Skip to content
DT Nova

Types of analytics: descriptive, diagnostic, predictive and prescriptive

There are four main types of analytics, and each answers a different question. Descriptive analytics tells you what happened. Diagnostic analytics explains why. Predictive analytics estimates what is likely to happen next. Prescriptive analytics recommends what to do. This guide explains each of the types of analytics with plain examples, the data each one needs, where each goes wrong, and how to decide which one your company actually needs now.

The four-part model is widely used in business teaching; Harvard Business School Online, for example, describes the same four types with the same questions. It is a useful map, not a ladder you must climb to the top. Most of the value for most companies sits in the first two.

The short version

Type and questionTypical outputExample
Descriptive: what happened?Reports, dashboards, KPIsSales by product and month, compared with last year
Diagnostic: why did it happen?Breakdowns, comparisons, drill-downsMargin fell because one supplier raised prices and discounts grew in one region
Predictive: what is likely to happen?Forecasts, risk scoresExpected orders per week for the next quarter, with a range
Prescriptive: what should we do?Recommendations, optimised plansHow much stock to order per warehouse to meet the forecast at the lowest cost

1. Descriptive analytics: what happened

Descriptive analytics summarises past and current data into numbers people can act on: revenue, margin, orders, stock levels, hours worked, tickets closed. It is the simplest type and the one the others depend on. Most dashboards are descriptive.

It sounds easy, and the hard part is not the chart. It is agreeing what each number means. Is revenue booked when the order is placed, shipped or paid? Does a returned item reduce last month's sales or this month's? Two reports that disagree destroy trust faster than a missing report. Good descriptive analytics starts with a written definition for each key number, one source for each definition, and the same number shown the same way everywhere.

Where it goes wrong: too many numbers, none of them owned; definitions that differ between finance and sales; dashboards nobody opens after the first month.

2. Diagnostic analytics: why it happened

Diagnostic analytics looks behind a change. Sales dropped: in which region, which product, which customer group, which channel? Did something else change at the same time, such as a price, a supplier, a campaign or a delivery problem? It compares, breaks down and correlates, and where the data allows, it tests a cause rather than assuming one.

This is where most management questions actually live. It needs data at a finer level than the summary: order lines rather than monthly totals, timestamps rather than weeks, and the attributes that let you slice, such as customer segment, region or product group. It also needs care: two things moving together do not prove that one caused the other, and the most useful diagnostic work often ends with a small test, such as changing one price in one region, rather than a chart.

Where it goes wrong: reading causes into coincidences; slicing until something looks significant; data that was summarised too early to break down.

3. Predictive analytics: what is likely to happen

Predictive analytics uses historical data to estimate future outcomes: next quarter's demand, which customers are likely to leave, which invoices will be paid late, which machine is likely to fail. The methods range from seasonal averages and regression to machine learning models. The output should always be a range or a probability, not a single certain number.

A forecast is only as good as the history behind it and the stability of the world it describes. Seasonal patterns need several seasons of clean, consistent history: with one year of data you cannot tell a season from a one-off. A change in pricing, product range or market can make older history misleading. Before building a model, measure the simple baseline, such as the same week last year adjusted for trend, and only keep a complex model if it beats that baseline on data it has not seen.

Where it goes wrong: models that look accurate on the data they were trained on and fail on new data; forecasts presented without uncertainty; nobody responsible for acting on the prediction. For prediction on the factory floor, see our guide to manufacturing predictive analytics.

4. Prescriptive analytics: what to do about it

Prescriptive analytics recommends an action, often by combining a prediction with the rules and costs of the business. Given the demand forecast, how much should each warehouse order? Given the expected workload, how many people should be on each shift? Given capacity and due dates, in which order should jobs run? The methods include optimisation, simulation and decision rules.

It is the most powerful type and the one that touches decisions most directly, so two rules apply. First, the recommendation must show its reasons and constraints, so a person can check it. Second, decisions about people need particular care. Under Article 22 of the GDPR, a person has the right not to be subject to a decision based solely on automated processing, including profiling, which produces legal effects concerning them or similarly significantly affects them. Prescriptive systems that touch hiring, credit, pay or access to services should keep a person in the decision.

Where it goes wrong: recommendations nobody understands, so nobody follows them; optimising one number while breaking others; automated decisions about people without review.

Which type of analytics do you need?

Work backwards from the decision, not forwards from the technique. Write down the three decisions that matter most in your business this year, who makes them and how often. Then ask what each decision needs:

  • If people argue about what the numbers are, you need descriptive analytics with agreed definitions first.
  • If the numbers are trusted but nobody knows why they move, you need diagnostic analytics with data at a finer level.
  • If the decision depends on what will happen next, such as stock, staffing or cash, you need predictive analytics with an honest baseline.
  • If the decision is complex and repeated, with many constraints, such as routes, rosters or replenishment, you may need prescriptive analytics.

For context on adoption: according to Eurostat, 19.95% of EU enterprises used AI technologies in 2025, and 55.03% of large enterprises did (Eurostat, Use of artificial intelligence in enterprises, data extracted December 2025, read 11 November 2026). Predictive and prescriptive tools are spreading, but for most companies the bottleneck is still trusted descriptive data underneath them.

The data each type needs

TypeData and historyPeople
DescriptiveSummaries from source systems, with definitions; months of historyAn owner per number
DiagnosticTransactions with attributes to slice by; a year or more for comparisonsSomeone who knows the business context
PredictiveClean, consistent history of the outcome and its drivers; several seasons if the outcome is seasonalSomeone who tests models against a baseline
PrescriptivePredictions plus rules, costs and constraintsA decision owner who reviews recommendations

How we help

We help companies choose the type of analytics their decisions need, then build it: the data model, the pipelines from your systems, the dashboards and, where it pays, the forecasts and planning tools. See data analytics consulting for the questions-first approach, dashboard development and Power BI consulting for descriptive and diagnostic work, and predictive real estate analytics or production scheduling software for examples of predictive and prescriptive systems. For AI more widely, see AI consulting. For one industry worked through use case by use case, read predictive analytics for retail; for sport, read artificial intelligence in sports. Sector and team examples: healthcare analytics solutions and predictive analytics for marketers.

Questions people ask

What are the four types of analytics?

Descriptive analytics answers what happened. Diagnostic analytics answers why it happened. Predictive analytics answers what is likely to happen next. Prescriptive analytics answers what we should do about it. Each builds on the one before, so the data and trust built for descriptive reporting are the foundation for the rest.

Which type of analytics should a small company start with?

Descriptive. Reliable numbers for sales, margin, cash, stock and customers, defined once and shown the same way every week, deliver most of the value for most small companies. Predictive models only help once those numbers are trusted.

Is predictive analytics the same as AI?

Not exactly. Predictive analytics uses historical data to estimate future outcomes, and it can use simple statistics as well as machine learning. Machine learning is one set of methods for it, and many useful forecasts need nothing more than a seasonal trend.

What is the difference between business intelligence and analytics?

The terms overlap. Business intelligence usually means the reports and dashboards that show what happened, which is descriptive analytics. Analytics is the wider practice, including diagnosing causes, forecasting and recommending actions.

Do we need a data warehouse for analytics?

Not at the start. A few sources can be reported on directly. A warehouse becomes worth it when several systems need to be combined, history has to be kept when source systems change, or reports slow down the systems they read from.

Sources