Quick answer: There are four main types of data analytics. Descriptive analytics tells you what happened. Diagnostic analytics explains why it happened. Predictive analytics estimates what is likely to happen next. Prescriptive analytics recommends what to do about it. They build on each other: each type needs the one before it to be reliable, and most businesses get the biggest return from doing the first two well before moving on to the last two.
Data analytics is the practice of turning raw business data, such as sales records, customer lists, website visits or production logs, into information that helps people decide what to do. The four types are not competing approaches; they are stages, each answering a harder question than the one before.
The Four Types at a Glance
| Type | Question it answers | Typical techniques | Example output |
|---|---|---|---|
| Descriptive | What happened? | Totals, averages, trends, dashboards, reports | "Sales fell 12% in October compared with September" |
| Diagnostic | Why did it happen? | Segmentation, drill-down, comparison, root cause analysis, correlation | "The fall came from two dealers in one district after a price change" |
| Predictive | What is likely to happen? | Forecasting, regression, classification, machine learning | "On current trends, November sales will be 8% below target" |
| Prescriptive | What should we do? | Optimisation, scenario modelling, simulation, decision rules | "Restore the old price for those dealers and add one sales visit a week" |
The value rises from top to bottom, and so do the data quality and skill needed. In practice, most of the questions a manager asks in a normal week are descriptive or diagnostic, which is why those two types deserve the most attention first. A forecast built on data nobody trusts is worse than no forecast, which is why the order matters.
1. Descriptive Analytics: What Happened?
Descriptive analytics summarises past data so people can see what happened. It is the foundation of every other type, and in most organisations it is still where the most value is left on the table.
Typical outputs: monthly MIS reports, sales by product and region, website traffic reports, inventory levels, attendance and productivity summaries, and the dashboards that bring them together.
Techniques: counts, sums, averages, percentages, growth rates, moving averages, comparisons with targets and with the same period last year, and charts that show trends and distributions.
Example. A distributor's weekly dashboard shows orders, invoiced sales, collections and outstanding payments by salesperson. Nothing in it is sophisticated, but it replaces three people rebuilding spreadsheets every Monday, and everyone sees the same numbers.
What makes it work: agreed definitions. "Sales" must mean the same thing in every report, whether orders, invoices or receipts, or descriptive analytics produces arguments instead of answers. See choosing KPIs for a business dashboard and business dashboard development.
Common pitfalls: too many metrics, reports that nobody acts on, and averages that hide the story. An average delivery time of three days can hide the fact that one route takes nine.
2. Diagnostic Analytics: Why Did It Happen?
Diagnostic analytics digs into descriptive results to find causes. When a number moves, it asks which segment, product, region, channel or customer group moved it, and why.
Techniques: breaking a metric down by dimensions (drill-down), comparing segments and periods, looking for the change point, correlation analysis, cohort analysis, and structured root cause methods such as hypothesis trees and the five whys.
Example. Website enquiries fall by a third in a month. Breaking them down by source shows organic search is stable, but paid search enquiries collapsed. Drilling further shows one campaign's landing page broke after a website update. The fix takes an hour; finding it without the breakdown could have taken weeks.
What makes it work: data at a fine enough level of detail (each order, each visit, each customer), and the discipline to test explanations against data instead of accepting the first plausible story. Our guides on business problem diagnosis, root cause analysis, diagnosing a revenue drop and why customers churn cover the methods in depth.
Common pitfalls: confusing correlation with cause, stopping at the first explanation, and looking only at the data that is easy to get.
3. Predictive Analytics: What Is Likely to Happen?
Predictive analytics uses historical data to estimate future outcomes or unknown values. It does not predict the future with certainty; it gives probabilities and ranges that are better than guessing.
Techniques: time-series forecasting (moving averages, exponential smoothing, ARIMA), regression models, classification models such as logistic regression, decision trees and random forests, and other machine learning methods.
Typical uses:
- Demand forecasting: how much of each product will sell next month, by location.
- Cash flow forecasting: when customers are likely to pay, based on their past behaviour.
- Churn prediction: which customers are likely to stop buying.
- Lead scoring: which enquiries are most likely to convert.
- Maintenance prediction: which machines are likely to fail, from sensor and maintenance data.
Example. A retailer with stores in Indore and nearby cities forecasts weekly demand for its top 300 products by store, using two years of sales, festival calendars and promotions. Stock-outs during Diwali fall, and so does excess stock in January.
What makes it work: enough clean history (often at least two years for seasonal businesses), a clear decision the forecast will drive, and honest measurement of forecast accuracy. See regression analysis explained.
Common pitfalls: building a model nobody uses, testing it only on the data it learned from, and ignoring events the history does not contain, such as a new competitor or a policy change.
4. Prescriptive Analytics: What Should We Do?
Prescriptive analytics recommends actions. It combines predictions with business rules, constraints and goals to suggest the best option, and sometimes acts automatically.
Techniques: optimisation (finding the best mix under constraints), simulation and scenario modelling ("what if we raise prices by 5%?"), decision rules, and recommendation systems.
Typical uses:
- Inventory and reorder decisions: how much to order, when, and from which supplier.
- Pricing and discounting: which discount maximises margin, not just volume.
- Route and delivery planning: the cheapest routes that still meet delivery windows.
- Workforce scheduling: staff rosters matched to forecast demand.
- Marketing budget allocation: where the next rupee of spend is likely to bring the most enquiries.
Example. A distributor combines its demand forecast with supplier lead times, minimum order quantities and warehouse capacity. An optimisation model recommends weekly purchase quantities that cut stock-holding by a fifth while keeping service levels.
What makes it work: reliable predictions, clear objectives (lowest cost, highest margin, best service) and people willing to follow or challenge the recommendation. Prescriptive analytics that managers do not trust is quietly ignored.
Common pitfalls: optimising one metric at the expense of others, hiding the logic so nobody trusts it, and automating decisions before the model has been checked against reality.
How the Four Types Work Together
The types are most useful as a loop:
- Descriptive shows that something changed.
- Diagnostic finds why.
- Predictive estimates what happens if nothing is done, and what happens under different actions.
- Prescriptive recommends the action.
- Descriptive then shows whether the action worked.
A business does not have to master each stage before touching the next, but each stage depends on the one before. Predictive models trained on data with inconsistent definitions produce confident wrong answers. Prescriptive recommendations built on poor forecasts optimise the wrong thing.
The Analytics Maturity Ladder
| Stage | What it looks like | Typical next step |
|---|---|---|
| 1. Spreadsheet reporting | Reports rebuilt by hand from exports; different numbers in different meetings | Agree metric definitions; automate the main report |
| 2. Automated descriptive | Dashboards refresh automatically; one version of the truth | Add drill-downs so managers can answer "why" themselves |
| 3. Routine diagnosis | Problems are broken down and explained quickly with data | Start forecasting the few numbers that drive decisions |
| 4. Predictive | Forecasts and scores used in planning and sales | Link forecasts to decisions with rules or optimisation |
| 5. Prescriptive | Recommendations or automated decisions, monitored and reviewed | Keep measuring, retraining and challenging the models |
Most small and mid-sized Indian businesses sit at stage 1 or 2. Moving from stage 1 to stage 2 is usually the single most valuable analytics investment they can make, because it saves time every week and makes every later stage possible.
Real-Time Analytics: Where Speed Matters
Real-time analytics processes data as it arrives, rather than overnight or weekly. It is a speed of delivery rather than a fifth type: real-time dashboards are descriptive, real-time alerts are often diagnostic, and real-time fraud scoring is predictive.
It is worth the cost where decisions are made within minutes: production line monitoring, payment fraud, logistics tracking, call centre queues, stock-outs on a busy e-commerce site and environmental monitoring alarms. For most management decisions, made daily or weekly, a reliable overnight refresh is cheaper and enough.
Big Data and Machine Learning: Do You Need Them?
"Big data" refers to data too large, fast or varied for ordinary tools, such as millions of transactions a day, sensor streams or large volumes of text and images. Most SMEs do not have big data; they have medium data in several systems that do not talk to each other. Their main problem is integration and quality, not volume.
Machine learning is a set of techniques used mainly in predictive and prescriptive analytics. It is powerful where there are many variables and plenty of history, such as demand across thousands of products. For a business with a few years of monthly data, simpler statistical methods are often as accurate and far easier to explain.
The right question is not "should we use AI?" but "which decision would improve if we could predict this number better, and do we have the data to do it?"
Data Analytics Examples by Industry
| Industry | Descriptive | Diagnostic | Predictive | Prescriptive |
|---|---|---|---|---|
| Retail | Sales by store and category | Why a store's margin fell | Weekly demand by product | Reorder quantities and markdowns |
| Manufacturing | Output, downtime, rejection | Root cause of a quality issue | Machine failure risk | Production scheduling |
| Distribution | Sales and collections by dealer | Why collections slowed | Late-payment risk by dealer | Credit limits and visit plans |
| Healthcare | Patient volumes and wait times | Why no-shows rose | Appointment demand | Staff rosters |
| Education | Enquiries, admissions, results | Why conversions fell for one course | Enrolment forecast | Counselling and marketing allocation |
| Financial services | Loan book and arrears | Why defaults rose in a segment | Default risk scores | Lending rules and pricing |
| Environmental compliance | Monitoring results against limits | Why a parameter exceeded | Pollution forecasts | Operational changes to stay compliant |
Example: Data Analytics for a Retail Business in Indore
A retailer with eight stores in Indore and nearby towns runs on a billing system, a spreadsheet of purchases and the owner's experience.
- Descriptive: a daily dashboard of sales, margin and stock by store and category replaces end-of-month reports. The owner sees for the first time that two categories make a loss after discounts.
- Diagnostic: breaking down a fall in one store's sales shows footfall is steady but average bill value fell after a competitor opened nearby, mainly in one category.
- Predictive: a simple weekly forecast by category and store, using two years of sales and festival dates, improves stock planning before Navratri and Diwali.
- Prescriptive: reorder suggestions based on the forecast, current stock and supplier lead times cut emergency purchases and dead stock.
None of this needs big data or machine learning. It needs clean data from the billing system, agreed definitions and a few well-designed views.
Building a Data Culture
Tools do not make an organisation data-driven; habits do. Practical steps:
- Start with decisions, not data. List the five decisions that matter most each month and build reports around them.
- One version of the truth. Agree definitions and stop the parallel spreadsheets.
- Make data easy to reach. Dashboards on phones, short weekly summaries, and alerts for exceptions.
- Ask "what does the data say?" in meetings. And accept answers that contradict opinion, including the owner's.
- Train people to read data. A short session on reading charts, averages and percentages goes a long way.
- Celebrate decisions improved by data. Share examples where data changed a decision for the better.
Tools for Each Type
| Type | Common tools |
|---|---|
| Descriptive | Excel and Google Sheets, Power BI, Looker Studio, Tableau, accounting and ERP reports |
| Diagnostic | Power BI and Tableau drill-downs, SQL, Excel pivot tables, Python or R for deeper analysis |
| Predictive | Excel forecasting for simple cases, Python (pandas, scikit-learn, statsmodels), R, cloud machine learning services |
| Prescriptive | Excel Solver, Python optimisation libraries, specialised planning software, rules built into ERP or CRM |
The tool matters less than the data and the question. See Power BI vs Tableau vs Looker Studio for dashboard tool choices.
Data Analytics vs Statistics, BI and Data Science
The terms overlap, and vendors use them loosely. A practical way to separate them:
| Term | Focus | Typical output |
|---|---|---|
| Data analytics | Using data to answer business questions, across all four types | Reports, analyses, forecasts and recommendations |
| Business intelligence (BI) | The systems and dashboards that deliver descriptive and diagnostic analytics | Data models, dashboards and self-service reports |
| Statistics | The mathematical methods for summarising data and drawing conclusions under uncertainty | Estimates, tests, confidence intervals and models |
| Data science | Building predictive and prescriptive models, often with programming and machine learning | Forecasting models, scoring systems and recommendation engines |
| Data engineering | Moving, cleaning and storing data reliably so the above can work | Pipelines, data warehouses and data quality checks |
A small business rarely needs separate teams for each. It needs someone who can get the data right, build the reports people use and answer "why" questions, with specialist help for forecasting or optimisation when a decision justifies it. See descriptive vs inferential statistics for the statistical side.
The Data Analytics Process, Step by Step
Whatever the type, most analytics work follows the same sequence:
- Define the question. What decision will this support, who will make it, and how often?
- Identify and collect the data. Which systems hold it, at what level of detail, and over what period?
- Clean and prepare it. Remove duplicates, fix codes, handle missing values and join sources. This is usually the longest step.
- Explore it. Look at distributions, trends and outliers before drawing conclusions.
- Analyse. Apply the methods suited to the question: summaries, breakdowns, statistical tests, forecasts or optimisation.
- Communicate. Present the answer in a form the decision-maker can use, with the key number first and the evidence behind it.
- Act and measure. Make the decision, then check whether the outcome matched the expectation, and feed that back into the next cycle.
Steps 1 and 7 are the ones most often skipped, and they are what turn analysis into value.
The Data Each Type Needs
| Type | Minimum data | Better with |
|---|---|---|
| Descriptive | Transactions or records for the period, with consistent codes | Targets, prior-year comparisons and master data such as products and regions |
| Diagnostic | Record-level detail with the dimensions you might break down by | Context data: prices, campaigns, staff changes, competitor events |
| Predictive | Enough history to see patterns, often two or more years for seasonal businesses | External drivers: festivals, weather, economic indicators, marketing activity |
| Prescriptive | Reliable forecasts plus costs, capacities, lead times and business rules | Feedback on decisions taken and their outcomes |
If the data for a type is not available, the first project is to start capturing it, not to buy a more advanced tool.
Measuring the Value of Data Analytics
Analytics should pay for itself. The returns are usually in four places:
- Time saved: hours no longer spent rebuilding reports; easy to measure and often enough on its own to justify automated reporting.
- Leakage stopped: overdue payments chased sooner, stock-outs avoided, discounts that lose money identified.
- Better decisions: inventory, pricing and marketing spend moved to what works, measured against what happened before.
- Risk reduced: problems spotted early, such as a customer about to leave or a quality drift on the line.
Set a baseline before the project, such as hours spent on reporting, days of stock held or average collection period, and measure the same numbers afterwards. Projects that cannot name a number they will improve usually do not improve one.
Common Mistakes in Data Analytics Projects
- Starting with tools instead of questions. Buying a BI platform or an AI product before deciding what it should answer.
- Skipping data quality. Building dashboards on duplicate customers and inconsistent product codes.
- Too many metrics. Dashboards with forty charts that nobody reads.
- Jumping to prediction. Forecasting before the basic reports are trusted.
- No owner. Reports that nobody is responsible for keeping accurate.
- Ignoring the people side. Analysis that contradicts how managers already decide, presented without explanation, is simply ignored.
Choosing a Data Analytics Partner
If you bring in outside help, look for a partner who:
- Starts by asking about your decisions and processes, not by demonstrating a tool
- Is tool-neutral and recommends based on your data, skills and budget
- Spends time on data quality and definitions before building visuals
- Explains methods in plain language and shows how results were reached
- Hands over documentation and trains your team, so you are not dependent on them
- Can show examples of similar work in your industry or with similar data
Data Analytics for Compliance and Environmental Data
The four types apply well beyond sales and marketing. Industries that report to regulators generate large volumes of monitoring data, such as ambient air quality, stack emissions, effluent quality and noise, and the same ladder applies:
- Descriptive: monthly and annual summaries against NAAQS or consent limits, exceedance counts and data capture rates.
- Diagnostic: why a parameter exceeded, by linking concentrations with wind direction, production rates, fuel changes or maintenance events.
- Predictive: forecasting high-pollution days, or estimating future concentrations with dispersion models such as AERMOD.
- Prescriptive: choosing operating changes, stack heights or control measures that keep the plant compliant at the lowest cost.
Because the results go into consent applications and audits, data validation and documentation matter even more than in business analytics. See our air quality analysis service for how this is done in practice.
Glossary of Data Analytics Terms
| Term | Meaning |
|---|---|
| KPI | Key performance indicator: a metric tied to a goal, such as monthly revenue or on-time delivery |
| Dimension | A way of slicing data, such as region, product, customer type or month |
| Measure | A number being analysed, such as sales value, quantity or count of orders |
| Drill-down | Moving from a summary to more detailed levels, such as from region to store to product |
| Cohort | A group sharing a starting point, such as customers acquired in the same month |
| Seasonality | A regular pattern that repeats, such as festival-season sales peaks |
| Forecast accuracy | How close forecasts were to actual results, often measured as average percentage error |
| Correlation | Two measures moving together; not proof that one causes the other |
| Model | A mathematical representation used to predict or explain, such as a regression equation |
| Data warehouse | A central store of cleaned, modelled data from several systems, built for analysis |
| ETL | Extract, transform, load: the process of moving data from source systems into an analysis store |
| Data quality | How complete, accurate, consistent and timely data is for its purpose |
How to Get Started
- Pick one decision that is made regularly and is costly to get wrong, such as purchasing, collections or marketing spend.
- Find the data behind it and check its quality: missing values, duplicates, inconsistent codes.
- Build the descriptive view that shows what is happening, with agreed definitions.
- Use it for a month and note the "why" questions people ask.
- Add diagnostic drill-downs for those questions.
- Forecast only the numbers where better foresight would change the decision.
- Measure the result: time saved, errors avoided, money recovered.
Our data analytics services help businesses in Indore and across India move up this ladder, from automated reporting to forecasting and decision support. For dashboards specifically, see our business dashboard development service.
Frequently Asked Questions
What are the 4 types of data analytics? Descriptive (what happened), diagnostic (why it happened), predictive (what is likely to happen) and prescriptive (what to do about it).
What is the difference between descriptive and diagnostic analytics? Descriptive analytics summarises what happened, such as sales falling 10%. Diagnostic analytics explains why, by breaking the change down by segment, period and cause.
What is the difference between predictive and prescriptive analytics? Predictive analytics estimates what is likely to happen, such as next month's demand. Prescriptive analytics recommends what to do, such as how much stock to order, taking the prediction and business constraints into account.
Which type of data analytics is most used? Descriptive analytics, in the form of reports and dashboards. It is also where most businesses gain the most, because every other type depends on it.
Is machine learning required for predictive analytics? No. Many forecasts and risk scores use classical statistical methods such as regression and exponential smoothing. Machine learning helps when there are many variables and a lot of history.
What is real-time analytics? Analysis of data as it arrives, used for decisions made within minutes, such as fraud detection, production monitoring and logistics. Most management reporting does not need it.
How do small businesses start with data analytics? By automating one important report with agreed definitions, using it regularly, then adding drill-downs and simple forecasts where they would change a decision.
Which type of analytics should a small business start with? Descriptive analytics: one automated, trusted report or dashboard for the decision that matters most, such as sales and collections. Once it is used every week, add diagnostic drill-downs for the questions people ask, then forecasts where better foresight would change a decision.
How long does it take to set up data analytics for a business? A first automated report or dashboard on one or two data sources can be live in one to two weeks. A set of dashboards on a proper data model typically takes four to eight weeks, and forecasting or optimisation projects follow once the base data is reliable.
What does a data analyst do? Collects, cleans and analyses data, builds reports and dashboards, investigates why numbers change, and, with more experience, builds forecasts and models that support decisions.
Something in the business is off and nobody can say why? DSTechnoverse runs diagnostic analytics engagements — segmenting the problem, testing hypotheses against your data, and finding the cause rather than the symptom. We are based in Indore, Madhya Pradesh and work with businesses across India. See our data analytics and BI implementation services, or talk to our team about what changed.