Most enterprises no longer struggle to start AI projects. They struggle to finish them. The evidence is consistent across research firms, and it points away from the model and toward everything around it.
What the research says
- IDC, in research with Lenovo, found that 88% of observed AI proofs of concept did not make the cut to widescale deployment.[1]
- S&P Global Market Intelligence found the share of companies abandoning most of their AI initiatives jumped to 42%, up from 17% the year before.[2]
- Gartner predicted in 2024 that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025, citing poor data quality, inadequate risk controls, escalating costs or unclear business value.[3]
- Gartner has since predicted that over 40% of agentic AI projects will be canceled by the end of 2027.[4]
- BCG found that only 5% of companies globally qualify as “future-built” for AI.[5]
A note on figures: these are the publishers' own, from surveys and predictions with their own methods. We have left out one widely repeated pilot-failure figure because the underlying report measured something narrower than the headline, and its data was preliminary and self-reported.
Four reasons the research keeps finding
1. The data is not ready
Poor data quality is the first reason Gartner gives for abandoned projects, and IDC points to low organizational readiness in data, processes and IT infrastructure.[3][1] A model that worked on a clean sample meets data spread across systems that disagree.
2. Security, privacy and risk controls
Inadequate risk controls appear in Gartner's predictions, and S&P respondents named data privacy and security among the top obstacles.[3][2] A plan that assumes access on day one loses weeks to accounts, network rules and review.
3. Cost and unclear value
Escalating costs and unclear business value are the other half of Gartner's list.[3] Projects started without an agreed measure of success are the easiest to cancel.
4. The work never reaches the people who should use it
BCG's finding that only a small share of companies are built to get value from AI at scale reflects how much of the work is organizational, not technical.[5] Systems that do not fit how people actually work are not used.
Where forward deployed engineers fit
Each of those four problems sits between the model and the business, which is exactly where forward deployed engineers work: inside the customer's environment, on its real data, through its security process, and alongside the people who will use the result. That is why AI labs, cloud providers and software companies have all built FDE teams in the last two years.
If you are hiring for this work, start with how to hire a forward deployed engineer.
Sources
- CIO.com, on IDC and Lenovo research, 25 Mar 2025. www.cio.com
- CIO Dive, on S&P Global Market Intelligence research, 14 Mar 2025. www.ciodive.com
- Gartner prediction of 29 Jul 2024, reprinted by Intelligent CIO. www.intelligentcio.com
- Gartner prediction of 25 Jun 2025, via IHL Group. www.ihlservices.com
- BCG press release, 30 Sep 2025. www.bcg.com