Correlation: A Handy Tool in the Performance Marketer's Toolkit

A correlation is a handy tool in my performance marketing kit. Just the relationship between two sets of data. It's another unsung hero.

It comes up constantly in performance marketing, because you're almost always holding two numbers from two different sources that were never going to match. Ad platforms with their own self-reported attribution versus your source of truth in the CRM. Different attribution models, different tracking, cookie issues. The B2B buyer journey is complicated to begin with, and it gets worse with channels that aren't direct response by default: zero click, LLMs, push, awareness.

You're not going to tie those together perfectly. A correlation tells you whether they at least move together, which is often the thing you actually need to know.

The 30-second version

Take two arrays of numbers over the same date ranges. You get back a value between 1 and -1.

• 1 is a perfect positive relationship. They move together.
• -1 is a perfect negative relationship. One goes up, the other goes down.
• 0 means no relationship.

It does not imply causality. This isn't meant to be sophisticated data science. It's a quick read on whether two things are connected.

Pearson or Spearman

Two flavors, and it's worth running both.

If you've ever used CORREL in a spreadsheet, you've run a Pearson. It uses the actual values and asks how close the points sit to a straight line.

If you've ever ranked both columns first and then run CORREL on the ranks, that's a Spearman. Same formula, different input. It asks whether the two columns put the weeks in the same order, not whether the relationship is a straight line.

I did both for years without knowing the second one had a name.

When the two disagree, that's information. A big outlier week or a curved relationship (think diminishing returns on spend) will drag Pearson down while Spearman holds up.

Use case 1: in-platform conversions versus your source of truth

This is the big one for me. You optimize in the platform, but leadership is looking at the CRM. Your source of truth data is rarely plumbed perfectly back into the ad platforms, so you need to feel confident that what you're acting on relates to what the business is counting.

The counts don't match, and they never will. But they move together, so the campaign that looks good in platform is the campaign producing MQLs. For me, anything consistently around 0.6 and up and I feel pretty good.

It matters even more when you hand accounts to someone else to manage. You want to know the in-platform numbers they're steering by actually track the business.

Here's where it gets practical. Say your source of truth cost per MQL target is $1,000, but in platform it reads closer to $500 because of the attribution delta. If the two series track each other tightly and the ratio is consistent, you can do the math and set a platform target that maps back to the number your company actually cares about.

Use case 2: LLM mentions versus direct traffic

As more of the journey goes zero click, you don't get a click to attribute. What you can do is line up LLM mentions against direct and branded traffic by week and see whether they move together.

Week 8 is a spike, probably one mention that traveled. Pearson gets dragged down by it. Spearman says every single week still ranks in the same order on both sides. Drop week 8 and Pearson goes to 1.00. That gap between the two numbers is the tell that one week is doing the damage, which is exactly why I run both.

Other examples: email sends against direct traffic, events against direct traffic and branded search. Use your imagination.

How to actually run it

The doing part is easy. Paste two columns into an LLM, ask for a Pearson and a Spearman correlation, and you'll have it in about three seconds. Back in the day it was a spreadsheet CORREL function, which is kind of an archaic way to do it in 2026.

What matters is understanding the concept. What 1 and -1 mean, and when you can use it. It's very simple. It's more about being aware it's a piece of armor in your toolkit.

If you want something repeatable, a short Python script does it. It's in the repo, free, with both example datasets from this post. There's also a free web version if you'd rather paste and go. No signup.

The honest caveats

It doesn't imply causality. Pearson won't catch a relationship that isn't a straight line. Spearman will, as long as it keeps going one direction. A handful of data points will tell you nothing. And if an email drops Tuesday and conversions land Friday, you may need to shift one series before the two line up at all.

If you've got a data scientist, use them, and get as robust as you can. I'm not offering this as a replacement for that. It's a quick tool that helps you get your head around what relates to what, and it's saved me a lot of time.

If you're not using it, give it a shot.

← All posts