Article
Most AI projects do not fail for technical reasons
Companies are spending heavily on AI, and a large share of that spending is being written off. The reason is almost never the technology.
In 2024, Gartner predicted that at least 30 percent of generative AI projects would be abandoned after the proof-of-concept stage by the end of 2025. That forecast turned out to be conservative. Gartner’s own later reporting put the actual figure at roughly half of generative AI projects abandoned after proof of concept, and a 2025 MIT study found that only about 5 percent of AI pilots produced real business value. The reasons Gartner named held throughout: poor data quality, inadequate risk controls, escalating costs, and unclear business value. Not one of those is a failure of the model. They are failures of data, governance, cost discipline, and business judgment.
The same pattern shows up in governance. Gartner also predicts that 80 percent of data and analytics governance initiatives will fail by 2027, because too many of them are run as a reactive, data-only exercise disconnected from the business outcomes they are supposed to serve. As Gartner’s analysts put it, a governance program that does not enable prioritized business outcomes fails.
Put those two predictions together and the message is hard to miss. The expensive failures in AI and data are leadership and organizational problems wearing a technology costume.
Why it keeps happening
The pattern is familiar. AI strategy gets built in a room that does not include the people who understand the data, the systems, the regulatory exposure, and what it would actually take to put a model into production. The model itself is the easy part. Whether the underlying data is usable, whether there is a governance process for when the model is wrong, whether the use case has a defined and measurable payoff, those are decisions someone senior has to own before the money is spent, not after.
When no one owns them, the project does what Gartner describes. It clears a proof of concept, runs into the data and the cost and the unclear value, and gets quietly shut down.
What a seasoned technology leader changes
Experience changes the order of operations. A leader who has done this before picks the use case with a defined business payoff instead of the most impressive demo. They know whether the data foundation can support the work before a budget is committed, because they have built data layers and seen what unready data does to a project. They own governance directly: the policy, the review, the escalation, and the accountability for when a model makes a bad call, which is not optional in a regulated industry. And they tie the work to a business outcome, which is the exact thing Gartner says governance programs fail without.
None of that requires a larger AI budget or a better model. It requires someone senior in the room when the strategy is set, holding the work to the standard that separates the projects that reach production from the ones abandoned after proof of concept.
The practical version
You do not need to hire a full-time chief information or technology officer to get that judgment. For most mid-market companies, the work of owning AI and data decisions does not yet fill a full-time executive seat. It needs experience applied at the decision points: which use case to build, whether the data foundation can support it, and who owns the governance once it is live. That is the kind of work a fractional CIO or CTO does, and it is usually the difference between AI that earns its keep and AI that becomes another abandoned proof of concept.
If your organization is making decisions about AI investment and wants an honest read on whether the foundation is ready, that is the conversation to have before the investment, not after.
Written by Jon McAnnis, Principal Advisor at Groundwork Technology Advisors.