Overview
Descriptive analytics tells you what happened. Predictive analytics tells you what will happen. Causal AI tells you why — and what you can do about it. Most organisations are stuck at the first level. We build predictive and causal AI systems that go deeper: identifying the factors that drive outcomes, quantifying their impact, and enabling decision-makers to understand the consequences of different actions before they take them. From credit risk models that explain why an applicant was declined, to clinical trial designs optimised by causal analysis of historical data, we build analytics that drive decisions — not just reports.
How It Works with a21
Problem Framing & Data Assessment
Define the prediction or causal question precisely. Assess available data — volume, quality, confounders, and selection bias. Design the modelling approach matched to the question and data.
Model Development & Validation
Develop predictive models and causal structures. Validate predictive accuracy and — critically for causal models — validate assumptions and test robustness to violations.
Deployment & Decision Integration
Deploy models into decision workflows — as real-time scoring APIs, batch processing pipelines, or interactive what-if tools — with monitoring for performance drift.
What We Offer
Predictive Modelling
Build and deploy machine learning models for classification, regression, survival analysis, and time-series forecasting — calibrated and validated for production use.
Causal Discovery
Apply causal discovery algorithms to learn the causal structure from observational data — identifying which variables drive outcomes versus those that merely correlate.
Treatment Effect Estimation
Estimate the causal effect of interventions — marketing campaigns, policy changes, clinical treatments — using econometric and causal ML methods.
Explainability & Decision Support
Provide feature importance, counterfactual explanations, and what-if analysis tools that make model outputs understandable and actionable for decision-makers.
Uplift Modelling
Identify which individuals are most likely to respond positively to an intervention — enabling targeted action that maximises return on investment.
Bias Detection & Fairness
Measure and address predictive disparities across demographic groups — ensuring models are fair and defensible in regulatory environments.
Why Choose a21
Causal, Not Just Predictive
Most firms stop at prediction. We go further — identifying the causal mechanisms that allow you to intervene, not just anticipate.
Regulated Industry Rigour
Our models are documented, validated, and built to satisfy model risk management requirements in financial services and pharma.
Decision-Integration Focus
We design models to be used, not admired. Every model is deployed into a decision workflow where it changes an actual business outcome.
Explainable by Design
We build explainability into models from the start — not as an afterthought. Every prediction comes with a clear, human-understandable explanation.
Success Stories
Problem
A consumer lender’s ML credit model was accurate but could not explain decisions — creating regulatory exposure and preventing actionable feedback to customers.
Solution
Rebuilt the credit model with causal variable selection, integrated SHAP-based counterfactual explanations, and designed an adverse action notice generator aligned with regulatory requirements.
Problem
A pharma company wanted to understand which patient characteristics were causally related to treatment response — to design more targeted future trials.
Solution
Applied causal discovery and treatment effect estimation to historical trial data across 8,000 patients, identifying three previously unknown causal moderators of treatment response.
Tech Stack & Tools
Scikit-learn / XGBoost / LightGBM
DoWhy / EconML
PyMC
SHAP / LIME
MLflow
FastAPI
Evidently AI
Get Started
Go beyond reporting. Talk to a21 about predictive and causal AI for your business decisions.
