What is a causality engine?
A causality engine is a system that infers cause-and-effect relationships from data, rather than describing the correlations inside it. Conventional analytics reports what happened and what moved together; a causality engine estimates what would have happened otherwise, and attributes the difference to a specific cause. The term is used in economics, epidemiology, large-scale experimentation, and marketing measurement.
Correlation, prediction, causation
Three questions are easy to confuse, and the difference between them is the whole subject:
| Question | Method | Answers |
|---|---|---|
| What happened? | Analytics, BI, dashboards | Description of the past |
| What will happen? | Machine learning, forecasting | Prediction, given no change |
| What happens if we change it? | Causal inference | The effect of an intervention |
Only the third question supports a decision. A model can predict revenue accurately and still be useless for deciding where to spend, because prediction does not separate a cause from a coincidence that reliably accompanies it.
How a causality engine works
Every causal method is an attempt to observe something that cannot be observed directly: the same subject, in the same period, without the intervention. That missing half is called the counterfactual, and the methods differ mainly in how they construct it.
- Randomised experiments build the counterfactual by assigning subjects to treatment or control at random, so the groups differ only in the treatment. This is the strongest method and is not always practical.
- Quasi-experimental methods — difference-in-differences, synthetic control, regression discontinuity, geo tests — construct a comparison group from data when randomisation is impossible.
- Causal graphs (directed acyclic graphs, do-calculus) encode assumptions about what causes what, then derive which effects can be identified from observational data and which cannot.
Causality Engine, the company
Causality Engine is a Dutch software company that applies causal inference to one specific problem: which marketing channels actually caused revenue for an ecommerce brand. It reads a Google Analytics 4 export and returns a per-channel causal view with confidence intervals in 5–10 minutes, using a proprietary causal-inference model rather than machine learning. It costs €99 per read, requires no pixel or SDK, and was founded in December 2023 in Utrecht.
The gap it measures — between what ad platforms claim drove revenue and what actually caused it — is what we call attribution debt.
Frequently asked
What is a causality engine?
A causality engine is a system that estimates cause-and-effect relationships from data rather than describing correlations in it. Where a conventional analytics tool reports what happened and what moved together, a causality engine estimates what would have happened otherwise — the counterfactual — and attributes the difference to a specific cause.
How is a causality engine different from analytics or BI?
Analytics and BI tools are descriptive: they aggregate, segment, and visualise what already occurred. A causality engine is inferential: it uses methods such as randomised experiments, quasi-experiments, or causal graphs to estimate the effect of an intervention. The practical difference is that analytics can tell you two things moved together, while a causal method can tell you whether one caused the other.
Is a causality engine the same as machine learning or AI?
No, although they overlap. Most machine learning optimises prediction: given these inputs, what is the likely output. Causal inference asks a different question: if we changed this input, how would the output change. A model can predict extremely well and still be useless for deciding what to do, because prediction does not distinguish cause from coincidence.
Where are causality engines used?
In economics and epidemiology, where causal inference originated; in technology platforms running large-scale experimentation; and in marketing measurement, where the question is which spend actually generated revenue rather than which channel was present at the sale. Causality Engine applies it to the last of these.