What will you do differently?

Someone asks for "a dashboard on customer retention". You can build that. It will be looked at twice and then never again.

Before opening anything, ask:

If this number comes back high, what do we do? If it comes back low, what do we do?

If both answers are the same, the analysis has no value — however good the chart is. You have found out something true and inert.

If the person cannot answer, that is not a reason to skip the work. It is the work. Keep asking until there is a decision attached, because until there is, no amount of SQL will make the output matter.

Turning a request into a question

A request sounds like: "How are we doing on retention?"

An answerable question names four things:

• The population. Which customers? Since when? Including trials?
• The measure. Retained meaning what, exactly — logged in, paid again, still subscribed on day 90?
• The comparison. Against last quarter? Against another segment? Against a target? A number with nothing to compare to cannot be high or low.
• The decision. What happens at each outcome.

Write those four down and read them back to the person who asked. This conversation takes ten minutes and routinely saves a week — usually because it surfaces that they meant something quite different by "retention" than you were about to measure.

The comparison is where analysis goes wrong quietly

"Retention is 62%." Is that good?

You cannot know. It is only meaningful against something: the previous cohort, a different segment, the industry, the target, or the same cohort before a change you made. Choosing the comparison is a judgement, it is arguable, and it belongs in the open where someone can argue with it.

A number presented without its comparison is not neutral. It quietly invites the reader to supply their own, and they will supply a flattering one.

Write it down before you query

One short paragraph: the question, the population, the measure, the comparison, the decision.

It takes five minutes, it makes the SQL nearly write itself, and three months later it is the only thing that explains why the number was defined the way it was.

Try this

Take a dashboard you maintain. For each chart, write the decision it informs. The ones where you cannot are candidates for deletion — and deleting them makes the rest more likely to be read.