Think about how your capital committee handles a big request. Operations wants a new production line. Sales wants the CRM rebuilt. Facilities wants more warehouse space. No one in that room would accept a process where each request is approved on its own merits and the total is simply whatever the requests add up to. You would ask what each one returns, over what horizon, at what risk, and what you are giving up to fund it. Then you would fund some, defer some, and say no to a few.
Now look at how many organizations built their AI budget for this year. A Copilot rollout. A content platform for marketing. An AI-assisted service pilot. A handful of departmental tools that came in on a corporate card. A proof of concept a vendor offered to run at a discount. Each one was justified individually, and together they became a number.
That is a wish list. You would never run your capital expenditure as a wish list, so it is worth asking why so many of us are running AI that way.
A wish list starts with what people want. Every item arrives with its own champion and its own business case, and most of those cases hold up fine when you read them one at a time. The trouble is that nobody reads them together. The approval question becomes "is this reasonable?" when the question that matters is "is this the best use of this dollar?"
A portfolio starts from the other end. It begins with a view of where the business needs to go, then allocates capital deliberately across risk, return and time horizon. It is managed as a whole rather than line by line. It concentrates on the few things that will actually matter. And it gets pruned, regularly and on purpose.
Same vendors, same price lists, same technology. What changes is who is asking the questions and what they are optimizing for. A wish list optimizes for coverage (every department gets something). A portfolio optimizes for the return on the capital deployed.
Here is the uncomfortable part. A lot of AI spend underperforms for reasons that have very little to do with AI. It underperforms because it is governed as an IT purchasing exercise when it is really a capital allocation decision.
IT can evaluate tools, negotiate licenses and run a secure deployment, and a good IT team does all of that well. What IT cannot do (and shouldn't be asked to do) is force trade-offs across business units. Telling a division head that their pilot is being stopped so another team can fund the one initiative that matters is an allocation call. So is deciding the company will accept a loss on two experiments to find the third that works. Those calls belong with the CEO and the CFO, and ultimately with the board.
The framing changes the questions. When AI is "a technology program we fund," the conversation stays on tools, seats and vendors. When AI is "a set of investments we manage for return," the conversation moves to the thesis, the expected return, the exit criteria and what we are choosing not to fund as a result. Boards are already starting to ask the second set of questions. They deserve a portfolio answer.
The cost of skipping this step is visible in the data. In its 2025 Voice of the Enterprise survey, S&P Global Market Intelligence found that 42% of companies had abandoned most of their AI initiatives, up from 17% the year before (CIO Dive coverage). That number usually gets cited as proof that AI is overhyped. To me it reads more like a capital allocation failure: many positions opened, few with a thesis, no exit criteria set in advance, and eventually a write-off that gets described as a change in strategy.
Over the past month we have been working through two pieces of this. The first was the real cost of AI, which runs well beyond the license: integration, data preparation, consumption, governance, training, change management, and the internal time it takes to actually change how work gets done. The second was the discipline to stop funding projects that aren't earning their keep.
Portfolio thinking is what connects them, and it is why both belong on the executive agenda. If you don't know the full cost of each position, you can't judge its return. If nothing ever gets stopped, what you have is a collection that only grows.
Budget season is where this becomes real. Most of us are setting 2027 numbers right now, and for a lot of organizations 2027 will be a reset year. The accumulated pilots, overlapping tools and quietly renewed licenses will get looked at together, many for the first time. The open question is whether that reset happens on your terms, as a deliberate rebalancing, or gets forced on you after another year of spend that is hard to defend.
The companies that enter 2027 treating AI as a managed portfolio are the ones that will reinvent. They will know what they own, what it returns and where the next dollar should go. The companies still working off a wish list will spend the year resetting, unwinding commitments nobody quite remembers making.
If you are working through your 2027 numbers now, our 2027 AI Budget Planning Guide is a practical place to start. Put your current AI spend in front of your leadership team and ask a simple question of every line: would we fund this again today, knowing what we know? And if it would help to think it through with someone, I am always glad to have that conversation.