Core technology
What is Causal AI?
Predictive AI finds what correlates. Causal AI estimates what happens when you act — the difference between a forecast and a decision. Here is how it works, and why your experts stay at the centre of it.
The problem
Correlation can’t tell you what to do
Predictive machine learning is built to find patterns: which variables move together. That is genuinely useful for forecasting — but it is silent on the one question institutions actually face when spending money or changing a policy: what happens if we intervene?
The reason is structural. Three completely different realities can produce the same correlation in your data, and a predictive model cannot tell them apart.
Standard ML cannot distinguish these
The Ladder of Causation
Three rungs, not one
Judea Pearl’s framework describes three levels of causal reasoning. Standard ML sits on the bottom rung; Causal AI is built to climb the upper two.
Pearl’s Ladder of Causation has three rungs. Rung one, Association: what is correlated — the level of standard machine learning. Rung two, Intervention: what happens if we act. Rung three, Counterfactual: what would have happened under a scenario that did not occur. Causal AI operates on the upper two rungs.
Methodology
Human-in-the-loop, end to end
A four-step process where your experts own the assumptions and our platform owns the rigour.
- 01
Define the structure
Your domain experts encode the causal relationships they already understand.
- 02
Identify what’s estimable
We determine which effects can be identified from your data — and which cannot.
- 03
Estimate the effects
The platform estimates causal effects and runs robustness and refutation checks.
- 04
Visualise & report
Results are turned into auditable, explainable reports for decision-makers.
Causal AI is not fully automated. The domain expertise of your team is not replaced — it is the foundation. Our platform makes that expertise computationally rigorous.
Foundation
Built on the scientific tradition of causal inference
We do not invent the science — we make it usable. 24may builds on decades of peer-reviewed causal inference and the open-source tools that implement it.
Judea Pearl
Turing Award 2011 — mathematical framework for causality
2021 Nobel Economics
Angrist, Imbens, Card — causal inference methods
Open-source foundation
DoWhy (Microsoft), CausalML (Uber), EconML (Microsoft)
24may is independent and not affiliated with these researchers or organisations.
Applications
Where causal questions live
Identify the causal drivers of student attrition and attainment, so interventions target cause — not correlation.
Estimate how design and process changes cause shifts in performance, reliability, and cost.
Model how district-level interventions cause changes in traffic, air quality, and demand.
Trace the causal pathways from site conditions to safety incidents and schedule delays.
Measure the causal effect of L&D programmes on capability, performance, and retention.
Quantify how policies and interventions cause environmental outcomes across space and time.
Bring us a real decision
Book a demo and we will walk through how a causal model would fit one of your institution’s actual questions — with your experts in the loop.