From data to decisions
Every organization generates records: sales, inquiries, deliveries, website visits, support requests, payroll, inventory. Data analytics is the practice of examining such records to find patterns, answer questions and support choices. It is not a single technique but a spectrum, ranging from simple counts and averages to statistical models and machine learning.
The value is not in the data itself, but in whether it changes a decision for the better. A report nobody acts on is an expense. This article is general information; results depend on the organization, the data and the people using it.
Four common levels of analysis
Analytics is often described in four stages, each answering a different question. The categories overlap, but they are a handy way to think.
Descriptive: what happened? Summaries of past activity, such as monthly sales by region, average call-handling time or the number of returns last quarter. This is the foundation, and many useful decisions need nothing more. Business intelligence dashboards usually sit at this level.
Diagnostic: why did it happen? Digging into causes. If sales fell in one region, analysts may compare it against product mix, pricing changes, stock shortages or seasonal effects. This often involves splitting data into groups and looking for differences. It generally gives clues rather than proof.
Predictive: what might happen? Using historical patterns to estimate future outcomes, such as expected demand next month or which customers may cancel. Forecasts rely on the assumption that the future resembles the past, and they carry uncertainty. Many predictive methods use machine learning, trained on past examples.
Prescriptive: what should we do? Suggesting actions, for instance recommending stock levels or scheduling staff to match expected demand. It combines predictions with rules, costs and constraints. Humans still need to judge whether the recommendations make sense.
What good analytics needs
A clear question. Starting with “what do we want to decide?” is better than starting with “what can we do with our data?” A focused question tells you which records matter.
Reliable data. Analysis cannot be better than its inputs. Duplicated customers, missing fields and inconsistent definitions produce misleading numbers. Data quality is often the largest practical obstacle, and data governance helps by deciding who owns definitions and standards. Where data comes from several systems, a data warehouse is a common way to bring it together.
Shared definitions. If two departments define “active customer” differently, their reports will disagree. Agreeing on terms avoids long arguments over whose number is right.
Suitable tools and skills. Spreadsheets handle a surprising amount. Larger or more complex needs may call for databases and specialized software. Skills in statistics and in the business area both matter, and ordinary staff benefit from a basic level of digital literacy.
A route to action. Someone needs to own the decision and be able to use the finding.
Where analytics can mislead
Numbers can feel authoritative even when they are shaky. Common traps include:
- Poor or incomplete data. If the data covers only the customers who answered a survey, it may not represent everyone.
- Correlation mistaken for causation. Two measures that move together do not necessarily cause each other. Ice cream sales and sunburn both rise in summer; neither causes the other.
- Cherry-picking. Searching through many comparisons until something looks significant, then reporting only that, tends to produce patterns that are due to chance.
- Small samples. Conclusions drawn from a handful of cases can swing wildly.
- Out-of-date models. A forecast trained on past conditions can fail when customer behavior, prices or the wider environment change.
- Hidden bias. Historical data may reflect past unfair or arbitrary decisions, and models trained on it can repeat them. Algorithmic bias and training data discuss how this happens when automated systems are involved.
- Misleading charts. Truncated axes and selective ranges can exaggerate small differences; data visualization covers how to avoid this.
- Measuring what is easy, not what matters. If an organization rewards a metric, people may optimize the metric rather than the goal behind it.
- False precision. A forecast of 1,243 units sounds exact, but the real range may be far wider. A range is usually more honest than a single figure.
- Privacy and fairness issues. Collecting and combining personal information creates obligations and risks. Data minimization suggests collecting only what you need, and rules vary by place, so qualified advisers should be consulted.
A hypothetical example
Suppose a small online shop sees orders rise 20% after a promotion. Descriptive analysis shows the rise. Diagnostic analysis might reveal that most of the increase came from existing customers buying earlier than they otherwise would, which means part of the gain is borrowed from future weeks. A predictive model might then estimate a dip next month. Without the diagnostic step, the shop could wrongly conclude that the promotion permanently raised demand. The numbers here are hypothetical, but the lesson, to ask why before extrapolating, is general.
Using analytics sensibly
Treat analysis as one input alongside experience and context. Ask where the data came from, what it leaves out, how uncertain the result is and what would change your mind. Start small with a question that matters, check the basics of data quality, and have someone other than the author review important analysis. Where decisions affect people significantly, human oversight remains important. Analytics can make decisions better informed, but it cannot remove uncertainty or the need for judgment.