The 4 Types of Data Analytics: Descriptive, Diagnostic, Predictive and Prescriptive

The four types of data analytics are four questions you can ask of data. Descriptive: what happened? Diagnostic: why did it happen? Predictive: what is likely to happen next? Prescriptive: what should we do about it? Each one builds on the one before it and needs more data, more skill and more judgment.

Most explanations of this framework list four definitions and move on, so it reads like theory. It makes more sense when you follow one business problem through all four questions, so this article uses a single example the whole way through.

The Scenario

An online store's revenue fell from $200,000 in February to $170,000 in March. Management wants answers by Friday.

1. Descriptive Analytics: What Happened?

Descriptive analytics summarizes past data: totals, averages, trends and breakdowns. It's the most common work in any analytics team and the bulk of what junior analysts do.

In our scenario: revenue dropped 15% month over month. Order count fell by about the same amount, and average order value barely changed. A breakdown by channel shows desktop revenue held steady, while mobile app revenue fell from about $104K to about $74K.

Typical tools: SQL, Excel, and BI dashboards in Power BI, Tableau or Looker. Typical outputs: KPI dashboards, monthly reports, variance tables.

Descriptive work doesn't answer "why", but it tells you where to look. The drop is concentrated in the mobile app. A team that jumps straight to theories about pricing or competitors without this step often ends up investigating the wrong thing.

2. Diagnostic Analytics: Why Did It Happen?

Diagnostic analytics drills into the "what" to find causes. It relies on segmentation, funnel analysis, comparing before and after, and checking which explanations the data rules out.

In our scenario: the analyst breaks the mobile funnel into steps. App visits are flat and add-to-cart rates are flat, but checkout conversion fell from 3.1% to 2.2%, starting on March 4, the day a new app version shipped. Desktop checkout conversion didn't change.

A quick check: mobile made up about 52% of February revenue. A drop in conversion from 3.1% to 2.2% is a 29% fall, and 29% of 52% is about 15% of total revenue. That matches the decline, so this one cause accounts for almost the whole drop.

Typical tools: SQL (joins, window functions), Python or Excel for segmentation, and drill-downs in BI tools. The skill that matters most: telling correlation from causation. The release date lining up with the drop is strong evidence, and it becomes convincing because other explanations (traffic, cart behavior, desktop) were checked and ruled out.

3. Predictive Analytics: What Is Likely to Happen?

Predictive analytics uses historical patterns to estimate future outcomes. That covers forecasts, churn probabilities, demand estimates and risk scores. This is where statistics and machine learning come in.

In our scenario: a simple forecast that combines the new conversion rate with April's usual seasonal dip projects around $165K in April if nothing is fixed. A churn model also flags that customers who hit a failed checkout are much more likely to stop buying, so part of the damage could outlast the bug.

Typical tools: Python (pandas, scikit-learn), R, time-series forecasting methods, and AutoML features in cloud platforms. Important caveat: a prediction is a probability, not a certainty. Every forecast should come with a range and a list of its assumptions. "$165K, assuming traffic stays flat" is more honest and more useful than "$165K".

4. Prescriptive Analytics: What Should We Do?

Prescriptive analytics recommends an action. It compares options by their expected results and costs, using simulation, optimization or experiments.

In our scenario: there are three options.

The recommendation: roll back now, ship the hotfix behind a feature flag, and A/B test it before rolling it out to everyone. Each day of delay costs roughly $1,000 in revenue, so speed matters more than keeping the new features.

Typical tools: optimization libraries, simulation, A/B testing platforms, and often just a well-built decision table like the one above. What it needs most: business context. The model can put numbers on the options, but deciding whether a lost feature matters more than $1,000 a day is a business decision.

The Four Types at a Glance

Common Misconceptions

"Descriptive analytics is the low-value kind." It's the foundation. A forecast built on an inaccurate revenue number is confidently wrong. Many well-paid analysts spend most of their time on descriptive and diagnostic work, because that's where most business questions sit.

"You must complete each stage before the next." It's a progression of questions, not a strict sequence. Mature teams run all four at once on different problems. Still, a prediction without a sound descriptive baseline is guesswork.

"Predictive means machine learning." Often it's a well-reasoned trend line or a seasonal adjustment in a spreadsheet. Use the simplest model that answers the question well enough.

"Prescriptive analytics makes the decision." It informs the decision. The trade-offs in our example (features vs revenue, speed vs certainty) still need a person to weigh them.

Where AI Fits In

AI tools are quickly getting faster at descriptive work: generating SQL, summarizing dashboards and drafting reports. They're less reliable at diagnostic reasoning, where you have to know which alternative explanations to rule out, and at prescriptive judgment, where business context matters. We look at what that means for jobs in Will AI Replace Data Analysts?. The short version is that the further right you can work in the table above, the harder you are to automate.

How This Maps to a Learning Path

• Descriptive and diagnostic make up the core of an entry-level data analyst skill set: SQL, spreadsheets, a BI tool and structured thinking. Our data analyst roadmap and SQL roadmap cover this stage. The Google Data Analytics Professional Certificate is built around it.
• Predictive adds statistics, Python and machine learning. The Google Advanced Data Analytics Certificate is a natural next step, and our machine learning engineer roadmap goes further.
• Prescriptive skills (experimentation, optimization) usually build up through experience on top of the other three.

Browse all options in our Data Analysis courses.

Frequently Asked Questions

Which type of data analytics is most common? Descriptive, by a wide margin. Most dashboards, reports and ad-hoc requests in a typical company are descriptive, with diagnostic follow-ups.

Which type pays the most? Roles that focus on predictive and prescriptive work (data scientists, analytics leads) usually pay more, but the pay reflects the broader skill set those roles need, not the label. Strong diagnostic analysts are well paid too.

Where does "cognitive" or "AI-driven" analytics fit? Some vendors add a fifth type for AI-driven or automated analytics. In practice, that's AI tooling applied across the four existing types rather than a separate type.

Can one project use all four types? Yes. The worked example in this article did. Most substantial business problems start descriptive and end with a prescriptive recommendation, even if nobody calls it that.

Bottom Line

The four types of analytics are four questions: what happened, why, what's next and what to do. Each needs more skill and judgment than the one before. Beginners should get very good at descriptive and diagnostic work first, because that's where most jobs and most business value sit. Predictive and prescriptive skills build on that foundation.