Freshness alerts and evidence on the AI assistant

A fictional quarter in a central data team at a mid-size company. Four requests arrived, each sized to fill one of the team's two delivery slots, so two are done and two wait on their interim paths. Three facts are not yet known: when annual planning starts, whether the staffing view would read from the pipeline that has run late, and whether the executive team will decide on the assistant with or without evidence. Each pair leans on some of them, and its review says which.

1. DeferredOne margin for planningThe CFO: Finance and Sales each report gross margin, and the two disagree by enough to change a conversation. Annual planning will set targets and spending for a year on one of them. The CFO wants the margin planning uses agreed, with an accountable owner and a stated bridge to the other view.Today An analyst reconciles the two by hand each month, about two days of work, and each report carries a footnote explaining the other.
2. ChosenFreshness alertsThe managers who run the morning meeting: Three times this quarter the morning report was stale, and the first person to notice was in the meeting that used it. They want to be told when a pipeline is late, before the report is opened, and told what to use instead.Today An analyst checks the load timestamps before eight. It works on the mornings the analyst remembers and is in, and when it fails the meeting finds out for itself.
3. DeferredA staffing viewThe head of operations: Next week's roster is set every Monday from a spreadsheet an analyst pulls that morning, and it is committed for the week: too many people on shift costs money, too few costs service. They want a view that puts the demand forecast against capacity while the decision is still open.Today The Monday extract. When the analyst is out, the roster is set on the previous week's numbers, and it is set either way.
4. ChosenEvidence on the AI assistantThe executive team: They have seen a demo of an assistant that answers questions from the warehouse and want to know whether to roll it out. Someone has to run the evaluation: the question set, the data the assistant may see, who in the business reviews its answers, and what evidence would justify proceeding, narrowing or stopping.Today Nothing is launched. The question comes up every month, and a vendor demo is the only evidence anyone has.