The thesis. A working demo proves that an artificial intelligence (AI) model can do a task. It does not prove that the organization will pay for, absorb and sustain the change around it. Most AI pilots stall in that gap, and closing it is a leadership decision rather than a technical one.
The gap is commercial, not technical
The evidence points one way. By some estimates, more than 80 percent of AI projects fail, about twice the failure rate of information technology projects that do not involve AI [1]. Gartner expects at least 30 percent of generative AI projects to be abandoned after proof of concept by the end of 2025, citing poor data quality, inadequate risk controls, escalating costs or unclear business value [2]. Boston Consulting Group (BCG) found that 74 percent of companies have yet to show tangible value from AI, and only 26 percent have built the capabilities to move beyond proofs of concept [3].
Adoption is not the problem. In McKinsey’s early-2024 survey, 65 percent of respondents said their organizations were regularly using generative AI, nearly double the share ten months earlier [4]. Use is broad. Value is narrow.
Look at the causes. RAND interviewed 65 experienced data scientists and engineers and identified five root causes of failure. First on the list: stakeholders misunderstand or miscommunicate the problem AI is meant to solve. The others are a lack of the data needed to train an effective model, a focus on the newest technology rather than real problems, inadequate infrastructure to manage data and deploy models, and problems too difficult for AI to solve [1]. Only the last is purely technical. The rest are decisions about scope, ownership and investment.
What a demo proves, and what it does not
A demo answers a narrow question under friendly conditions: curated data, a patient audience, no integration and no liability. A business case answers a different set of questions. Treating one as evidence for the other is where most stalls begin.
| Question | A demo answers | A business case needs |
|---|---|---|
| Does it work? | On a clean sample | On live data, at volume |
| Who owns it? | The project team | A named business owner |
| What does it change? | One task | The workflow and roles around it |
| What does it cost? | Build effort | Run cost, support and risk controls |
| How do we know? | It looks right | A baseline and an agreed metric |
Every row on the right is a management decision. None can be settled by a better model.
Where the money actually goes
Executives often budget AI as a technology purchase. The evidence says it is mostly an operating change. BCG reports that AI leaders put about 10 percent of their resources into algorithms, 20 percent into technology and data, and 70 percent into people and processes [3]. If a pilot budget is mostly model work, the organization has funded the 10 percent and assumed the 70 percent will happen on its own. It rarely does.
Data is the second hidden cost. Gartner predicts that through 2026, organizations will abandon 60 percent of AI projects that are not supported by AI-ready data [5]. A demo can run on a hand-cleaned extract. Production cannot. The cost of making data fit for purpose belongs in the business case from day one, not in a later phase that never gets approved.
Patience is the third. Deloitte’s survey of 2,773 director- to C-suite-level leaders found that more than two-thirds expected 30 percent or fewer of their generative AI experiments to be fully scaled in the next three to six months. Sixty-nine percent said fully implementing a governance strategy would take more than a year [6]. Leaders who expect a demo to become a product within a quarter will be disappointed, and disappointment ends more pilots than weak models do.
Three trade-offs leadership has to make
Pilots stall when these choices are left implicit. They move when someone with authority makes them explicit.
Breadth versus depth
Running many small pilots feels like progress and spreads risk. It also spreads attention, data work and change capacity thinly. BCG found that AI leaders pursue, on average, only about half as many opportunities as their less advanced peers, and it locates 62 percent of AI’s value in core business functions [3]. The trade-off is fewer bets, each funded to production, against many bets, none funded past the demo.
Speed versus readiness
Moving fast on a vendor demo shortens the time to a first result. It can also commit the organization to a solution before anyone has checked the data, the controls or the exit terms. Moving slowly protects the balance sheet but cedes learning to competitors. The right answer depends on the use case. A customer-facing process with regulatory exposure warrants readiness. An internal drafting tool warrants speed.
Central control versus business ownership
A central AI team brings standards, reuse and risk discipline. A business unit brings process knowledge and the incentive to change. Pilots owned only by the centre tend to produce polished demos that no operating leader asked for. Pilots owned only by the business tend to duplicate effort and skip controls. The workable pattern is shared: the centre owns platforms and guardrails, and the business owns the outcome and the budget line.
What changes when leadership engages
Working where enterprise strategy meets technology commercialization, I see one pattern more than any other: a pilot moves when a senior executive puts a business number next to their own name. Before that, it is an experiment. After that, it is a commitment.
In practice, that means four changes:
- Name the business owner before any build starts, and make the outcome part of that person’s objectives.
- Write the production threshold in advance: the metric, the baseline and the level of improvement that earns rollout funding.
- Budget the full path, including data preparation, integration, training and run costs, not only the model.
- Set a decision date. On that date, the pilot scales, changes scope or stops.
The fourth change matters most. A pilot without a decision date never fails, so it never releases its people and budget for something better.
Questions for the leadership team
- Which of our current pilots has a named business owner with a target in their objectives?
- For each pilot, what result, measured against what baseline, earns production funding?
- What share of our AI budget goes to people and process change, compared with models and tools?
- Is the data each pilot depends on ready for production, and who pays to make it ready?
- When is the next decision date for each pilot, and who makes the call?
References
- J. Ryseff, B. F. De Bruhl and S. J. Newberry, “The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed,” RAND Corporation, Research Report RR-A2680-1, Aug. 2024.
- Gartner, “Gartner predicts 30% of generative AI projects will be abandoned after proof of concept by end of 2025,” press release, Jul. 29, 2024.
- N. de Bellefonds et al., “Where’s the Value in AI?,” Boston Consulting Group, Oct. 2024.
- McKinsey & Company, “The state of AI in early 2024: Gen AI adoption spikes and starts to generate value,” May 2024.
- Gartner, “Lack of AI-ready data puts AI projects at risk,” press release, Feb. 26, 2025.
- Deloitte AI Institute, “The State of Generative AI in the Enterprise: Generating a new future,” Deloitte, Jan. 2025.