The 4 Types of Data Analytics: Descriptive to Prescriptive

“Use data to make better decisions” is advice everyone agrees with and few people can act on. The problem is that “analytics” covers very different kinds of work. A sales report and a demand forecast are both analytics, but they answer different questions, need different data and cost very different amounts of effort.

The cleanest way to sort it out is the 4 types of data analytics: descriptive, diagnostic, predictive and prescriptive. Each answers one question. In this guide, we take a single business problem and answer it 4 ways, so you can see exactly what each type adds and when it’s worth the extra effort.

The 4 types at a glance

The types form a ladder. Diagnostic work depends on accurate descriptive numbers. Predictions are only as good as your understanding of what drives the outcome. Prescriptive recommendations need a prediction plus your business constraints. Skipping rungs usually means building a sophisticated model on numbers nobody trusts.

One question, 4 answers: the example

Picture an online home goods store with about 5,000 customers. The owner asks a simple question: “Repeat orders feel slow this quarter. What’s going on, and what should we do?”

Here’s how each type of analytics answers it. The numbers below are illustrative, not survey data.

Descriptive analytics: what happened?

Descriptive analytics summarizes historical data into totals, rates and trends. IBM describes it as summarizing and visualizing past data to reveal patterns. It’s the work behind every sales report, dashboard and monthly review.

For our store, descriptive analytics answers: The repeat purchase rate fell from 31% last quarter to 26% this quarter. First-time orders held steady. Average order value is flat.

That’s already useful. The owner now knows the problem is real, how big it is and that it’s specific to returning customers rather than overall traffic.

What you need:

• Clean transaction data with customer IDs and dates
• Consistent definitions (what counts as a “repeat” customer and over what window)
• A spreadsheet, the reports built into your ecommerce platform or a BI dashboard

Most of the value in analytics lives here, and most problems start here too. If 2 reports show 2 different repeat rates, fix the definition before moving on.

Diagnostic analytics: why did it happen?

Diagnostic analytics digs into the data to explain causes. IBM’s definition centers on uncovering root causes, patterns and relationships in historical data, using techniques such as drill-down analysis, correlation analysis and root cause analysis.

For our store, diagnostic analytics answers: Breaking the repeat rate down by segment shows the drop is concentrated among customers whose first order shipped late. Their repeat rate is roughly half that of customers who received orders on time. The late shipments trace back to a carrier switch 4 months ago.

Common diagnostic techniques:

• Drill-down: split a number by region, product, channel or customer type until the change concentrates somewhere
• Cohort comparison: compare groups who started at different times or had different experiences
• Correlation: check whether 2 measures move together, such as delivery time and repeat rate
• Root cause analysis: keep asking “why” until you reach something you can fix

Some BI tools automate part of this. Microsoft’s Power BI key influencers visual, for example, runs a regression behind the scenes to rank which factors are most associated with an outcome such as a low rating or churn. Microsoft recommends at least 100 observations for the state you’re analyzing.

One caution: correlation isn’t causation. Late delivery and low repeat rates moving together is a strong lead, not proof. A quick test, such as offering faster shipping to a sample of customers, confirms it.

Key takeaway: Descriptive and diagnostic analytics answer most everyday business questions. Get those right, with clean data and agreed definitions, before paying for predictive or prescriptive tools.

Predictive analytics: what is likely to happen?

Predictive analytics uses historical data with statistical and machine learning models to estimate future outcomes. IBM gives the example of a retailer forecasting demand for a new product from past customer behavior. Our full guide to predictive analytics covers churn, demand and risk models in more detail.

For our store, predictive analytics answers: A churn model scores every customer on their likelihood of not ordering again in the next 90 days, based on order history, delivery experience, support tickets and email engagement. About 600 customers are flagged as high risk. A sales forecast shows that if repeat rates stay at 26%, next quarter’s revenue will come in below plan.

What changes at this level:

• You need more history, usually at least a year or more for seasonal businesses
• You need to measure accuracy by testing the model on data it hasn’t seen
• Predictions come with uncertainty. Excel’s Forecast Sheet, for example, shows a confidence interval around each forecast value, set to 95% by default

Predictive analytics doesn’t require a data science team for every use. Excel’s forecasting uses exponential smoothing on time-based data, and many CRM, ecommerce and BI tools include built-in forecasting or churn scores. Custom models make sense when the decision is frequent and valuable.

Prescriptive analytics: what should we do?

Prescriptive analytics recommends actions. IBM calls it the most advanced type, combining predictions with optimization algorithms, business rules and constraints to recommend the best decision.

For our store, prescriptive analytics answers: Given a $3,000 retention budget, which customers should get which offer? An optimization model compares the expected value of keeping each high-risk customer against the cost of each action: a free-shipping code, a 15% discount or a personal email from the owner. It recommends free shipping for customers whose first order was late, a discount only for high-value customers at the highest risk, and no offer for customers likely to come back anyway. It also flags the carrier contract for review, because fixing delivery times is cheaper than paying for retention forever.

Prescriptive analytics ranges from simple to complex:

• Business rules: “If churn risk is above 70% and the first order was late, send free shipping.” Easy to build, easy to explain.
• Optimization: tools such as Excel’s Solver add-in find the best value for an objective, like maximum profit, while respecting constraints you set, like a budget cap.
• Automated decision engines: pricing, inventory replenishment and ad bidding systems that act continuously, with humans setting the rules and reviewing results.

The higher you go, the more important human judgment becomes. A model can optimize the number you give it, but it won’t know about a supplier relationship or brand risk unless you build that in.

How the 4 types compare in cost, data and value

Where to start

You don’t need to climb all 4 rungs at once. Here’s a practical path:

1. Pick 1 decision that matters. Staffing, reordering, pricing or retention. Analytics without a decision attached becomes a dashboard nobody opens.
2. Get the descriptive number right. Define the metric, confirm the source and check it against a second report.
3. Ask “why” before “what next.” Spend a week segmenting and drilling down. You’ll often find the fix here.
4. Add prediction where the decision repeats. Weekly inventory orders or monthly retention campaigns benefit from a forecast. One-off decisions usually don’t.
5. Automate prescriptive rules last. Start with simple, explainable rules and review results before scaling up.
6. Measure the outcome. Compare results with and without the analysis so you know what it’s worth.

If you’re building these skills yourself, our guide on how to become a data analyst covers the tools to learn, and data analytics for small business has 7 starter analyses using data you already have. AI features in spreadsheets can speed up the first 2 rungs too, as shown in our guide to using AI in Excel and Google Sheets.

Conclusion

The 4 types of data analytics are a sequence, not a menu. Descriptive tells you what happened, diagnostic tells you why, predictive tells you what’s likely next and prescriptive tells you what to do. In our example, each step added something real: the size of the problem, its cause, who was at risk and how to spend a limited budget. Start at the bottom, earn trust in the numbers and move up only where a repeated decision justifies the effort.

Sources

• IBM: What is prescriptive analytics? — definitions of the 4 types and techniques used in prescriptive analytics
• IBM: What is diagnostic analytics? — diagnostic techniques including drill-down, correlation and root cause analysis
• Microsoft Learn: Key influencers visualizations in Power BI — how the visual ranks drivers and its minimum observation guidance
• Microsoft Support: Create a forecast in Excel for Windows — exponential smoothing forecasts and default 95% confidence interval
• Microsoft Support: Define and solve a problem by using Solver — optimization with objectives, variables and constraints in Excel