# 89% of AI pilots never reach production > The reason is almost never the model. It is five conditions around it, and all five can be diagnosed before you sign anything. [Platform](https://www.shimliapp.com/en/blog/category/plataforma) ## 89% of AI pilots never reach production The reason is almost never the model. It is five conditions around it, and all five can be diagnosed before you sign anything. **Luis Aguilar** September 15, 2026 4 min read IN SHORT - Deloitte puts the share of AI agent pilots that never reach production at 89%; independent 2026 studies place it between 86% and 88%. - 78% of companies have at least one active pilot, but only 14% have scaled one across the organisation. - Gartner estimates that over 40% of agentic projects are at risk of cancellation before 2027. - The cause is rarely the model: it is data access, project ownership, continuous evaluation and cost control. There is a conversation that repeats in committees this year. Someone presents an AI agent pilot that went well, the team approves continuing, and twelve months later the project is still a pilot. It is not bad luck or bad execution. It is a pattern, and it has been measured. ### The size of the problem | Deloitte tech trends 2026 | 89% | | --- | --- | | Independent studies range reported in 2026 | 86% to 88% | Only the two sources measuring the same thing. Gartner separately estimates that over 40% of agentic projects risk cancellation before 2027, but that measures something else and putting it on the same bar would make them look comparable. Deloitte Tech Trends 2026 · independent studies compiled in 2026 The figure that best frames the conversation is another one: 78% of companies have at least one pilot running, and only 14% have scaled one across the organisation. Almost the entire market sits at the same halfway point. ### What fails is not the model It is worth saying plainly because it saves months: **the same model that runs in the pilot runs in production.** If the pilot worked, the model is fine. What changes is everything around the model. The five most repeated reasons in the 2026 analyses are always the same: - Integration with systems that already exist. The pilot read a copy of the data. Production demands reading and writing in the real system, with its permissions, its latency and its edge cases. - Inconsistent quality at volume. A hundred hand-picked cases behave differently from a hundred thousand real ones. - No monitoring. If nobody sees what the agent decided and why, nobody can authorise it to continue. - Nobody owns it. The pilot belonged to innovation or technology. The operation that should absorb it never took part, and does not claim it. - Not enough domain data. The agent knows about everything and nothing about this institution. None of those five is fixed by changing model. All five can be spotted before signing. | Criterio | Pilot | Operation | | --- | --- | --- | | Where the data comes from | A copy prepared for the trial. | The real system, with its permissions and its latency. | | Who supervises | Someone reviews every case. | Nobody watches. The limits are declared and auditable. | | Who is accountable | Innovation or technology. | A named person inside the operation. | | What it costs | Calculated on the trial. | Calculated at real volume. | A pilot proves something can work. An operation proves it works with nobody watching. The gap between them is not technical: it is who defines the limits and where they are written down. **CrediAmigo** Agent · actions live buscarUsuario Database Look up the customer by ID clasificarUsuario Logic Classify the credit level requisitos Data Validate approval requirements aprobacion Logic Conditions to approve credit cuotaMensual Compute Compute the monthly payment Code Interpreter Web Search In an operation, what the agent can and cannot do is declared and auditable: what it checks, how far it negotiates, when it hands the case to a person. A pilot usually works because someone supervises every case, and that does not scale. [See the controls and traceability →](https://www.shimliapp.com/en/seguridad) Four questions for your current pilot - Is the agent reading and writing in the real system, or in a copy prepared for the trial? If it is the second, integration remains untested. - Who, by name, inside the operation is accountable for the outcome? If the answer is a department and not a person, there is no owner yet. - What does each task cost at real volume, not in the trial? Pilot economics are almost never production economics. - Can someone review today why the agent decided what it decided, case by case? Without that, no committee will authorise it to run alone. [See the four processes already running in production →](https://www.shimliapp.com/en/casos) Shimli operates today with 160 active clients across 15 countries, with agents in production on collections, credit, support and internal operations. That is the difference this article tries to make visible: not a better model, but a process that already made it from trial to operation. ### Questions on this topic Why does a pilot that worked never reach production? Because a pilot is assessed with hand-picked cases and constant supervision, and production demands the opposite: volume, edge cases, real data and nobody watching. The five most repeated reasons are integration with legacy systems, inconsistent quality at volume, lack of monitoring, no clear owner inside the organisation, and a shortage of domain data. What sets apart a pilot that will scale? Three things you can measure before signing: that the agent already reads and writes in the real systems and not a copy, that a named person inside the institution is accountable for the outcome, and that cost per task is calculated at volume rather than in the trial. Useful to someone on your team? Share it with whoever decides on this process. [WhatsApp](https://wa.me/?text=89%25%20of%20AI%20pilots%20never%20reach%20production%20https%3A%2F%2Fwww.shimliapp.com%2Fen%2Fblog%2Ffrom-pilot-to-production) [LinkedIn](https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fwww.shimliapp.com%2Fen%2Fblog%2Ffrom-pilot-to-production) [Email](mailto:?subject=89%25%20of%20AI%20pilots%20never%20reach%20production&body=89%25%20of%20AI%20pilots%20never%20reach%20production%20https%3A%2F%2Fwww.shimliapp.com%2Fen%2Fblog%2Ffrom-pilot-to-production) Copy link ### Keep reading [See all →](https://www.shimliapp.com/en/blog) - Platform #### FINNOSUMMIT 2026: one in three sessions was about AI Instead of summarising impressions, I counted the sessions: 22 of the 67 had AI in the title. That says a lot about where banking in the region stands. September 25, 2026 - Platform #### The next AI advantage is: trust What the Volcano Innovation Summit 2026 left behind: when capability gets cheap, the advantage becomes trust. And trust has two sides. September 16, 2026 - Support #### Your next banking channel is already installed: WhatsApp AI agents are doing to apps what apps did to websites. What that means for an institution running real volume, and what separates a chatbot from an agent. September 25, 2026 ### Put an agent to work on this process. Create the account and start today. Or we take 30 minutes and define which metric you'll move. [Create account](https://shimli.app/user/auth/create-account) Book a demo