1,700 items in one management view

Request: „Merge ticket system, roadmap, board and repository into one dashboard: progress, mapping, risks, technical maturity.“

1. The AI pulls items from the ticket system, the code-hosting roadmap, the planning board and the repository
2. It joins them on keys and makes the mapping between the sources visible
3. It computes progress, risk and technical maturity per epic and feature
4. It generates the dashboard as a static page that is rebuilt on every run — a single file without external calls; twice a week a run re-collects all figures and publishes them internally, the measured live run took around a minute
5. A manager cockpit adds, as a pilot, open staffing needs from the resource system, a decision queue, data-source status and risk signals

Every system comes with its own reporting; in most organisations the overall picture is compiled by hand in slides.

Which KPIs matter, what a value means and what “done” means is decided by management, which also judges data quality itself; qualification ran along several management roles.

A board status was misread — “slicing tasks” counted as done; after a query, one progress figure dropped from a good 90 to just under 60 percent. A keyword match counted around a sixth of the tickets twice. And the dashboard is a blueprint for one product; the KPIs are not hardened yet and scaling to further products is still open.

Any management team whose truth is spread across several tools. The value is the transparency that emerges when the sources are joined.

Decisions based on individual perception are expensive. For one of the unit's products, the agentic PMO merges four sources for the first time: ticket system, code-hosting roadmap, planning board and repository. Around 1,700 items were consolidated into one delivery status; around 130 epics and features are measurable within it — progress, mapping, risks and technical maturity at a glance. From the first prompt to a running app took a good week. How much transparency only emerges when the sources are joined shows in the detail: which item belongs to which epic, and which feature already has a trace in the repository, becomes visible across sources. It gets honest with the mistakes: a misread board status made progress drop from a good 90 to just under 60 percent once someone asked; exactly these exposed data questions are often more valuable to leaders than the curve itself. Twice a week a run re-collects all figures, in around a minute. The dashboard is still a blueprint; hardening the KPIs and scaling to further products come next.