There is a graveyard in every business software stack: the dashboards. Built with real enthusiasm, demonstrated once, opened rarely. The problem is rarely the tooling. Dashboards answer questions nobody was asking on the day they looked. Dashboards are supply. Decisions are demand. Most reporting projects build supply and hope.
So flip it. Find the one document your leadership team actually asks for: the monthly report someone assembles by hand, the Friday numbers email, the board pack section that takes two days and three chasing messages to produce. That document has already passed the only test that matters: a real decision waits on it. It has demand.
How it actually works #
The Decision Document Test is simple: could this document generate itself from live data? Not a dashboard version of it. The actual document, in its actual format, with the numbers current instead of two weeks stale. If the answer is yes (and it almost always is, once the underlying systems are connected), that is your first automation. One self-assembling decision document beats ten dashboards, because it inherits an audience, a cadence, and a consequence.
The second-order effect is the interesting one. When the document assembles itself, the two days of gathering become two hours of judging. The senior person who used to compile it now interrogates it. That is the honest promise of AI in operations: removing everything that was queued in front of judgement. The businesses getting real returns from AI are the ones who picked artefacts with decisions attached, not the ones with the most pilots.
Running it on Monday #
On Monday: list every recurring report in the business. Cross out the ones no decision waits on and retire them outright; that is pure found time. Take the top survivor and scope what it would take for it to build itself. That scope is worth more than any AI vision statement.
Do not automate what nobody reads. Automate what somebody is waiting for.