Why Marketing Attribution Models Disagree With Each Other

Marketing attribution models disagreeing with each other isn’t a bug in any one of them. Each model applies a different rule for assigning credit across the same underlying customer journey, so disagreement between models is the expected outcome, not a sign one of them is broken.

• Touchpoint weighting rule: the formula a model uses to divide credit for a conversion across every interaction that preceded it.
• Lookback window: how far back before a conversion a model will still count an interaction as contributing to it.
• Cross-device stitching: whether a model can recognize the same person across multiple devices, or treats each device as a separate, disconnected visitor.

1. Models are built to disagree, by design

HubSpot’s own attribution reporting documentation states plainly that comparing multiple models on the same data is informative precisely because they disagree: a channel that looks like the best performer under first-touch attribution can look mediocre under last-touch analysis on the identical set of customer interactions. Neither reading is wrong. Each is answering a different question about the same journey.

Google Ads’ own help documentation confirms the mechanism: an attribution model is a rule, or set of rules, that determines how credit for a conversion gets divided among the touchpoints along the path to it. Change the rule and the same path produces a different credit assignment, which is the entire source of the disagreement.

2. Confidence in last-click has collapsed, but it’s still the default

A Snap and eMarketer Media Measurement Survey of 282 US marketers, fielded June-July 2024 and published October 9, 2024, found only 21.5% of respondents confident that last-click attribution is a reasonably accurate reflection of a platform’s long-term impact on the business. 74.5% said they’re either actively moving away from last-click attribution or would like to.

That gap between stated skepticism and continued use matters because last-click remains the default reporting view in many analytics tools, meaning the model most marketers trust least is often the one they’re still looking at first.

Metric	Value
Marketers confident last-click reflects real impact	21.5%
Marketers moving away from or wanting to move away from last-click	74.5%

3. What actually causes the disagreement

Three mechanical differences between models account for most of the disagreement in practice:

• Touchpoint weighting rule: last-click assigns 100% of credit to the final interaction; linear splits it evenly across every touchpoint; time-decay weights recent touchpoints more heavily. Same data, three different answers.
• Lookback window: a 7-day window and a 90-day window can include entirely different sets of touchpoints for the same conversion, changing which channels appear in the analysis at all.
• Cross-device stitching: a model that can’t recognize the same person across a phone and a laptop will undercount that person’s earlier touchpoints, understating whichever channel first reached them.

4. A practical order for choosing a model

Rather than debating which model is “correct,” work through this sequence:

1. Pick one model as the primary standard for budget decisions, and document the choice.
2. Check that model’s lookback window matches the actual length of the typical sales cycle.
3. Confirm cross-device stitching is enabled if customers plausibly research on one device and convert on another.
4. Use a second model only as a cross-check for directional agreement, not to relitigate the primary model’s numbers every reporting cycle.

This is the same measurement discipline behind Performance Marketing Metrics That Matter More Than Click-Through Rate and Auditing Your Funnel Before You Add Another Marketing Tool: define the measurement standard before using it to make a spending decision. Founders untangling which model to trust for their own funnel are welcome to start with our Performance Marketing service, which begins with exactly this kind of attribution audit.

What the evidence doesn’t yet support

Two claims worth resisting. First, that multi-touch attribution is simply “more correct” than last-click: multi-touch models require accurate cross-device stitching and a well-chosen lookback window to be reliable, and a poorly configured multi-touch model can be less accurate than a well-understood last-click view, not more. Second, that the eMarketer survey’s findings generalize to every industry: the sample skews toward marketers already engaged enough with measurement to respond to an industry survey, which may understate confidence in last-click among less measurement-focused teams.

Frequently Asked Questions

Which attribution model is the most accurate?

None is universally most accurate — each answers a different question about the same customer journey. The practical fix is picking one as a consistent primary standard rather than searching for a single “correct” model.

Why do marketers keep using last-click if confidence in it is so low?

It’s the default view in many analytics tools, and switching models requires deliberate configuration. The eMarketer survey found 74.5% want to move away from it, which suggests inertia rather than continued trust explains its ongoing use.

Is multi-touch attribution always better than last-click?

Not automatically. Multi-touch models depend on accurate cross-device stitching and an appropriate lookback window — misconfigured, they can produce less reliable numbers than a well-understood last-click view.