The meeting with three versions of the truth
Last year, one of our consultants sat in a steering committee where one program appeared with three different budget numbers. The slide said 1.8 million spent. The finance pack said 2.1 million. The program manager's own tracker said 1.6 million. They spent twenty minutes debating which number was real and zero minutes making the decision that the meeting was supposed to make. The sponsor closed her laptop and said something I have repeated to every PMO team since: "Call me when you agree with yourselves."
That portfolio was not short on tools. It had dashboards, registers, trackers, and a reporting deck that took days to assemble. What it did not have was a single number anyone was willing to defend. And that same portfolio, twelve months later, was being pitched an AI layer to "accelerate insights."
Every 2026 portfolio management trend report leads with AI. Predictive capacity planning, scenario modelling in seconds, automated status synthesis. I use these capabilities, and I want PMO leaders to use them. But the readiness conversation almost always skips the unglamorous part: AI does not fix bad portfolio data. It distributes it faster, with more confidence, to more executives.
Why AI makes bad data more dangerous, not less
A human analyst who doubts a number hesitates. She cross-checks, asks the PM, and softens the claim. An AI assistant reading a stale risk register does none of that. It produces a fluent, confident summary of information that is no longer true as of March. The polish hides the rot. When a wrong number arrives in bad formatting, people question it. When it arrives in a well-written executive summary, they act on it.
I saw this play out at a client that piloted an AI status assistant before touching its data. The pilot produced weekly portfolio summaries that read beautifully and quoted a risk exposure figure that had been superseded two months earlier. Nobody caught it for three cycles, because the writing was too good to question. The pilot was paused, not because the AI failed, but because the portfolio fed it fiction.
So the sequencing matters. Automate a portfolio with trusted data and you multiply its value. Automate a portfolio with conflicting data and you industrialize the confusion. The PMOs winning with AI in 2026 are not the ones with the best prompts. They are the ones whose data was worth reading in the first place.
Is your portfolio data ready for AI?
Here is the test I give every PMO leader who asks. Open last month's portfolio review. Pick any three numbers: a budget figure, a forecast, a status date. Now trace each one back to its source in under five minutes per number. Where does it live? Who owns it? When was it last updated by someone accountable for it?
If you cannot do that, your portfolio is not AI-ready, no matter what tool you buy. The good news is that fixing it is not a two-year data program. It is five disciplines, applied with some stubbornness.
The five disciplines of AI-ready portfolio data
1. One system of record per data type. Budget numbers live in exactly one place. Status lives in one place. Risks live in one place. Every report, deck, and dashboard references the record; nothing re-keys it. The moment a number gets retyped into a slide, you have created a second version of the truth, and second versions always drift.
2. Define every field once. In the three-budgets portfolio, "forecast" meant four different things depending on which team you asked. Write a one-page data dictionary: field name, definition, owner, update cadence. One page. Argue about it once, in a room, then stop arguing about it forever.
3. Collect at the source, automatically. Manual re-keying is where truth goes to die. Intake through forms, not emails. Status through structured updates, not paragraphs pasted into decks. Integrations move numbers between systems so a human never retypes them at midnight before the steering committee.
4. Give every number an owner and a date. Unowned data rots on a predictable schedule. Every key field shows who owns it and when it last changed, visibly, on the view executives actually read. A number with a stale date stops being a fact and starts being a question, which is exactly what it should be.
5. Audit before you automate. Once a quarter, run a data trust check on your top projects: trace the key numbers, score each field red, yellow, or green, and fix the reds before you wire any AI to them.
A word on effort, because this is where most leaders overestimate the mountain. In the three-budgets portfolio, the full cleanup took six weeks, and most of that was conversation, not configuration. Deciding which system owns the budget number took one meeting and some political capital. The data dictionary took two workshops. The retype ban took a standing rule and a month of enforcement. None of it required a data engineering team or a transformation budget. It required a PMO leader willing to make five decisions and hold them. The resistance you will meet is real but shallow: teams defend their private trackers for about three weeks, until the first month they do not have to assemble a report by hand. After that, nobody volunteers to go back.
Where Smartsheet fits
This is the work Smartsheet was built for, which is why we lead most data cleanups with it. One sheet becomes the system of record per data type. Forms handle intake so requests arrive structured on day one. Automated workflows chase owners when a field goes stale, so freshness stops depending on my nagging. Control Center provisions every new project from the same template, which means every project is born speaking the same data language, and portfolio rollups actually mean something. Dashboards read live cells, not pasted screenshots. Our clients typically see about a 50 percent increase in portfolio visibility from this work alone, before any AI is involved. Then, when you do add AI on top, it is reading a portfolio that tells the truth.
Your move next week
Pick your top ten projects. For each one, trace three numbers: total budget, current forecast, and the date of the last status update. Score every number green (traced to a governed source in under five minutes), yellow (traced, but with detours), or red (could not trace it or found conflicts). Thirty numbers, one afternoon. The reds are your real AI readiness plan, and fixing one red field per week will do more for your portfolio than any tool purchase this year.
If you want a second set of eyes on it, this is exactly what the Assess step of our Vision2Value Framework does. The PMO Value Blueprint is a focused 4 to 8 week engagement that maps where your portfolio data, governance, and delivery stand today, often including a Smartsheet pilot on your real data. Start with a conversation at pmoevolution.com/smartsheetsolutions or book 30 minutes at calendly.com/pmoevolution.
Click the image to book a call
Upcoming Live Session
Most PMO training tells you what a value story should contain. On Sept 16, you will actually build one.
The 30-Day Answer is a free two-hour live session from Bruno Freitas and Dr. Tony Prensa, built around one scenario: Helix Renewables, where the CFO has given the PMO 30 days to prove it adds value to the organization.
Working in small breakout groups with other PMO leaders, you take that company from flagged for defunding to a defensible rescue plan, using the PMO flywheel from PMI's PMO Practice Guide.
You leave with the plan you built, the thinking behind it, and a much clearer picture of what your own PMO's answer would need to look like.
