Most businesses do not have a data problem. They have a question problem - they collect plenty of numbers but never frame a question sharp enough for those numbers to answer. Here is the sequence that turns statistics from a reporting chore into a problem-solving tool.
Step 1: Turn the complaint into a measurable question
Business problems arrive as complaints: "customers are churning", "the new hires are not working out", "marketing spend feels wasted". None of these can be analysed. Rewrite each as a question with a number in it:
- "Customers are churning" → What percentage of customers who joined in Q1 were still active at month six, and how does that compare with last year?
- "Marketing spend feels wasted" → Which channels produced customers whose first-year revenue exceeded their acquisition cost?
If you cannot write the question with a number in it, you are not ready to analyse anything yet.
Step 2: Decide what answer would change your decision
Before touching the data, write down what result would make you act differently. "If churn at month six is above 20%, we pause new acquisition and fix onboarding." This one habit eliminates most wasted analysis, because it forces you to distinguish curiosity from decisions.
Step 3: Find the smallest dataset that answers it
You almost never need everything. A well-chosen sample of 200 customer records will answer most questions as well as 200,000, and you can pull it this afternoon. Watch for the classic trap: only analysing the customers who are easy to reach, which quietly biases the answer.
Step 4: Look at the spread, not just the average
The average is where analysis begins, not ends. Average delivery time of four days can hide the fact that one order in ten takes a fortnight - and it is that tail that generates complaints and refunds. Segment the data, look at the distribution, and check whether a handful of extreme values are driving the whole picture.
Step 5: Test whether the difference is real
The most common expensive mistake in business is acting on a difference that is just noise - a good week, a small sample, a seasonal blip. Before you roll out a change, ask whether the improvement is bigger than the normal week-to-week variation. That is exactly what statistical significance measures, and it is why controlled tests beat before-and-after comparisons.
A worked example
A retailer believes a new checkout layout increased basket size. Averages: £42 before, £45 after. Statistics adds three things: the variation in daily basket size (large), the sample period (nine days), and a test - which shows this difference is well within normal fluctuation. The retailer runs a proper split test for four weeks instead, finds a genuine £1.80 lift, and rolls it out with confidence rather than hope.
Building the habit
Problem-solving with data is a repeatable discipline, not a talent. The Statistics for Business course at London School of Business UK teaches it through exactly this kind of case work. Enquire today.