Steering many AI initiatives

AI portfolio management means steering many AI initiatives at once: knowing what is running, what it costs, what it delivers, and which initiatives deserve more capacity. A project lead looks at one initiative. A portfolio view looks at all of them together.

Why do AI pilots stall after the first demo?

I once worked with two lab chemists and a development team on an AI tool for image analysis in laboratories. It saved time, and it worked well in practice. It never scaled beyond a single team. The model was good, and so was the usability. What was missing sat around the tool: the organization had not built the conditions for scale.

In an article in March 2026, I described these conditions as three questions:

• Sustainable operations: Does the initiative have a real operating model, with one accountable owner, a named sponsor and protected capacity after launch?
• Proof of value: Can it defend its value when priorities compete?
• Adoption behavior: Are people set up to use it well?

If one of the three is weak, scale becomes difficult.

What does a portfolio view show that a single project cannot?

A problem in one pilot can look like a product problem when its cause sits in the surrounding system: the enterprise infrastructure, communication at large scale, a governance process, unclear ownership or missing trust.

I find it useful to look at three systems across all initiatives:

• Habits shape whether people engage with AI repeatedly and whether local ownership forms.
• Patterns shape whether teams build reusable methods, coherent infrastructure and stronger proof of value.
• Governance shapes whether the system that enables AI is clear, trustworthy and fast enough to create value.

Product and project leads often work inside their own reporting line. Sometimes the better move is to evolve the surrounding system.

What does scaling look like in practice?

Scaling can seem like an easy step after a successful first demo. In my experience, it is a sequence of steps that need decisions and collaboration from several sides: creating training formats, identifying user champions, enabling local experiments, evaluating in a representative way, collecting evidence, broadening demand, and then widening the rollout.