Two reports land in the same meeting. One says paid search drove 60% of last quarter’s revenue. The other says paid search drove 25%, and content and email did most of the work. Same data, same date range, same conversions. The instinct is to figure out which one is right. The better question is which one was built to answer the thing you’re deciding today, because they weren’t built to answer the same thing.
What a Model Actually Is
An attribution model is a rule for splitting credit for one conversion across the touchpoints that came before it. That’s all. Last click gives everything to the final touch. First click gives everything to the first. Linear splits it evenly. Time decay weights recent touches more. Data-driven uses the account’s own conversion patterns to work out which touches tended to matter, then weights accordingly.
None of these is a measurement of what caused the sale. They’re conventions for allocating credit, and each convention is more useful for some decisions than others. Treating any of them as the truth is where the arguments start.
Last Click Answers “What Closed It?”
Last click is easy to dismiss now, and it’s still the right model for one specific question: what was the touch that finally converted a person who was already warm? That’s a useful thing to know when you’re deciding what your bottom-of-funnel should look like, which landing page to invest in, or whether a retargeting campaign is actually catching people at the right moment.
What it can’t tell you is what made them warm in the first place. A branded search that closes a deal gets full credit under last click, and the six months of content that made someone search your brand name get nothing. If you set budgets on last click, you’ll keep cutting the things that create demand and keep funding the things that harvest it, until there’s nothing left to harvest.
First Click Answers “What Started It?”
First click is the mirror image. It’s the model to look at when the question is where new people are coming from, which channels are opening relationships rather than closing them, and whether top-of-funnel spend is finding anyone at all. It’s also the model that makes content and organic search look best, which is why it’s rarely the one a paid media team volunteers.
Data-Driven Answers “What Tended to Matter?”
Data-driven attribution is the most defensible model for budget allocation across channels, because it’s derived from your own conversion paths rather than a rule someone picked. It’s also the least transparent. You can’t easily explain why it gave 30% to email this month and 18% last month, and when the pattern shifts, you’re trusting a model you can’t see inside.
It needs volume, too. Below a few hundred conversions a month, the weights get noisy, and the model quietly falls back to something simpler without making a fuss about it.
Match the Model to the Decision
The practical fix is to stop asking “which model do we use” and start asking “what are we deciding.” A short mapping covers most cases:
- Setting channel budgets for next quarter: data-driven if you have the volume, position-based (40/20/40) if you don’t.
- Deciding which landing page or offer to improve: last click.
- Judging whether a new top-of-funnel channel is worth continuing: first click, alongside a plain count of new-to-file contacts it produced.
- Explaining results to someone outside marketing: whichever model you used last time. Consistency matters more than sophistication when the audience is comparing quarter to quarter.
Whatever you pick, write it down next to the number. A revenue figure without its attribution model attached is a figure nobody can check.
The Thing Attribution Can’t Fix
Every model divides up conversions that were tracked. If the tracking chain breaks before the CRM, if UTMs are inconsistent, if half your leads come in through a route that never got tagged, the model is splitting credit for a fraction of reality, and it doesn’t know that. That’s why the UTM discipline and the traffic-to-pipeline chain come first. Attribution is the last step in measurement, not the first, and it’s only as good as the data walking into it.
If two of your reports are arguing with each other, the fix is usually to decide what question you’re asking. We can help with that.