Search Precision Consulting
Type to search · ↑↓ to navigate · Enter to open · Esc to close
AI for the PMO and Transformation Office
AI for the PMO and transformation office
one bar for every AI tool the office runs
Is your portfolio office getting the most from AI, and would you know if it were not? Every major portfolio platform now ships its own AI agent, and each governs itself and sees only its own data. We set one bar for all of them, review your business cases against written assessment criteria, and design the data spine that lets every tool work from the same numbers.
The first conversation is confidential, without obligation and without charge.
Independent by rule. After we review a programme's business case, we take no readiness, recovery, build or operating work on that programme for 24 months. Read our independence and conflict rules
How can AI be used in a PMO?
To draft packs and RAID narrative from source records, screen business cases against written criteria, compare the status a programme reports with the status its data shows, and forecast burn and cost at completion as ranges, back-tested on your own closed programmes before you rely on them.
Drafting now ships inside the major portfolio platforms. The harder work is holding every tool to one standard and knowing how good each output is.
What goes wrong when every tool brings its own AI?
AI starts competing with AI. Forecasts disagree, verdicts cannot be compared, proposals learn to game the screens, and the same problem is paid for twice.
- Competing numbers. Finance, the portfolio office and a vendor's platform each forecast the same programme with different models and data, and the committee spends its time reconciling them.
- Competing verdicts. Divisions score business cases or risks with their own prompts and thresholds, and scores cannot be compared across the portfolio.
- Adversarial loops. Proponents tune AI-drafted business cases to pass AI screens.
- Duplicated build. Several teams pay to solve the same problem on different tools, and none reaches production quality.
Each platform's agent governs only its own work, so the answer is to set the bar once, for the whole office, before every team builds its own: one register, one data spine, one set of criteria, and every AI tool held to the same standard. Teams then build inside the bar.
AI held to account. One bar. One baseline. One role per system.
What each one meansWho is the AI Business Case Review for?
It is for the people who decide which investments go ahead: an investment committee chair, a portfolio office lead, a senior responsible officer or a chief financial officer.
It fits when:
- a business case, or a batch of them, is coming to committee and you want it tested against written criteria before the decision;
- divisions score cases with their own prompts and thresholds, and the scores cannot be compared across the portfolio;
- cases may have been drafted with AI, and you want every claim in them traced to its evidence;
- you want the view from reviewers who will take no further work on that programme.
It does not fit when:
- you want help to write or strengthen a case. That is business case development, which we offer separately and never call a review;
- your framework requires a formal review of the case. This review does not replace one, and it gives no access to one;
- we have built, fixed or run the system the case is for. We never independently review our own work, so that review goes to another firm;
- you want us to deliver the programme afterwards. After a review, our rules bar readiness, recovery, build and operating work on that programme for 24 months.
What is an AI Business Case Review?
A review of one business case, or a batch, against written assessment criteria, with every finding traced to the text of the case. AI-assisted analysis does the tracing; an experienced assurance reviewer weighs it, signs the view, and a second reader checks every citation and figure before you receive it.
- A signed view of the case against each criterion. A person decides every rating.
- Every finding traced to the passage of the case it rests on, with every citation and figure checked by a second reader before release.
- The criteria cover strategic alignment, the options considered, the cost basis, the benefits logic and how benefits will be measured, deliverability, dependencies and risk, with cost and schedule assumptions tested against a reference class of comparable investments.
- AI-drafted cases. The review notes indicators that parts of a case may be AI-drafted, for the reviewer to weigh. It makes no finding that a person used AI. We recommend a disclosure rule, so proponents declare AI use and cases drafted with AI cannot game a screen run by AI.
For the whole pipeline, the Business Case Screen applies the same criteria to every case in the pipeline, as a proof of value planned at four to six weeks.
In government, an assessment that materially influences a funding decision is itself an AI use case under the Commonwealth policy, so a person decides every rating.
Fees are scoped to the case, or the batch, and agreed before work starts.
How does a review run?
Over one to three weeks a case. An experienced assurance reviewer leads, with an analyst, and a second reader checks every citation and figure before the view is released.
Planned elapsed time One to three weeks a case
| Stage | What happens |
|---|---|
| Scoping | You name the case, or the batch, and the decision it supports. We check the programme against our conflict register before the review is scoped. |
| Analysis | AI-assisted analysis traces each claim in the case to its evidence, and tests the cost and schedule assumptions against a reference class of comparable investments. |
| Review | The reviewer reads the case in full, weighs the findings against each criterion, decides every rating and signs the view. |
| Release | A second reader checks every citation and figure. You receive the signed view and discuss it with the reviewer. |
One to three weeks a case is the planned elapsed time. We replace it with measured times as reviews complete.
What does a Transformation Office AI Maturity Review cover?
Where your office stands against the six dimensions of the Blueprint, what AI it already uses, and the two uses most worth proving next. It is planned at four to six weeks. A Precision director signs every view and answers for how it was produced.
You receive:
- a baseline against the six dimensions of the Blueprint;
- an inventory of the AI in use in the office, sanctioned and not;
- a sample of up to five recent business cases, assessed for how the office's criteria were applied, with no verdict on any case;
- two candidates for a proof of value, each with a baseline.
Optional modules, scoped separately: a draft Consolidated AI Bar for the office; data spine readiness (shared identifiers and monthly snapshots); a sample comparing reported status with the status the evidence shows; the obligations the office's AI touches, set out as questions for your counsel; and a roadmap.
Who does it: an EPMO director leads, with an experienced assurance reviewer on the business case sample, a solution architect and an analyst. If we might go on to build, we say so when the review is scoped.
Without a formal EPMO? A portfolio scan covers the same ground in a planned two to three weeks for organisations that run a portfolio of projects without a formal office.
Fees are scoped to your office and agreed before work starts.
Our Managed EPMO clients can add this work to their existing service. About Managed EPMO
Can AI forecast programme cost and burn?
Yes, as a range, once the method has been back-tested on your own closed programmes.
Forecasts of burn and cost at completion are available on that condition. We test the method on work you have finished before we forecast work you have not.
The Blueprint
What is the Blueprint?
Our maturity model for AI in the portfolio office: five levels, from Individual to Orchestrated, across six dimensions.
The AI-Enabled Transformation Office Blueprint is an AI maturity model for EPMOs and transformation offices. It describes what good looks like for AI in a portfolio office, and we publish it here in full. This is version 1.0, published by Precision Consulting Corporation on 30 September 2026.
A Maturity Review places your office on it from evidence, and a Blueprint engagement designs the move to the Consolidated level and beyond.
The Blueprint
What does good look like for AI in a portfolio office?
One portfolio data spine, every AI output traced to its source, forecasts with a track record, and people holding every decision, all against one published bar. Seven principles describe it.
- One portfolio data spine. A single taxonomy and set of identifiers joins schedule, cost, risk, benefits, resources and decisions. AI is only as good as this join.
- Evidence-traceable outputs. Every AI-generated statement links to its source record. This is the control against fabricated content, and it is why a finding can be defended.
- Forecasts with a track record. Every forecast is a range, back-tested on closed work, and its accuracy is reported. A forecast without a track record is an opinion.
- People hold the decisions. A written decision rights matrix says what AI may draft, flag, recommend or carry out, and what only a person may decide.
- One published bar. Criteria for business cases, risk scoring and status ratings are written down once and applied the same way by every tool and every person.
- Hosting decided and written down. Programme data stays in environments the office has approved, with where each stores and processes data recorded, and it is never used to train public models.
- Value measured against a baseline. Hours per reporting cycle, forecast error, the lead time of warnings and how long decisions take are measured before go-live and after it.
The Blueprint
What are the five levels?
Individual, Sanctioned, Consolidated, Predictive and Orchestrated. Each level adds a capability and the governance that makes it safe to rely on.
| Level | In one line |
|---|---|
| Level 1: Individual | People use AI tools on their own. The office cannot say which tools, or what data went into them. |
| Level 2: Sanctioned | Approved tools, in approved environments, draft and summarise. Every use is registered and every output is reviewed by a person. |
| Level 3: Consolidated | AI works from one portfolio data spine against one published bar, and every output traces to its sources. |
| Level 4: Predictive | Forecasts of burn, cost at completion, schedule and benefits carry ranges and a published track record. |
| Level 5: Orchestrated | Bounded agents run defined workflows inside a decision rights matrix. People decide, and the office's AI is independently reviewed each year. |
The Blueprint
Why is level 3 called Consolidated?
Because level 3 is where the bar is built once, centrally, and every tool is held to it. Below it, each tool and each team sets its own standard, and AI starts competing with AI; above it, forecasting and orchestration have something consistent to stand on.
What is set centrally is the standard: the bar, the register, the data spine, the approved platforms and models, the evaluation standard and decision rights, each set once. Teams then build inside the bar.
The Blueprint
What are the six dimensions?
Value and use cases, data foundations, platform and security, governance and the bar, people and operating model, and measurement. An office can sit at different levels on different dimensions.
| Dimension | From level 1 to level 5 |
|---|---|
| Value and use cases | Level 1: ad hoc drafting. Level 2: status packs, meeting notes and RAID drafting. Level 3: business case screening, reported status compared with evidenced status, and early warning flags. Level 4: burn and cost-at-completion forecasting, and schedule risk. Level 5: agents validate data, chase owners and draft gate papers, with people approving. |
| Data foundations | Level 1: spreadsheets and email. Level 2: controlled document repositories. Level 3: one taxonomy and identifier set, data quality rules and monthly snapshots. Level 4: enough closed-programme history to back-test, and reference class data. Level 5: event logs and interfaces to source systems, with identities and permissions for agents. |
| Platform and security | Level 1: tools chosen by individuals. Level 2: approved tools in approved environments. Level 3: portfolio, finance and schedule sources integrated, with access by role. Level 4: models hosted with versioning and monitoring. Level 5: an orchestration layer with a kill switch. |
| Governance and the bar | Level 1: none. Level 2: an AI use policy, a register and an accountable owner. Level 3: the full Consolidated AI Bar, with impact assessments where needed. Level 4: model risk management and calibration reporting. Level 5: decision rights enforced in the workflow, and independent review each year. |
| People and operating model | Level 1: enthusiasts. Level 2: trained users who review what AI drafts. Level 3: portfolio analysts own the data spine, with an AI product owner in the office. Level 4: forecasting and data capability, in-house or managed. Level 5: the office organised around exceptions and decisions. |
| Measurement | Level 1: none. Level 2: hours saved per cycle. Level 3: cycle time, traceability and correction rates. Level 4: forecast error, calibration and the lead time of warnings. Level 5: how long decisions take, and portfolio outcomes. |
The Blueprint
What data does an office need to move up a level?
Mostly history it is not yet keeping. Monthly snapshots of portfolio data, kept from now on, are the single step that opens level 4 later.
- To reach level 3: a portfolio taxonomy; identifiers shared by schedule, finance, RAID and benefits records; data quality rules with owners; and monthly snapshots retained from now on.
- To reach level 4: enough closed programmes with consistent snapshots to back-test a forecast. The published study we draw on used 110 projects. Smaller portfolios can pool with reference class data and accept wider ranges.
- To reach level 5: stable interfaces to source systems, an identity and permission model for agents, and event logging.
The Blueprint
What is a Blueprint engagement?
A defined piece of work, planned at eight to twelve weeks, that designs your office's move to the Consolidated level and beyond, with every artefact yours to keep and any provider free to use it.
You receive:
- the target operating model for the AI-enabled office: the bar set centrally, with teams building inside it;
- the data spine design;
- the platform decision, including an exit path from Project Online or Project Server where you need one, made with no referral arrangement with any platform or implementer;
- the full Consolidated AI Bar for your office and a decision rights matrix;
- a governance pack that addresses the Commonwealth policy and ISO/IEC 42001 where they apply to your office;
- a benefits case with baselines, and an implementation plan.
Who does it: an EPMO director, a solution architect, an analyst and a governance lead. We do not implement the portfolio platform; your platform partner does.
In government, this work defines requirements for a later procurement, so we declare our interest in that later work and accept any probity measure you set, including exclusion from it.
Blueprint engagements are taken on one at a time, on request.
Our portfolio platform already has an AI agent. What does this add?
One standard across every tool you run, including the ones your platform cannot see.
Each platform's agent governs itself and works from its own data; the committee needs one set of numbers, one set of criteria and a person who answers for them.
We do not implement portfolio platforms. Your platform partner implements the tool; we set the bar the office holds every tool to, review the cases that go through it, and design how the data joins up across tools.
We have no referral arrangement with any platform or implementer.
If you have just moved off Project Online, which Microsoft retired on 30 September 2026, this is the point to set the bar, before each team configures its own AI in the new tool.
How do you keep a business case review independent?
By the same rules as every independent review. After we review a programme's business case, we take no readiness, recovery, build or operating work on that programme for 24 months, and we never independently review an office or a system we built.
A Maturity Review, a Blueprint engagement and any build are advisory or delivery work, and we say at scoping whether they could lead to our delivery work.
Common Questions
The questions we are asked most
Will AI replace the PMO?
No. It changes the office's work. AI takes over drafting, chasing and reconciling, and the office moves toward exceptions, evidence and decisions. People still own every decision: a written decision rights matrix says what AI may draft, flag or recommend, and what only a person decides.
Which AI tool is best for project management?
The one that fits the tools you already run, held to a standard you set. We do not sell a portfolio tool; we set the bar every tool in the office is held to, and review the cases and outputs that go through them.
What does an AI Business Case Review cost?
Each review is scoped to the case, or the batch, and its fee is agreed before work starts. A screen of the whole pipeline is scoped separately, as a proof of value.
Can you tell whether a business case was written by AI?
The review notes indicators that parts of a case may be AI-drafted, for the reviewer to weigh. It makes no finding that a person used AI. The stronger control is a disclosure rule, so proponents declare AI use in what they submit.
Does this replace a review our framework requires?
No. It is a review we carry out for you against written criteria. It does not replace a review that a framework requires, and it gives no access to one.
How is this different from our Managed EPMO or EPMO service?
It adds AI to the work those services already do: one bar for the office's AI tools, and business cases reviewed against written criteria. Existing clients can add it to their service.
How we work, in writing
Would you know if your office's AI were getting it wrong?
Start with one business case.
The first conversation is confidential, without obligation and without charge.