Reporting that writes itself

Part of a single argument: the engagement gap is a workload problem.

Every team that starts measuring its sessions produces a beautiful first report. The second is thinner. The third does not exist.

This is so consistent that it is worth treating as a law rather than a series of individual failures. The cause is always the same: the report was assembled manually, and manual assembly has a half-life.

Manual reporting has a half-life

A report that takes two hours will be produced when someone has two hours, which is roughly never once the initial enthusiasm has passed and the person who cared has moved on to the next thing.

This is predictable enough to plan around. If a report cannot be produced in under ten minutes, assume it will not exist after month two, and design accordingly rather than being disappointed later.

It is better to have a thin automatic report forever than a rich manual one twice.

Decide the shape once, then stop

Automation is only possible if the questions are stable. Agree the three or four things you report on for every session and then resist improving them.

Teams that redesign the report each cycle guarantee it stays manual, because there is never anything consistent enough to automate. The improvement instinct is the enemy here, which is counterintuitive and worth saying out loud to whoever keeps having good ideas about the format.

Push the exception, not the summary

The most useful automated reporting is not a dashboard. Dashboards require someone to remember to look, which is the same failure mode in a new costume, and the looking stops at about the same rate the manual reports did.

What works is a message when something is unusual. A topic where scores dropped. A session with abnormal drop-off. A question suddenly being failed by everyone, which usually means the question broke rather than the audience did.

Exceptions arrive without being requested and they carry their own justification for attention. Summaries have to be sought out and compete with everything else.

Keep a human in the sending

Fully automatic distribution goes stale and gets filtered. Recipients learn within a month that the weekly email is machine-generated and stop opening it, at which point you have automated the production of something nobody reads.

The version that survives is generated automatically and sent by a person who has glanced at it. That glance costs a minute, catches the occasional nonsense, and keeps the thing trustworthy enough to be read.

It also means somebody sees the numbers every week, which turns out to be most of the value of reporting in the first place.

The minimum worth building

One automatic message, to one person, when one thing is unusual.

That is a smaller ambition than most reporting projects start with and considerably more likely to still be running next year. You can add to it once it has survived a quarter, which is the only evidence that matters.