Premiering soonAI Projects PMO: Portfolio, Governance and Delivery
Many organisations now have several AI pilots running at once, often with no shared view of cost, risk or value. An AI PMO, or AI programme management office, brings order to that work.
In this video we explain how to set one up and run it. We cover building a single intake for AI ideas, a simple scoring model for value, feasibility and risk, stage gates from pilot to production, governance for data, security and model use, and the reporting a leadership team actually needs.
We also show how AI tools can help the PMO itself: summarising status reports, tracking actions and drafting steering-committee packs, with people making the decisions.
This is a practical guide based on how Egisedge runs delivery for AI and Oracle programmes. Templates referred to in the video are available in the Egisedge resource library.
Learn more: https://egisedge.com
#AIPMO #ProgrammeManagement #AIGovernance
Transcript
[Scene: Presenter on camera. On-screen title: "How to run an AI PMO".]
PRESENTER: Most organisations we speak to have more than one AI pilot running. Finance has one. HR has one. IT has two. Each has its own budget, its own vendor and its own idea of success. Nobody has the full picture. That is the problem an AI PMO solves. In this video we explain how to set one up and run it.
[Scene: Graphic. Scattered boxes labelled "Pilot" with different colours, then they snap into a single board with columns.]
PRESENTER: An AI PMO is a programme management office for AI work. It does not build the solutions. It makes sure the right ones are built, safely, and that value is tracked.
[Scene: Chapter card, "One intake for AI ideas".]
PRESENTER: Start with a single intake. Every AI idea comes in through one simple form. What problem does it solve? Who owns it? What data does it need? What happens today without it?
[Scene: Screen recording of an intake form built in Oracle APEX, with fields: Title, Business owner, Problem, Data sources, Current process, Expected benefit.]
PRESENTER: Keep the form short. If it takes more than ten minutes, people will go around it. We often build the intake as a small APEX application, so it sits next to the rest of the organisation's data.
[Scene: Chapter card, "Scoring value, feasibility and risk".]
PRESENTER: Next, score each idea. We use three headings. Value: what changes for the business, in time, cost or quality. Feasibility: is the data available and good enough, and do we have the skills? Risk: what could go wrong, for customers, staff or compliance?
[Scene: Scoring matrix. Ideas plotted on value against feasibility, with bubble colour for risk.]
PRESENTER: Plot the ideas. High value and high feasibility go first. High value but low feasibility may need data work before anything else. Low value ideas wait, however exciting they sound.
[Scene: Chapter card, "Stage gates: pilot to production".]
PRESENTER: Then set stage gates. We use four. Idea. Pilot. Scale-up. Production. Each gate has clear entry criteria. For example, a pilot cannot move to scale-up without a measured result against a baseline, a named business owner, and a support plan.
[Scene: Horizontal flow with four gates and a short checklist under each.]
PRESENTER: Gates stop pilots from running forever. They also stop half-tested tools going live because someone senior likes them.
[Scene: Chapter card, "Governance: data, security, models".]
PRESENTER: Governance is where an AI PMO earns its place. Agree the rules once, and apply them to every project. Which data can be sent to which model? Where may data be processed? Who reviews prompts and outputs? How are models and vendors approved? What must be logged?
[Scene: One-page policy summary on screen with five headings: Data, Location, Review, Approval, Logging.]
PRESENTER: Bring your security, legal and risk teams in at the start. It is much faster than asking them to approve something already built.
[Scene: Chapter card, "Reporting that leaders use".]
PRESENTER: Reporting should fit on one page. How many ideas are at each stage. What each live solution costs to run. What benefit it is delivering against its baseline. And the top risks and decisions needed. Leaders do not need a slide for every pilot.
[Scene: One-page portfolio dashboard with four sections matching the narration. Sample data.]
[Scene: Chapter card, "Using AI inside the PMO".]
PRESENTER: The PMO can use AI too. We use it to summarise weekly status reports, pull out actions and owners from meeting notes, and draft the steering committee pack. A person checks every output. The decisions stay with people.
[Scene: Screen showing a set of status reports and an AI-drafted summary with actions and owners highlighted.]
[Scene: Chapter card, "Recap". Presenter on camera.]
PRESENTER: To recap. One intake. A simple score for value, feasibility and risk. Clear stage gates from pilot to production. Governance agreed once and applied every time. One-page reporting. And AI to cut the admin, with people deciding. If you would like help setting up an AI PMO, visit egisedge dot com.
[Scene: End card with egisedge.com.]