Data Analytics Basics: Descriptive, Diagnostic, Predictive

Four questions analytics answers

Data analytics is not one activity but a ladder of questions. Each step builds on the previous one:

Most companies get the most value from the first two steps done well. Forecasts built on messy descriptive data are worse than no forecasts.

Descriptive analytics

This is reporting: totals, averages, trends and comparisons over time. Dashboards in BI tools such as Power BI, Metabase or Looker Studio are mostly descriptive.

Good descriptive analytics is consistent: the same metric is calculated the same way in every report, and everyone knows the definition.

Diagnostic analytics

Here you look for causes. Typical techniques:

• Segmentation — split the metric by channel, device, region, customer type.
• Drill-down — go from the total to the level where the change actually happened.
• Funnel analysis — find the step where users drop off.
• Cohort analysis — compare groups of customers who started in the same period.

The main trap is confusing correlation with causation. Two metrics moving together does not mean one drives the other. When the decision is important, confirm the cause with an experiment such as an A/B test.

Predictive analytics

Predictive analytics uses historical data to estimate the future: demand forecasts, churn probability for each customer, expected revenue. Methods range from simple trend lines and seasonality to machine learning models.

A forecast is always a range with uncertainty, not a promise. It is only as good as the history it learns from: if the business has changed sharply, old data may mislead.

Prescriptive analytics

The final step recommends actions: optimal prices, stock levels, delivery routes, which customers to offer a discount. It combines forecasts with constraints and goals. This level usually requires mature data and clear rules for how decisions are made.

Metric vocabulary

• Metric — any measurable value: orders, revenue, visits.
• KPI — a key metric tied to a goal; a team should have only a few.
• Dimension — what you slice a metric by: date, city, channel.
• Conversion rate — share of users who complete a target action.
• Retention and churn — share of customers who stay or leave over a period.
• AOV (average order value) — revenue divided by the number of orders.
• LTV (lifetime value) — revenue a customer brings over the whole relationship.
• CAC (customer acquisition cost) — marketing and sales spend per new customer.
• DAU / MAU — daily and monthly active users.
• Vanity metric — a number that looks good but does not inform decisions, like total registrations without activity.

From question to insight: the workflow

1. Formulate the question. "Why did repeat purchases fall in Q3?" is workable; "let's look at the data" is not.
2. Define the metric precisely. What counts as a repeat purchase, in what time window, which orders are excluded.
3. Collect the data from the CRM, website analytics, accounting, ad accounts.
4. Clean it: duplicates, test orders, missing values, different time zones.
5. Analyze: compare periods, segment, look at distributions, not only averages.
6. Visualize with the simplest chart that answers the question.
7. Formulate the insight and the action: what we learned and what we will change.
8. Measure the effect of the action and return to step 1.

Common mistakes

• Starting with a dashboard of everything instead of a specific question.
• Different departments calculating the "same" metric differently.
• Trusting averages that hide outliers and different customer groups.
• Drawing conclusions from very small samples.
• Building forecasts before the basic reporting is reliable.