Rung 1 — answerable with data

“Which customers tend to churn? Which risks score highest? Which claims are unusual?”

Rung 2 — requires a causal model

“If we change this pricing, what happens to retention? If we invest here rather than there, what changes?”

Rung 3 — requires an Structural Causal Model (SCM)

“Would this customer have churned anyway? Was this control the reason we had no breach? Did this claim result from our policy or would it have happened regardless?”

The gap between rungs is not a gap in sample size, model sophistication, or algorithmic power. It is a gap in the formal structure of the question. A Rung 3 question asks about a specific individual under a specific counterfactual condition — what would have happened to this customer, this asset, this claim, under different conditions. No dataset contains counterfactual observations. The counterfactual world did not happen. Data describes what did happen. Answering Rung 3 questions requires a model of the mechanism — how causes produce effects — not a larger sample of effects.

A Structural Causal Model. The graph encodes which variables cause which. The structural equations encode the mechanisms. The unobserved U variables encode the individual-specific factors that make each case unique. Together they make the counterfactual question computable. Without the U variables you can answer population-level questions. With them you can answer individual-level ones — the ones boards actually ask. See Structural Causal Models for the formal treatment.

The Engagement

Bring the decision your tools cannot currently support. Thirty minutes to determine whether the gap is bridgeable and what it would take.

info@rung3.ai