What is likely?
- Forecast an outcome
- Score risk or propensity
- Find predictive variables
Causal Machine Learning
Prediction tells you what may happen. Causal analysis asks what will change it.
Graphite Note uses causal AI and causal machine learning methods when the business question is about an intervention: which driver matters, which customer treatment works, or which operational change is expected to move the outcome.
Predictive vs causal
Two different questions need two different model types.
A predictive model can be excellent at forecasting scrap, churn or demand without telling you which intervention will improve it. Causal work starts with a counterfactual question: what would likely change if we changed X while other relevant factors were accounted for?
What changes the outcome?
- Define a treatment or intervention
- Control for confounding assumptions
- Estimate treatment effects
A governed workflow
Causal answers need diagnostics, not just a coefficient.
Graphite Note treats causal work as an evidence pipeline rather than a single model call.
Frame the causal question
Define the outcome, candidate intervention, timing and decision the analysis is meant to support.
Check identification and overlap
Review confounders, treatment support and whether the available data can credibly answer the question.
Estimate and stress-test effects
Use appropriate estimators, diagnostics and robustness checks before turning an effect into a recommendation.
Where it fits
Use causal ML when the decision is an intervention.
Promotions and targeting
Estimate whether a campaign, offer or treatment changes behaviour rather than merely correlating with it.
Process changes
Identify which controllable variables are likely to change quality, waste, throughput or another KPI.
Incremental impact
Separate customers who would act anyway from those whose behaviour is changed by an intervention.
Direct answers
Causal ML in plain English.
Can causal machine learning prove causality from any dataset?
No. Causal conclusions depend on the design, available variables, assumptions and data support. A responsible workflow should make those limitations explicit rather than presenting every association as causal.
Is A/B testing still useful?
Yes. Randomised experiments are often the strongest way to identify causal effects. Causal ML is useful when experiments are unavailable, incomplete, expensive, or when treatment effects vary across customers or operating conditions.
How does causal ML connect to Decision Intelligence?
Causal analysis helps determine which levers are plausible interventions. Decision Intelligence then combines that evidence with expected impact, constraints and execution.
Decision IntelligenceFrom signal and prediction to a governed business action.AutoMLAutomate repetitive model-building work without losing business context.Causal Machine LearningSeparate drivers from correlations and estimate intervention effects.Predictive AnalyticsForecast demand, score customers and anticipate business outcomes.Decision OptimisationCompare feasible actions and rank them by expected business impact.Decision APIDeliver recommendations, expected impact and evidence into workflows.Embedded Decision IntelligenceAdd a decision layer to ERP, CRM, BI, workflow and vertical software.Build vs BuyA practical framework for deciding whether to build, buy or embed.
Stop asking which variable moved. Ask what you can change.
Bring a business outcome, candidate levers and the data you already have.