Enterprise AI is creating a new management problem. Companies can start AI projects faster than they can decide which ones deserve to keep going.
AI budgets are growing, but the stronger signal is how companies are beginning to narrow their focus. BCG’s latest IT Spending Pulse, based on a May 2026 survey, found that IT buyers expect 2026 budgets to grow 5.8% year over year, with 66% of respondents expecting to increase spending on AI and machine learning, including GenAI and AI agents. BCG also found that buyers are narrowing where they apply GenAI and agents, concentrating on fewer high-impact use cases as measured ROI rises.
That is the management problem. The number of possible AI initiatives keeps expanding, but the resources that make them work do not expand at the same pace: technical talent, domain expertise, integration capacity and management attention.
XBP Global CEO Andrej Jonovic believes that creates a different test for leadership. An AI-first organization needs to know when an initiative has earned more resources and when the result no longer justifies continuing.
Starting an experiment is becoming easier. Keeping one alive should require evidence.
Starting is cheaper. Continuing should have a higher bar.
A technically successful AI project can still be a poor use of resources.
The model may work while the cost base does not. An automation may produce an acceptable result and still require too much manual review. A use case can solve a real problem without solving one important enough to justify the integration effort.
Those distinctions become harder to make when the organization has pressure to show AI activity everywhere.
Jonovic tends to bring the conversation back to what happens after the demonstration. What resource does the initiative consume? What changes when it reaches volume? Does the quality hold? Does the economics make sense inside the operating environment where the system has to run?
The answers should determine whether the company expands the work, changes it or stops.
There is already evidence that many projects eventually reach that decision. In January 2026, Gartner reported that at least half of generative AI projects had been abandoned after proof of concept by the end of 2025. Gartner cited poor data quality, inadequate risk controls, escalating costs and unclear business value among the reasons.
Some of those projects probably needed to end.
For Jonovic, stopping an initiative does not automatically mean the experiment failed. A company can learn that the economics are wrong, the data is not ready or the problem is better solved another way. Continuing to consume resources after that point creates the larger problem.
The useful discipline is deciding what evidence the project needs before the next stage of investment.
AI projects should earn the next dollar of investment
Jonovic looks for the point where AI begins changing the economics of the operation.
XBP Global has been applying AI-led automation across its own organization while converting legacy workflow platforms into more intelligent pipelines. In its second-quarter 2026 results, the company raised its expected annualized operational efficiencies from AI-led automation to $65 million to $75 million, with approximately $35 million expected to be realized during calendar year 2026.
Those numbers give management something concrete to follow.
When an initiative removes repeatable effort, changes the cost base or allows the same resources to handle substantially more work, the benefit can continue after the initial deployment. Jonovic has described those automation-driven efficiencies as compounding over time.
That is the kind of result that can justify additional investment.
Other AI projects will produce weaker economics. Some will become too expensive when usage increases. Others will depend on integration work that outweighs the benefit. A system may perform well technically while creating new work somewhere else in the process.
Leadership has to follow the result far enough to see that.
The initial use case should therefore include the conditions for continuing it. Leaders need a view of the cost, resource requirement, expected operating benefit and level of performance the initiative has to reach. Once the company has evidence, the next investment decision becomes easier to make.
More AI creates a resource-allocation problem
The number of available use cases can become misleading.
A company may identify dozens of places where AI could technically be applied. Each one creates a claim on people, infrastructure and management attention.
Research from Boston Consulting Group found that companies generating stronger AI results tend to concentrate their efforts. Leading companies prioritized an average of 3.5 use cases, compared with 6.1 among other companies, and expected 2.1 times greater ROI from their AI initiatives.
For Jonovic, that concentration makes sense.
Every organization has a finite amount of domain expertise. The people who understand a healthcare process, financial workflow or mission-critical client environment well enough to help redesign it are valuable. So are the architects and technical teams capable of putting the system into production.
Spreading those resources across too many initiatives can slow down the work that has the strongest case for scaling.
As AI tools become more capable, the temptation will be to add more projects. Management discipline may require the opposite response in some areas.
The company needs to learn quickly enough to determine which experiments deserve deeper investment, which need to change and which should release their resources back to the organization.
That skill becomes more important as the cost of experimentation falls.
Enterprise AI has spent the last few years expanding what companies can try. The next phase will put more pressure on leaders to decide what deserves to survive.
The companies that develop that discipline can put more of their people, capital and attention behind the AI initiatives that materially improve the business.
