Engineering time, by process, person and estimate

The leak

The technology head wanted a plain answer to a plain question: what is engineering spending its time on, which parts of the business is that time serving, and are the estimates we give the business anywhere near what the work actually takes. The raw material existed. Every engineer logged work against tickets, and tickets were structured as parents and children under business initiatives. But the ticketing system's own reports could not roll logged time up through that hierarchy into a view by process, team and person, or set it against the original estimates, so the question was answered by feel.

The constraint

The data was messy in the ordinary ways: work logged against children rather than parents, estimates on some tickets and not others, teams that changed mid-quarter. The dashboard had to be honest about coverage, showing how much logged time it could attribute and how much it could not, rather than presenting a tidy chart over incomplete data. And it had to serve one reader well before serving many.

The system

A pipeline that pulls the raw work logs and ticket structure, walks each log up through its parent chain to the business process it serves, and attributes time by process, by team and by person. Estimates and actuals are set side by side at every level, so the question "how far off were we" has an answer per initiative and per team.

On top of it, the views the technology head asked for: where time went this month and this quarter, which processes consumed the most engineering, which teams and people are concentrated where, and where estimates diverge most from actuals. Each view carries a written summary generated from the data, so the reader gets the conclusion first and the charts second. Unattributable time is shown as its own bar, not hidden.

The number

Engineering time by business process, team and person, and estimate against actual at every level, for the person who makes staffing and commitment decisions. The gap between estimate and actual is now a number per team rather than a feeling.

What I would do differently

Fix the logging conventions before building the roll-up. A short agreement on where time is logged and where estimates live would have raised attribution coverage from the first month. The dashboard made the gaps visible, which drove the conventions, but the order could have been reversed.