What is the difference between correlation and attribution, and when do you use each?
Attribution assigns credit for a specific sale to a specific touch you observed. Correlation looks at whether two measures move together over time without claiming a path between them. Use attribution when an identifier links the lead to a click or call. Use correlation when no identifier exists, such as brand awareness, offline media or weather. Neither one establishes cause, and confusing them is the most common reporting error.
One works record by record, the other works period by period
Attribution is a joining operation. It needs a thread connecting a person to a touch: a click identifier, a tracking number, a form submission, a matched phone number. When the thread exists you can say this job came from that campaign, with a stated confidence.
Correlation is an aggregate operation. It compares two series across time and asks whether they rise and fall together. It needs no identifier at all, which is exactly why it is the tool for spend that produces no click, such as radio, direct mail or vehicle wraps.
What correlation is good for
Correlation earns its keep in three places. It measures channels with no click path, it detects lagged effects where spend today produces calls three weeks from now, and it surfaces relationships nobody thought to look for, such as call volume tracking a temperature threshold rather than a campaign.
That last case is where automated analysis helps most. Scanning many pairs of series is tedious for a person and trivial for a system, which is what AI business intelligence work is often actually doing under the hood.
The failure modes, and they are not subtle
- Seasonality drives everything. Spend, leads and revenue all rise in summer. Any two of them correlate beautifully and none of it means much.
- Too few periods. Twelve monthly data points cannot support a confident claim about a relationship, especially with seasonality in the mix.
- Multiple comparisons. Test a hundred pairs and some will look strong by chance alone. A correlation found by scanning needs a plausible mechanism before you act on it.
- Reverse direction. Rising leads often cause rising spend, because budgets get increased when things go well. The chart looks identical either way.
The honest hierarchy
Rank your evidence and say which rung you are standing on. Correlation is the weakest: two things moved together. Attribution is stronger: a specific identified path existed. A controlled test is strongest: you changed one thing and compared against a control that did not change.
Most reporting fails not because it uses correlation but because it presents correlation in attribution's voice. Writing the word next to the number costs nothing, and it is the difference between analysis people trust and analysis people quietly discount. Our intelligence briefs label the rung explicitly for that reason.
Topics: correlation · causation · analysis · measurement
Have a version of this question about your own business?
The useful answer usually depends on which systems you run and how they're connected. That's a conversation, not a blog post.