A status and escalation agent built against a map of how the PMO actually decides. It reads project, cost and risk signals, escalates with a written reason, and leaves the calls that belong to people with people.
In most PMOs the weekly status is stitched together by hand from three systems, the escalation threshold lives in one senior person's head, and when a sponsor asks why something went red, the honest answer is a shrug.
Two PMs looking at the same signals make two different calls. When an escalation is challenged there is no record of the reasoning, and when one is missed there is no threshold to point to.
The escalation decision was broken into five layers: what triggers it, how the situation is framed, the real criteria behind the call, the action taken, and what is left behind to learn from. That map, not a prompt, became the build spec.
Every PMO decision was then catalogued and given a tier: the AI acts, the AI recommends, or a human decides. The most important outcome was a "no": the portfolio review verdict was examined and deliberately left with people. The agent escalates; people decide.
Boundaries keep mistakes cheap. The agent reads across systems but writes only to its own status view and escalation channel, and it steps back on initiatives whose context lives outside the data.
Each is something the PMO can point to when the agent's behaviour is questioned. Click a stage.
Explainability was the spec, not a feature added later.
Each decision is tiered before anything is automated. The most valuable result was the decision it declined to automate.
The decision map is the agent's spec and its audit trail, so judgment does not drift between the two.
Track initiatives, never grade people. That line was drawn with the people affected.
Proven on ResonAi's own delivery practice before it went to a client.
The same pattern fits status escalation, stage-gate reviews and risk triage. Start with a conversation about which of your calls should be mapped first.
📅 Book a call