Project Fact or Fiction

Work items could sit in progress indefinitely without anything in the system noticing. Nothing tracked how long something had been stalled, so nothing raised a hand when it started to slip.

Status surfaced one of two ways: somebody remembered to go looking, or the delay had already become a problem people were reacting to. Both of those are the same failure — the signal arrived because a person went and got it, not because the system sent it.

Every work item is pulled and checked against how long it has been sitting in its current in-progress state. That age is then bucketed into a risk tier using a percentile model — the P70, P80 and P90 marks of how long this kind of work actually takes — rather than a flat day count picked by hand.

An automated flow turns those tiers into weekly digests, framed differently for each audience: several delivery-management levels, plus a separate digest written for leadership. An agent tuned to a specific coaching methodology reads each aging item and writes an assessment of why it is stuck, rather than restating what it says.

Delivery risk stopped being something a person had to remember to go and check, and became something that arrived on a schedule whether anyone thought to ask or not. Leadership went from reconstructing where things stood to opening a digest that already said so.

The reason it gets read is in how it was built. A percentile model rather than a flat threshold means a flagged item is genuinely unusual for that kind of work — the tier reflects the real distribution instead of a number somebody guessed, which is what makes it defensible when it lands in front of a delivery manager. Framing each digest for its audience is why four different levels each get something worth their attention rather than one report written for nobody in particular.

And the agent is tuned to judge rather than summarize. A digest that restated status would have been one more thing to skim; one that says why an item is stuck starts a coaching conversation instead. That was a decision about how the agent was prompted, made at the same time as the decision to send the digest at all.

Delivery risk became a standing, audience-aware signal that arrives every week on its own, instead of a question someone had to think to ask.

The work that is quietly slipping now announces itself, with a reading of why, while there is still time for that to matter.