budget across upper and lower funnel investments, you need a model that credits awareness activities, which means moving beyond last-touch.

Data availability constrains your options more than most teams admit. Algorithmic attribution requires substantial conversion volume—typically thousands of converting paths monthly—to produce stable estimates. Below that threshold, sophisticated models produce sophisticated-looking noise. Smaller businesses are often better served by position-based models that encode reasonable assumptions without overfitting to sparse data.

Use multiple models simultaneously rather than searching for the one true answer. Compare last-touch and first-touch reports side by side: channels that look strong in both are genuinely productive, while channels that only appear in one are playing specific funnel roles. Triangulating across models reveals more than any single view.

Validate attribution outputs against incrementality testing whenever possible. Geo-holdout experiments, conversion lift studies, and media mix modeling provide ground-truth checks that pure attribution cannot. When attribution and incrementality testing disagree, trust the experiment—it's measuring causation while attribution is measuring correlation dressed up in business clothing.

Takeaway

Attribution tells you what happened alongside conversions. Incrementality testing tells you what actually caused them. The mature analytics function uses both and never confuses the two.

Attribution models conflict because they're answering subtly different questions while appearing to answer the same one. The model that maximizes apparent ROI on paid channels isn't necessarily the model that allocates budget most effectively across your full marketing mix.

The strongest analytics teams treat attribution as a portfolio of perspectives rather than a single source of truth. They know their models' biases, validate findings through experimentation, and resist the temptation to declare any single number definitive.

Budget decisions worth millions deserve more than a default setting in a reporting tool. Choose your attribution approach as deliberately as you choose your marketing strategy—because in practice, the two are inseparable.