GLOSSARY

Predictive Analytics

Predictive analytics uses statistical and ML models on historical data to forecast outcomes — churn, conversion, demand, risk — governed via MLOps and NIST AI RMF.

Last updated:

Quick answer
Predictive analytics uses historical data and machine-learning or statistical models to forecast future events or outcomes — customer churn, purchase propensity, demand volume, credit risk, fraud, equipment failure. The output is a probability or expected value for each record the business cares about, and the practical value is realized only when the prediction is wired into a business decision or workflow.

WHAT IT IS

Techniques include regression, decision trees, random forests, gradient boosting (XGBoost, LightGBM, CatBoost), neural networks, and survival analysis. Model choice depends on data shape, explainability requirements, and latency constraints; in regulated contexts (credit, insurance, healthcare), model governance often forces explainability-first methods and full documentation per ISO/IEC 42001 or NIST AI RMF.

HOW IT WORKS

A production predictive-analytics program requires feature engineering infrastructure, training/serving pipelines (MLOps), monitoring for data and concept drift, fairness and bias checks, and a deployment path that turns predictions into decisions or experiences automatically.

WHEN TO USE

Deploy predictive analytics when decisions recur at volume, when the cost of mis-prediction is measurable, and when labeled history or reliable feedback loops exist to train and retrain.

RELATED

SOURCES

Related questions.

What is predictive analytics?
Predictive analytics uses historical data and machine-learning or statistical models to forecast future events or outcomes — customer churn, purchase propensity, demand volume, credit risk, fraud, equipment failure. The output is a probability or expected value for each record the business cares about.
How does predictive analytics differ from BI?
BI reports what happened and what is happening. Predictive analytics forecasts what is likely to happen next and, when combined with decision logic, recommends what to do about it. The two are complementary layers of the data stack, not substitutes.
What use cases return the most value?
Customer churn prediction, lifetime-value estimation, lead scoring, recommendation and personalization, demand forecasting, fraud detection, credit and risk scoring, and predictive maintenance. Each ties directly to a revenue or cost line, which is why they tend to survive procurement review.
What does a predictive deployment require?
A clean feature store, a labeled outcome variable, train/validate/test splits, offline evaluation against business metrics, an online deployment with monitoring, and a retraining cadence. Most failures happen at deployment and drift, not at modeling.
How does NUUN Digital deliver predictive analytics?
We tie every model to a named business outcome, build vendor-neutral using cloud-native tooling, instrument monitoring for data and concept drift before release, and define a retraining cadence up front. Predictive programs without retraining cadence decay fast.

Need this term in action?

Book a call