1. Home
  2. Glossary
  3. Predictive analytics
AI glossary · Using AI at work

Predictive analytics

Predictive analytics: Predictive analytics uses historical data and machine learning models to estimate how likely a future outcome is: which customers will churn, which invoices will be paid late, what a store will sell next month. The output is a probability, not a fact.

Predictive analytics predates the current AI wave by decades. Credit scores, insurance pricing, and demand forecasts are all predictive models, and most use techniques such as regression and decision trees rather than anything resembling a chatbot. The recipe is consistent: assemble past records with a known outcome, find the patterns that preceded that outcome, and apply them to current records to score what is likely to happen next.

Today it lives inside tools you already use. CRM platforms score leads and flag at-risk accounts, ERP systems forecast demand, HR software estimates attrition risk, and maintenance systems predict which machine will fail next. Spreadsheet users can build a simple version with a forecast function. The value is in acting on the score: calling the at-risk customer, ordering the part before the failure.

The pitfalls are predictable too. Models trained on the past assume the future looks similar, so a pricing change, a pandemic, or a new competitor can quietly break them. Scores can become self-fulfilling once the business acts on them (the leads you ignore never convert). And in hiring, lending, and housing, predictive models can reproduce historical discrimination and collide with fair-lending and EEOC rules. Ask what the model was trained on, how well it did on data it had not seen, and how often it is retrained.

Example at work

A B2B software company builds a churn model from three years of usage and support data. It flags accounts whose logins dropped and who opened two or more billing tickets. Customer success calls the top fifty flagged accounts each month with a specific offer. The team tracks whether flagged accounts that were called renew more often than flagged accounts that were not, which is the only way to know the model is earning its keep.

Why it matters

Predictive scores show up in more decisions every year, often without the label "AI". Understanding that they are educated guesses from past data, with known failure modes, lets you use them confidently and challenge them when they are wrong.

Related terms