Confidence intervals: reading a range instead of a point

A single number feels precise. It usually isn't. When a test says a display added 12 units per store per week, the honest version is closer to "somewhere between 4 and 20, most likely around 12." Confidence intervals express that range. They show how much uncertainty surrounds an estimate, which is often more useful than the estimate itself.

The myth is that the point estimate is the answer and the range is a technicality. In practice, the range tells you whether you can act.

What is a confidence interval?

It's a range, calculated from the data, that's likely to contain the true value you're estimating. The standard interpretation is about the method: if you repeated the same study many times, about 19 of every 20 intervals built this way would contain the true value. That's the conventional level most analysts report.

The interval has a center, usually the point estimate, and a width. The width depends on how much the data varies and how much data there is. More stores, more weeks and less noise produce narrower intervals.

The confidence interval entry covers the technical details. For most business decisions, the key ideas are the center, the width, and whether the range includes zero.

How do you read one in a retail test?

Here's a hypothetical. A brand tests a new shelf placement in 25 stores against 25 matched controls. The estimate is that the new placement adds 12 units per store per week, with an interval from 4 to 20.

Three things follow. First, the whole range is above zero, so the placement very likely helps. Second, the effect could be as small as 4 or as large as 20. Third, if the placement only makes sense financially above, say, 10 units, the result is promising but not conclusive, since a meaningful part of the range sits below that line.

Now imagine the same estimate of 12 with an interval from negative 2 to 26. The center is the same. But the range includes zero and even a small loss. The test hasn't ruled out that the placement does nothing.

Same point estimate, very different decisions. That's why the range matters.

What makes an interval wide?

Variation. Stores differ, and each store's sales move from week to week. The more they bounce around, the wider the interval.

Small samples. With few stores or few weeks, the estimate is less certain. Doubling the number of stores doesn't halve the width, but it narrows it noticeably.

Poor matching. If test and control stores aren't similar, extra differences between them add noise. Matching on size, location type and recent sales helps.

If an interval is too wide to support a decision, the answer is usually a bigger or better-designed test, not a different way of reading the same result.

What are the common misreadings?

That the true value is equally likely anywhere in the range. It isn't. Values near the center are generally more plausible than values near the edges.

That overlapping intervals mean no difference. Two estimates with overlapping intervals can still differ meaningfully. Comparing them properly requires an interval for the difference itself.

That a narrow interval means the estimate is correct. A narrow interval means the estimate is precise. If the test design was flawed, for example with poorly matched stores, it can be precisely wrong. The monthly reports from NRS Insights provide category context that can help you judge whether a test period was unusual, and the July 2026 same-store sales report is the most recent.