Causal inference is the difference between:
“I ran a facebook ad campaign and leads went up." (That’s correlation), and
“The facebook ad campaign made leads go up." (That’s causation).
Causal inference helps us move from the first to the second.
Who cares?
It really important to know if the facebook ad campaign worked. If I ran the facebook ad as part of a wider strategy to test out a new go to market strategy or customer profile I want to know if it worked to know whether to continue investing in it, and what that tells us at our ICP. It's equally as useful to know if it hasn’t worked - as that enables to quickly move to invest elsewhere.
When trying to find what works, there is always a lot of noise in the data this is because there are a lot of factors at play that can influence your leads - user experience, weather, seasonality, referrals, traffic split, ads and so we will extrapolate which of these is the real cause.
Why It’s Hard
Just because two things happen together doesn’t mean one caused the other.
Everyone’s favourite example of this is the chart showing how the number of Nicolas Cage movies released each year correlates with the number of people who drowned in swimming pools. It’s completely meaningless, of course, but it shows how easily we can be fooled by patterns in data that have no actual connection. See the graph here.
But in all seriousness, if we use the example of the facebook ad campaign, there is likely a whole load of things you are doing in your business, and things happening outside of your business that drive leads at any one point. Leads increasing at the same time as the marketing campaign doesn’t prove that is the cause.
How We Try to Find Cause
One of the main concerns with assuming that correlation is causation is that of confounders.
Let’s explore how these 2 methods work
1. Experiments (Randomized Controlled Trials/ AB tests)
Effort: Full experimental setup
Results: Not guaranteed
The most well known method is experimentation
-
Split people into two groups randomly.
-
One group gets the treatment (my facebook ad).
-
The other group doesn’t.
By doing this we assume we have isolated the effect of the ad since we have equally split any other factors into seperate groups. So if weather has a large effect on my business, I assume I have randomized these perfectly into my 2 groups.
(In reality this randomization is extremely difficult and expensive and doesn’t always perfectly randomize those confounders)
2. Causal Inference
Effort: Collecting observational data with confounders
Results: Understand causal drivers
Sometimes, we don’t want to invest in the full experimental set-up, we just want to understand what works and what doesn’t- quicky.
-
Collect observational data
-
Mark confounders within the observational data
-
✨Causal Inference✨
We use novel generative causal AI to learn how each of the confounders affect your leads, perhaps the Facebook ad is the cause, or maybe it’s a sunny day, and that caused it instead.
If you’re tired of guessing what’s working in your funnel - get in touch to see how causal inference can accelerate your path to finding your ICP