John McIntyre
Founder, HotPMO
Every month I trawl through the research papers, reports, and podcasts so you don’t have to. Here’s what caught my eye this month.
There’s a learning thread running through April. Six decades of Olympic Games data asking whether projects ever learn from each other, a PwC study on why most companies still haven’t learned how to get value from AI, an essay on where all that individual AI productivity actually goes, and a Financial Times piece arguing that the coordination work PMOs have always done is becoming the most valuable skill in the building. Plus a packed events list for May.
Do the Olympics ever learn?
Atif Ansar, Bent Flyvbjerg and Alexander Budzier have published a big data analysis in Public Management Review asking whether projects learn across space and time. Their answer, based on Olympic cost overruns from 1960 to 2024: mostly no. Every Games with reliable data went over budget, with no improvement across six decades. Their explanation gives you your word of the week: spatiotemporality. I challenge you to drop that into your Friday PMO report. Each Games is run by a brand-new temporary organisation, in a new city, four years after the last one, and all that distance in space and time blocks the higher-level learning.
The conclusions don’t quite land for me. Cost overrun is one metric, and I’m not convinced it’s the right one for judging learning. Delivering an Olympics involves an enormous amount of novelty: stadiums and event spaces on a scale most host cities have never attempted, and architecture that’s deliberately experimental because every host wants to make a splash. Add the political habit of lowballing a bid to get it approved, and I’d question whether the overruns correlate more with novelty than with the distance between Games.
Two things the headline hides. First, the dataset only covers cities that hosted. Cities are also learning by walking away: Brisbane took the 2032 Games unopposed because nobody else bid. A decision to walk away from a project is a lesson learned too, and we shouldn’t lose sight of that when we count lessons. Second, Paris looks like an example of learning from the past. They borrowed, they reused, they were cautious, and while the final numbers are still being crunched, their overrun sits below the historical average.
Paris 2024: on the paper’s own estimates, an overrun below the historical average. Perhaps somebody did learn something.
To be fair to the authors, even the abstract concedes there’s plenty of lower-level learning: accessibility, crowd management and anti-doping have all improved markedly from Games to Games. The lessons are being learned; they just don’t bubble up to the bottom line. The paper is free to read and worth your time, even if you end up arguing with it like I did. And a nice connection for regular viewers: Alexander Budzier has been a guest on the PMO HotHouse before, and he’s keynoting at the PMO Conference in London this June.
The question it left me with: if someone judged your portfolio’s learning purely on cost variance, what would they conclude, and would they be right? Check that your lessons process can show the learning a single metric would miss, including the projects you decided never to run.
Lots of AI pilots, not much payback
PwC’s 2026 AI performance study, Decoding ROI from AI, surveyed 1,217 executives across 25 sectors and found that around three quarters of AI’s economic gains are being captured by roughly a fifth of companies. Everyone else is running pilots.
The patterns will feel familiar. Tools thrown out to see what sticks, with weak links to revenue, cost savings or better decisions. A conversion problem, where something works for one team and never becomes scalable or reusable for the wider organisation. And an efficiency fixation: the companies seeing the least return are using AI to shave time off existing work, while the leaders use it for growth, new markets and new capabilities.
All of this is bread and butter for a PMO, yet plenty of organisations are chucking AI tools around like Smarties with no portfolio discipline in sight. Put every AI experiment into one programme or portfolio view, each initiative mapped to a business objective with a measurable outcome. Run an explicit scale-or-stop cadence, because too many experiments get tried once and left festering in a corner when the decision should be to kill the broken ones and double down where value is showing. Then build the foundations the study points to: investment discipline, skill building, and safety and governance. Responsible AI policy belongs at organisational level, and where it doesn’t exist yet I’ve seen PMOs picking it up, because their projects are the ones paving the way.
A check to run this week: list your organisation’s AI experiments and see whether each one has a named objective and a measurable outcome. If the answer is no, your portfolio has found its next job. On last month’s show we talked about a PMO chosen to lead its organisation’s AI implementation precisely because it sits across everything; this study reads like the business case for that decision.
Everyone’s 10x more productive. So where did it go?
George Sivulka’s essay Institutional AI vs Individual AI opens with a puzzle: AI just made every individual ten times more productive, yet no company became ten times more valuable as a result. So where did the productivity go? Strictly speaking it’s a March piece, but I only saw it in April, and I’m in charge of the news, so it counts.
His argument is that tools which speed up individuals don’t speed up institutions. Everyone develops their own AI habits and their own confidently held, AI-assisted opinions, and the models are agreeable enough to tell each of them they’re absolutely right. That’s an echo chamber. Across an organisation it entrenches people in different directions instead of aligning them, which makes the whole slower even as every part gets faster.
We ran into exactly this at Monday’s PMO meetup, before I’d read the essay. We were asking what’s accelerating delivery in 2026 and what’s holding it back, and the observation from the room was that individuals are quicker and software teams are shipping more releases, yet portfolios show no capacity uplift. Generating documentation is faster than ever; somebody still has to read it, analyse it and take accountability for it. The work survives, it just moves up a level, and it lands on the people who were already the bottleneck.
Take this question into your next steering meeting: is AI making your individuals more productive, or your portfolio more productive? If it’s only the first, you’re paying for speed you can’t bank.
“Glue work”: the PMO’s time to shine?
Sarah O’Connor’s column in the Financial Times picks up “glue work”, the term coined by engineer Tanya Reilly in her 2019 talk Being Glue: the coordinating, aligning, relationship-building and misalignment-fixing that holds projects together. Reilly’s warning was that glue work is expected when you’re senior and career-risky when you’re junior, because the person who gets promoted is the one who wrote the great code after the glue person unblocked everything. Regular viewers will recognise our old friend the boundary spanner.
What’s changed is who writes the code. With engineers increasingly supervising AI tools that do the typing, O’Connor argues the support work is becoming the core work. What’s left is managing the agents, documenting the wider context, coordinating, and seeing the customer’s needs and the bigger picture. She quotes Brittany Ellich, a staff engineer at GitHub: “Now, those are the most important skills. In fact, those are the skills.”
You can see where I’m going. That list is a decent job description for a PMO. Software engineering is pivoting towards the thing we’ve done for organisations all along, and as AI spreads into other professions I’d expect the same revaluation to follow. Our moment to shine, possibly.
Two questions to ask about your own team this month: who is doing the glue work, and are they getting credit and progression for it? If the answer makes you wince, fix it before AI makes the gap obvious.
May: worth leaving your desk for
Fancy a free ticket to the PMO Conference? House of PMO’s AI Movie Poster Challenge runs until 8 May: create an AI-generated, PMO-inspired movie poster and share it on LinkedIn with #PMOMovieChallenge.
6 May: PMO at Sea, with the Roke PMO team aboard a Red Funnel ferry out of Southampton. Yes, an actual boat.
7 May: Digital AI Analytics in Action for the PMO at AtkinsRéalis in Edinburgh.
12 May: International PMO Day, marked with an in-person London evening followed by networking drinks.
14–15 May: the new Practical AI Skills for the PMO course runs as a virtual classroom. It was already close to full when we recorded, so move quickly.
17–19 June: PMO Conference London, with PMOLearn on the 17th, the conference itself on the 18th (Alexander Budzier keynoting), and Laura Barnard’s IMPACT Engine Fundamentals workshop on the 19th.
A quick nod as well to Project Forum, House of PMO’s new free event for the whole project community, which had its first outing at the Business Design Centre on 21 April. Watch for it returning next year.
This roundup is based on my monthly PMO News segment from the PMO HotHouse, run by the House of PMO. If you’re not already part of that community, it’s well worth a look.
If April’s news is right and coordination is becoming the core work, it pays to get better at it on purpose. That’s exactly what we built the PMO SuccessHub for: practical support for PMO teams, month in, month out. Come and take a look.
