Somebody says it near the end of a management meeting. Usually the CEO, occasionally the COO, and almost always in the same words: I think we need a PMO.
Nobody argues. That is the first problem.
Around the table sit four different pictures of what was just agreed. The CFO heard a reporting function that will finally produce a reliable number for what the company spends on change. Operations heard more project managers. In IT somebody heard a methodology and a tool. Somebody heard headcount; they will be asked to fund it out of a budget they have already committed. The person who said it was describing a feeling — that the organization is busy with too many things and nobody can say which of them are working.
Every one of those is a legitimate version of a PMO. That is exactly why the meeting ended in agreement.
The disagreement does not disappear. It waits. It reappears six months later, when three proposals arrive that nobody can compare, because each supplier answered a different question, or a year later, when the function is running, and half the executive team quietly considers it a disappointment. By then the argument costs money and reputations. In the meeting where it belonged, it would have cost an uncomfortable afternoon.
The hunch is usually right and the brief is usually missing
Chief executives are rarely wrong about this. If you sense that the organization has lost track of its own change agenda, you are almost certainly correct — you are closer to the reporting than anyone, and you are the one who signs off the next initiative without knowing what the last eleven produced.
The hunch does not contain a specification. It is a symptom, accurately felt. And a symptom is not something a supplier can convert into a design on your behalf, because that conversion is not a technical exercise. It is a series of choices about who in your company gets to decide what, and nobody outside the company can make those choices for you.
Which is why this starts as an internal conversation, not a procurement exercise. The value of the conversation is not the document it produces. It is that four people who thought they agreed discover, in a controlled setting, that they did not.
Chair it yourself, and do not let the room be polite
Half a day. You in the chair, or the COO. Your executive team, plus whoever currently runs the largest initiatives — the people who will live inside whatever you build, and who know things the executive layer does not.
Set one rule at the start: this session may conclude that the answer is no. If the room thinks you are asking it to ratify a decision you have already taken, it will ratify it, and you will learn nothing.
Then work through these, in order, and hold the room on each one until it produces an answer somebody would repeat outside the room.
- What decision did we make badly, or too late? A specific one. A date, a piece of work, a consequence.
- Who is this function for, and what do they need from it? Name people, not layers. Their needs will conflict.
- What will it not do? The list you refuse is more revealing than the list you accept.
- Where does it sit — near the strategy, or near the delivery? Check that against the first answer.
- What may it decide without asking? And what may it only recommend, by when?
- What is it allowed to stop?
- Who inside this company owns it in eighteen months?
Seven questions, and the room will move quickly through four of them. The other three are where the session earns its cost.
A complaint is not a decision, and the room will offer you complaints
Ask what decision went badly and you will get atmosphere. We lack visibility. Priorities are unclear. Things take too long here. All true, all unusable, and all comfortable to say because nobody is implicated.
Push for the specific instance. The week two directors were both promised the same engineer. A capital request approved in March that nobody could evaluate, because nobody had ever reviewed the last three approvals. The initiative ran for five months before anyone noticed that its sponsor had left.
Those examples do the design work for you. They tell you whether you have a prioritization problem, an accountability problem, or an information problem — and those need different functions. If the room cannot produce a single concrete example, that is worth knowing before you spend anything. The discomfort may be real and still not yet be a structural problem.
The room will agree on what it does and go silent on what it does not
This is the moment I look for, and it arrives reliably.
Asked what the function should cover, everyone contributes. Reporting, planning support, risk, methodology, portfolio reviews, benefits tracking, resource visibility. The list grows because each item is individually reasonable and nobody wants to be the person who objects.
Asked what it will not do, the room goes quiet.
That silence is the finding. Anything you do not exclude will arrive anyway, sent by whoever has no other place to put it, and you will then judge that function on outcomes it never had the resources to control. A PMO expected to cover everything has not been given a mandate. It has been given a wish list with a reporting line attached.
There is a decent instrument for forcing the choice here, if the room needs help. PMI’s current guidance — reworked in 2025 around the idea that a PMO is a service provider shaped by the needs of the people it serves — catalogues the services such a function commonly provides, grouped by whether they support strategy, capability or delivery. Print it and make the team select. The argument that breaks out over which items to cross off is the entire point of the session, and it is far better held in front of you than in front of a supplier.
Nobody will volunteer an answer to what it can stop
The authority question is the one your team will try to defer, and they will defer it in good faith. It sounds premature. Let us define the scope first and work out the governance later.
Later never comes. It cannot, because by then the answers have winners and losers.
Ask it plainly. When two functions want the same resource in the same week, who settles it? When an initiative is failing, who can halt it, and does that person need anyone’s agreement? And when a functional director declines to cooperate, what actually happens next?
If the honest answer to all three is you, personally, then what you are building is a reporting desk that will forward those decisions back to your desk, and you will have added a step rather than removed one. That is a legitimate choice — but make it knowingly, and do not later blame the function for having no grip.
I write this into every charter as a table: one row per decision area, each marked as decided alone, recommended with a deadline for the board’s answer, co-decided with a named person, escalated, or reserved to the owners entirely. The reserved rows matter as much as the rest, because naming what is excluded is what makes the rest credible. It is the section most templates leave out, and it is the reason the charter belongs before the appointment rather than after it.
Name the internal owner in the room, while everyone can hear it
The last question is the one most often answered with a job title that does not yet exist.
Someone has to hold this when the external help leaves, so name that person while the executive team is in the room and can react. If nobody in the room is willing to own it, you are not building a capability. You are renting one, and the arrangement will end when the invoice does.
The answers turn a tender into a specification
Everything above has a commercial consequence, and it is the reason the afternoon pays for itself.
A request that asks the market to design and implement a PMO invites suppliers to bring their own process. It will be a competent process, refined across other people’s organizations, and adapting yours to it is a substantial share of what you will pay for. You will also be unable to compare the responses, because each will have answered a different question.
A request built from the answers above reads differently. This is what the function covers, this is what it explicitly does not, these are the decisions it takes without asking, this is who sponsors it, this is who owns it by the end of next year. Suppliers now quote for your design. And you know what you are shopping for, which is execution rather than another round of design.
Somebody has to say out loud that it worked
One last question before the room disperses, and it is not a formality.
These functions are rarely closed because they delivered nothing. They are closed because, eighteen months on, nobody senior can describe what they changed. Recognition does not follow good work automatically; somebody has to arrange it.
So decide it now. Who will state, a year from today, what this changed — to whom, and on what evidence — while nobody yet has a result to defend?
The hunch is worth acting on. Make sure the answer is yours before the market puts a price on it.