AI hype to AI value: Escaping the activity trap

At nearly every board meeting now, CIOs are walking leadership through AI progress decks filled with familiar numbers: tools deployed, pilots underway, adoption rates rising quarter after quarter. At the same time, Gartner forecasts that global AI spending will reach $2.52 trillion in 2026, up 44% from the prior year. The investment is accelerating fast. The more important question is whether the value is keeping up?

Even with all the momentum, the results are far less impressive than the activity suggests. CXOTalk research reported in early 2026 that while 88% of companies are using AI in some form, only 6% are seeing clear financial returns. MIT study found that nearly 95% of projects fail to produce measurable results within the first six months. At the same time, pressure is building. The 2026 Kyndryl Readiness Report found that 61% of senior business leaders feel growing pressure to prove that AI is delivering value, while the Teneo CEO and Investor Outlook Survey showed that 53% of investors expect returns within six months.

Taken together, these point to a growing disconnect. Boards are hearing confidence, CFOs are asking for returns and CIOs are often reporting activity instead of impact. The problem is not AI itself. The real problem is what I would call the Activity Trap: the assumption that if AI is being adopted, it must be creating value.

That trap is easy to fall into because activity is much easier to measure than outcomes. Companies count how many AI tools they have purchased, how many pilots are underway and how many licenses are being used, then present those numbers as proof of progress. But more tools do not automatically lead to better business results. More pilots do not mean returns have been achieved. Higher adoption does not, by itself, create value. The board hears momentum, the CFO receives numbers that are difficult to tie to ROI and spending continues without a clear answer to the most important question: what has improved?

3 ways the activity trap shows up

1. The productivity measurement gap

This pattern is already playing out in real boardrooms. Take a large U.S. financial firm that rolled out a major AI productivity platform to 40,000 employees in 2025. Six months later, 78% of licensed users had opened the tool at least once. On paper, that gave the CIO a strong story to tell the board: adoption was high, usage looked healthy and the rollout appeared successful. 

But when the CFO asked a simpler question, the story fell apart: what had the company gained? No one could say how much time had been saved, whether work had become faster or whether costs had come down. The company had tracked usage, but not value. There had been no baseline before deployment, no agreed method for measuring impact and no clear owner responsible for proving results.

So, the issue was not that the technology failed. The issue was that the organization never defined success in business terms before it invested. And that is not unusual. In many companies today, this is exactly how AI is being approached.

2. Pilot purgatory — Where 73% of AI projects stay

The McKinsey 2025 State of AI report suggests that nearly 73% of AI initiatives never make it beyond the pilot stage. The reason is usually not that the technology cannot perform the task. It is that the organization never clearly defined what business success was supposed to look like in the first place.

Too often, pilots are designed to answer a narrow question: Can the tool do this? But that is only half the question. The more important one is: Does this create value for the business? If the pilot is not tied to a business case from the start, there is no real basis for deciding whether it deserves to move into production.

This is how the Activity Trap shows up at the pilot stage. A pilot is considered successful if it ran smoothly, produced output or demonstrated technical capability. But the real outcomes that matter, such as revenue generated, cost avoided, process time reduced or risk lowered, were never defined as success criteria. So, the pilot “works,” yet the business still does not know whether it was worth doing.

3. The board confidence gap

There is a growing gap between confidence and measurement in AI adoption. For example, a recent Logicalis report shows just how wide it is: 94% of surveyed CIOs say they are actively pursuing AI, yet 89% also admit they are still “learning as they go,” and many believe adoption is moving faster than their organizations can properly manage.

And yet, success continues to be reported upward.

That is where the real disconnect begins. The board hears momentum. The organization feels progress. But underneath that confidence, the actual business impact often remains unclear. No one is necessarily being misleading. This is usually not about exaggeration or bad intent. It is a more subtle problem: visible activity starts to look like measurable success.

That is the Activity Trap at the executive level. The more effort an organization puts into displaying new tools, pilots, dashboards and adoption numbers, the easier it becomes to create the impression that AI is working, even when the outcomes have not been clearly defined, measured or proven.

5 questions that expose the activity trap

Before the next AI update goes to the board, it is worth pausing and asking a few harder questions:

1. What value did AI deliver last quarter in real terms? Not projected benefits. Not vendor claims. Not assumed future upside. What changed in the business because of it? Did revenue increase? Did costs fall? Did turnaround times improve? Did errors decline? If those results cannot be shown clearly, then the organization may be reporting motion, not value.
2. What was the baseline before implementation? Every real improvement needs a “before” and an “after.” Without a baseline, even honest progress becomes difficult to prove. The story may sound persuasive, but it remains largely interpretive. A baseline keeps the conversation anchored in evidence.
3. How much effort has gone into measuring outcomes as opposed to simply deploying tools? Deployment is visible. It creates announcements, dashboards and board slides. Measurement is quieter work. It is slower, less glamorous and often postponed. But that is where value is either confirmed or exposed as wishful thinking. If no one is seriously measuring outcomes, the Activity Trap is already in place.
4. How many pilots were deliberately stopped because they failed to deliver? Every serious investment portfolio should include some efforts that were tested and discontinued. If an organization claims that none of its AI pilots failed, that usually does not signal exceptional success. More often, it suggests weak measurement or an unwillingness to shut things down. That is how zombie pilots accumulate projects that remain active on paper but no longer create meaningful value.
5. What is being reported upward? Outcome-based metrics or activity-based metrics? Go back and review the last few board presentations. Were leaders shown business impact, or were they shown rollout statistics, user counts and implementation updates? That pattern reveals more than the slide deck itself. It shows what the organization truly values and what it may still be avoiding.

The escape: Outcome-first AI governance

Getting out of the Activity Trap does not require better AI. It requires better governance.

The first shift is ownership. Every meaningful AI investment should have a business leader accountable for outcomes, not just a technical owner responsible for implementation. Deployment matters, but deployment alone is not the point. Someone on the business side must own the question of whether the investment delivered value.

The second shift is clarity before launch. Success should be defined upfront, not reconstructed later under pressure. That means identifying in advance what the investment is expected to change: revenue, cost, error rates, turnaround time, customer experience or risk exposure. If success cannot be described clearly before deployment, it will be almost impossible to measure honestly afterward.

The third shift is discipline around stopping. Not every pilot deserves to become a program. Organizations need explicit criteria for continuation, scale and termination. Otherwise, they end up with zombie pilots—initiatives that consume budget, remain technically alive and create the appearance of progress without producing meaningful results.

That is where governance maturity really begins: not with launching more pilots, but with assigning clear accountability, measuring what matters and being willing to stop what is not working.

Recent research points to how wide this gap still is. A recent Info-Tech Research Group report found that leaders rate AI governance as highly important, but far fewer believe their organizations are executing it effectively. The companies starting to close that gap are usually the ones that make the shift early from tracking activity to tracking outcomes.

That will likely be the dividing line in this next phase of AI investment. The organizations that succeed will not necessarily be the ones that deploy the most tools. They will be the ones that learn to measure outcomes early, govern AI with discipline and separate real value from visible motion. The ones that remain stuck in the Activity Trap will keep spending through one of the biggest technology investment cycles in recent memory, only to find themselves unable to answer the simplest question when finance asks: what did all this produce?

And that is the deeper lesson. This is not primarily a technology failure. It is a governance failure. It starts with what gets measured, what gets reported and what gets challenged in the next board presentation. If the CFO cannot clearly explain what the AI program is worth, then the organization is not managing value. It is managing activity.

This article is published as part of the Foundry Expert Contributor Network.
Want to join?