Marco van Nieuwenhuijzen is a C-level technology executive with over 25 years of progressive leadership across Fortune 500 organisations in the pharmaceutical, consumer goods, and media sectors. He currently serves as VP and Chief Information Officer of a pan-European pharmaceutical wholesale business operating across 9 countries, where he leads large scale. technology transformation programmes. Marco is a Fellow of the British Computer Society (FBCS) and a Chartered IT Professional (CITP), the UK’s most selective professional designations for IT practitioners, awarded by peer assessment to individuals of demonstrated outstanding achievement.
The productivity gains from AI inside transformation programmes are real and significant. But adoption moves faster than governance, and closing that gap is now one of the more important things a CIO can do.
I’m a strong believer in AI inside large-scale transformation programmes. I’ve been applying it across my own programmes for some time now, I actively encourage my teams to use it, and I think practitioners who aren’t yet engaging with it seriously are missing a genuine productivity opportunity. That position hasn’t changed.
What has changed is my understanding of what responsible promotion looks like. Encouraging adoption without putting the right governance in place first, or at least in parallel, creates a different kind of problem. Not a catastrophic one, but a real one. And it’s one I’m actively course-correcting on, which is what prompted me to write this.
I’m sharing it as a practitioner, not as someone with a finished answer. The organisations getting this right are building as they go. So are we.
What AI is doing inside our programmes
The applications are broader than most external commentary suggests. We’re using AI to evaluate and score software packages against functional requirements, producing structured comparisons that previously took days. We’re generating effort estimates from requirements sets and target solution architectures. We’re using it for meeting minutes, communications, presentation drafts, and workshop synthesis. More recently we’ve been applying it to requirements documentation, process mapping, and quality engineering.
The productivity gains are genuine. Outputs that took days take hours. The cumulative effect across a large programme is significant. I’ve been vocal about that internally and with peers, and I stand by it.
What I underestimated was how quickly teams run ahead of any structure when a senior leader is actively promoting adoption. That’s not a criticism of the teams. It’s on me for not building the infrastructure before pushing the accelerator.
The governance gap, and how it surfaces
An effort estimate produced by AI looks identical to one built through rigorous bottom-up analysis. A vendor scoring document generated in twenty minutes carries the same visual authority as one that took a week. The output format is standardised by the tool. The quality of the reasoning behind it is not, and without governance, that distinction is invisible until something lands on your desk that doesn’t quite add up.
Teams encouraged to be creative will be creative in ways you didn’t anticipate. Outputs enter programme workflows before anyone has defined what “good” looks like, who is accountable for the review, or what level of scrutiny applies to which type of output. The AI is producing confident-looking documents. The governance to validate them hasn’t been built yet.
A 2026 Deloitte survey of more than 3,000 senior leaders found that only one in five organisations has a mature governance model for autonomous AI agents. Most are experiencing exactly this: adoption ahead of the infrastructure to make it trustworthy.
What’s working, and I would recommend
The highest return action is standardising outputs before promoting adoption. Define what a good AI-assisted output looks like for each type: a requirements document, an effort estimate, a vendor evaluation, and quality assurance becomes a comparison exercise rather than a judgment call. This single step removes most of the ambiguity that allows poor outputs to circulate undetected.
Classify outputs by risk, not by tool. Meeting minutes and presentation drafts are low risk. Effort estimates feeding a board-level business case are high-risk. Define those categories explicitly and apply proportionate review standards. Make it clear that a named human is accountable for any output that informs a decision, AI generates, a human decides, and those two things should never be mixed.
Require disclosure: if an output is AI-generated or AI-assisted, say so on the document. Not as a disclaimer, as a signal that specific scrutiny is needed.
On training; invest in teaching teams how to evaluate AI outputs, not just how to generate them. The skill gap in most programmes right now isn’t in production, it’s in knowing whether to trust what’s been produced. This is a different kind of training from anything most organisations have run before, and it’s worth investing in properly.
Beyond training, it’s worth considering whether you need dedicated AI practitioners embedded in your programme team, people whose job is to guide the correct use of AI across workstreams, maintain quality standards, and stay ahead of what the tools can and can’t do. If building that capability internally isn’t yet feasible, specialist external support is a credible alternative. The organisations getting this right tend to have someone who owns it.
Lead it yourself
The most important thing I’d say to any practitioner in this position is this: learn how to use AI well yourself, before asking your teams to. Not at a surface level but genuinely engage with it, apply it to your own work, understand where it produces good outputs and where it doesn’t. That experience is what allows you to lead credibly, set realistic expectations, and make sound judgments about where governance needs to be tightest.
The CIOs and programme leaders who will get the most from AI are not the ones who delegate the learning. They’re the ones who stay close enough to the technology to guide its use with genuine authority.
Promote adoption. Build the infrastructure. Lead from the front. And treat the governance not as a constraint on AI use, but as what makes AI use sustainable.