← All projects

Explainable Customer Retention Pipeline

Churn-risk scoring with built-in reasoning traces - symbolic + Bayesian + LLM layers.

Why it mattersGives customer-success teams a why alongside the score, not just a number. Critical when acting on a prediction requires sign-off from a human who needs to justify the call.

PythonBayesian NetworksLLMs

What it does

A churn-risk pipeline that doesn't just score customers - it explains why. Every recommendation can be traced from the symbolic features through a Bayesian causal layer up to the language framing on top.

Where it applies

  • Customer-success and retention teams that need auditable reasoning before acting on a risk score.
  • Finance and telecom - industries where models that affect customer treatment are increasingly held to a "show your work" standard.
  • Any business context where a confident black-box prediction isn't good enough to act on.

How it works (high level)

Symbolic feature extraction turns raw signals into clearly-named inputs. A Bayesian network exposes the causal structure between those features and churn outcomes. An LLM layer handles narrative framing and personalisation - but it never drives the underlying score. That separation is what makes the trace readable.

Stack

Python · Bayesian networks · LLMs · symbolic feature extraction.