How do I interpret data at work?
Ask five questions of every number you are shown: what changed, compared with what, how certain is it, what else could explain it, and what would you do differently because of it. Decoding a chart tells you what it says. Interpreting it tells you what you are entitled to claim, which is the part most people were never taught.
The guides that rank for chart questions stop at decoding. They teach a reading order, title then axes then units then legend then trend. But knowing what a chart says is not the same as knowing what it means. In the forums where people ask how to learn data interpretation, nobody teaches the second half.
That second half is a decision habit, not a talent. A dashboard shows a change, not whether it matters. As one reader put it in a data forum, "Just because you look at the charts doesn't mean you understand them." One written line per question, before the meeting, turns looking at a dashboard into a position you can defend.
Which five questions should I ask of any chart?
What changed? Compared with what? How certain is it? What else could explain it? What would the evidence justify doing? Those five cover the checklist an AI answer already gives you: the topic, the measurement, the axes, the scale, and the source. They add the part the checklist leaves out, which is what the number is evidence for.
Take them in order and answer each in one written line. The first four describe the claim and its limits. The fifth forces the claim to become a decision. Written answers survive the meeting, because a number you can only half explain is one you will agree with rather than question.
Each question narrows the next. If you cannot say what changed, there is no comparison to test. If you cannot say how certain the change is, there is no reason to act. Skipping to the fifth question is how a team commits to a trend that was never there.
What should every comparison be compared with?
A comparison needs a named baseline: last month, the same month last year, the plan, the other team, or the same metric from before the change. A rise from 3 percent to 4 percent is a one point move and a one third increase, and both descriptions are true. Only the comparison tells you which one is worth saying.
Baselines change the story more than the data does. A window that opens after a bad quarter shows a flattering trend that a two year view erases. A raw count with no denominator makes a small team look like the worst performer.
So ask two things plainly: compared with what, and why that choice. Then find the denominator. Rates travel between teams and periods. Raw counts often do not.
How do I separate a pattern from an explanation?
A pattern is what the data shows. An explanation is your story about why it happened. Keep the two apart until the explanation makes a prediction the data could contradict. Anscombe's quartet makes the point cleanly: four small datasets with nearly identical averages, variances, and correlations produce four very different graphs. The summary hides the shape.
Name the mechanism, then say what else you would expect to see if it were true, and check whether you do. Then look for a lurking variable. Simpson's paradox is the standard warning, because a trend can hold inside every subgroup and reverse when the groups are combined.
Kahneman's phrase for the failure is "what you see is all there is." The mind builds a confident story from the evidence in front of it and does not notice the evidence that is missing. That is why the explanation deserves its own line, and why "what would change my mind?" is worth asking out loud.
How can I spot scale tricks and missing context?
Check four things before you believe a visual: where the axis starts, what the units are, what time window is shown, and what is missing. A y-axis that begins at 40 instead of 0 turns a small move into a cliff, and a window that opens after the last dip flatters a flat trend. Both are choices.
Edward Tufte's work on graphical integrity named the gap between the size of an effect as drawn and its size in the data. The mismatch is visible once you look for it. Truncated axes, dual scales that line up two unrelated series, and overlapping pie slices are the common forms.
Missing context is quieter than a distorted axis. Ask about the sample size, the response rate, who was left out, and whether the definition of the metric changed partway through the series. A definition change can produce a step in the line that looks exactly like performance.
What action does the data justify, and what remains uncertain?
State your claim in one sentence, then state your confidence in plain words: this is clear, this is likely, or this is a guess. Say what you would do and what would make you reverse. A decision made on stated uncertainty is safer than one made on false precision, because it stays open to evidence.
Research on risk communication, including work by Gerd Gigerenzer and colleagues, has found that people understand a risk far better as a plain count than as a percentage or a relative change. The same applies to your dashboards. "Four of the ninety customers who signed up in March have churned" is a decision-ready sentence.
Confidence has a practice problem, though. Reading about these questions does not make them automatic, the way reading about a serve does not let you return one. Ten minutes a day on a real chart, scoring yourself, turns the routine into a reflex. Absorb is a daily comprehension practice on the App Store that runs that loop for visual material.