Taking Admin Off a Growing PMO with AI
How a pilot-first AI operating model gave project managers hours back each week without changing the tools they already used.
- Engagement
- AI for PMOs
- Stage
- Pilot complete, rollout under way
- Stakeholders
- Partners and C-suite, PMO leadership, project managers, internal AI build team
The situation
A fast-growing ERP and CRM implementation firm was taking on more projects every quarter. Leadership wanted the PMO to keep scaling without adding more admin work or more headcount.
What was really going wrong
- Project managers were spending hours every week on status reports, meeting notes, budget tracking and chasing updates, instead of on clients and delivery.
- Risk was hiding behind green statuses: projects marked green could have budget, timeline or go-live problems that only surfaced late.
- Reporting looked different from one PM to the next, so leadership had no single, reliable view of the portfolio.
How we fixed it
Step 1: Find the work worth automating.
- Mapped a typical PM week to see where time actually went.
- Identified eight PM workflows that AI could take over or speed up, including status reporting, meeting notes, budget tracking and portfolio risk.
- Ranked them by time saved and risk, so the first automations would show value quickly.
Step 2: Design the operating model.
- Built an AI in Project Management operating model and roadmap for the partners and C-suite.
- Planned it in two phases, starting with a small pilot group before any wider rollout.
- Kept the firm's existing project tracker as the single source of truth, with AI working on top of it rather than replacing it.
Step 3: Build the first automations.
- Turned the weekly portfolio exception review into a reusable AI skill that checks every project against six rules, such as green projects with yellow or red scope, budget or timeline, go-live dates at risk or already passed, and status updates older than 14 days.
- Set up meeting summaries that turn call transcripts into decisions, action items, owners and next steps.
- Automated weekly budget burn, testing and issue-log reports from raw project data.
Step 4: Pilot, measure, then roll out.
- Ran the automations with a pilot group, with weekly working sessions with the internal build team to fix issues and refine outputs.
- Measured the time saved before planning the rollout to the rest of the PMO.
What we refused to allow
AI sending anything to clients without a project manager's approval, or replacing the tools the team already relied on. Every output was a draft until a person signed it off.
Tools we used
Claude, the firm's existing project tracker, meeting transcripts, and automated budget, testing and issue-log reporting.
The results
Four-step flow: Map the PM week, Design the operating model, Pilot with a small group highlighted as 8 to 10 hrs/week saved, then Roll out to the PMO. Illustrative, not to scale.
Illustrative, not to scale.
8 to 10 hrs/week
saved across the pilot group
2 to 4+ hrs
expected saving per PM per week once rolled out to the full PMO (modeled)
Weekly review
A portfolio exception review that now runs the same way every week, catching hidden risk before clients escalate
The lesson
AI in a PMO works when it fits the tools and rules people already use, starts small, and keeps a person in charge of every output.
