A simple idea transforming everything: Do Not Act on Correlations
SONOS’ service quality ratings was the most correlating factor towards likelihood to recommend. Should the brand invest in service to boost upselling and word of mouth?
Lottery providers’ survey proves that its customers are much older than the general population. Obviously this is their most receptive target group, isn’t it?
Telco customers who received an “Are you still satisfied” call, showed a higher churn rate than those who did not. Did the call awaken “sleeping” churners?
This kind of information comes to us every day and intuitively we think we “know” the answer. There must be a reason behind the correlation, right?
You may think: “Come on, let’s not split hairs. Facts are facts. And behind facts lies somehow the truth.”
You will be surprised.
SONOS’ service was primarily used by new customers in the setup process. The initial enthusiasm of new customers is expressed by a high NPS rating. Service and NPS rating are both results of being a new customer. They correlate but have just a minor causal relationship.
Lottery customers are more likely to buy a ticket if they are habituating playing and when they experience winning. Both grow over time, which is why age correlates with playing. But if everything else is the same, older people are LESS likely to start playing the lottery.
Telco customers who picked up the landline to take the call belonged to a social group with lower economic wealth who per se had a higher churn rate. Picking up the phone was an indicator of having a high likelihood of churning. The telephone conversation itself LOWERED churn.
Do not take your knowledge from correlations
Everybody has heard it “Correlation is not causation”. But it feels like a quote from Chinese Cookies “The journey is the reward”. Everybody quotes it, just a few act on it.
The simple reason why we ignore this is that we don’t know how to solve the dilemma.
“Do not take your knowledge from correlations”
But this is slowly changing. A new group of methods is becoming increasingly popular. The picture shows the frequency of two keywords on three internal company tech blogs.
It is “Causal AI” that the techies tinker with these days. Also, Gartner has picked up the trend. While GenAI is at its peak of hype, Causal AI is on the rise.
One of the scientific pioneers in this field is Judea Perl who has formalized causality in mathematical terms. He recently explained that Causal AI will grow first in two fields: Marketing and Medicine.
One German Causal AI software company is focusing and leading on MEDICAL application: XPLAIN. Another German Causal AI software is focusing and leading on MARKETING applications: It’s us – Success Drivers.
The big money instead flows into the general platforms. The company CausalLens raised in 2022 over 50m dollars and leads in building awareness for the Causal AI category.
Statistical Modeling: Like a Racing Car Going Offroad
“Ok, I get it, correlation is not causation. But we have statistical modeling since decades” you may argue. Good point!
Statistical and econometrical modeling have been around for a while and tens of thousands of dissertations have been made around it. In the 1970s even a statistical discipline of causal analysis evolved and led to Structural Equation Modeling and Partial Least Square modeling.
All this should now be worth nothing?
Of course not. Those methods are great. They are very accurate and precise. They are like Formula One Racing cars. Just amazing.
The only problem is when you try to drive an F1 car downtown. You don’t get far.
“Statistical Modeling: Like a Racing Car Going Offroad”
Like F1 cars, statistical models are tuned by assumptions nobody in real business life could ever justify. Even worse they are designed to validate hypotheses instead of learning new things.
The future of AI has one name: Causal AI.
Artificial Intelligence (AI) changes the paradigm from validation to exploration. It’s like a 4-weel Jeep, much more versatile in business landscape.
Still, AI suffers the same illness as most statistical modeling. No matter if we are talking about Model Drift, Discrimination by Machine Learning, or wrong explanations of AI models, it all has the same source: Lack of causality.
“The future of AI has one name: Causal AI.”
Causal AI as the methodology is solving the following issues of conventional statistical modeling and Artificial Intelligence:
Multicolinearities between drivers lead to biased causal attribution in statistical models and AI.
Confounders: Statistical models assume complete models. Any outside factor that influences dependent and independent variables at the same time can lead to directionally wrong findings. Causal AI methods enable to include all available variables or try solve the issue through other ways.
Causal Direction: Whether a variable is dependent or independent is not always clear. In Causal AI we have methods to test this. In any way, it forces you to make a conscious decision.
Indirect effects: Conventional driver analysis just has an input-output-view. A causal effect consists of those direct but also indirect effects. That is why Causal AI models just like conventional SEM are always modelling a causal network.
To make a long story short:
Causality is intuitive most leads you quickly to wrong conclusions.
Statistical modeling is limited and for most business applications cumbersome and not very practical.
Artificial Intelligence's weak spot and source for failure is the lack of Causality.
Causal AI is a new class of methods and brings better and richer findings and better predictions at lower costs in a shorter time.
Applications are anywhere in Marketing: Product optimization, brand positioning, attribution and mix modeling, ad effectiveness, or CX drivers. You may even better than ever understand why your category buyers truly buy.
You find more depth on Causal AI and its applications read my latest book “Think Causal, Not Casual”.
A simple idea can transform everything “Do Not Act on Correlations”.
If you are cultivating this idea in your organization. If you are offering solutions to this problem statement. If you are creating a causal mindset in your team …
Then you are 10x Insights at ease!
---Frank
p.s. Among those who comment on this article on LinkedIN or Substack, I will select three of you who I will be sending a hard copy to your postal address. You can be critical. I will select those who I consider to be the most constructive contribution.