In June 2026 the State Comptroller published a number that belongs on the wall of every CEO in Israel: 68% of the AI projects in the organizations audited never got past the pilot. Only 32% made it into production. The same report found that just 18% of organizations have a written AI strategy, and that 41% run with no data governance at all.
I read those numbers as a CEO, not as a technologist. 25 years of running companies teaches you exactly what a project looks like when everyone is pleased with it and nobody is making money from it. A stalled AI pilot is that project: a good deck, an impressive demo, and not one line in the P&L that moved.
The Israeli paradox: the employees are sprinting, the organization is standing still
Here is what makes the 68% so strange. Anthropic's usage index ranks Israel number 1 in the world for individual AI adoption, 4.9 times higher than the country's size would predict. Industry surveys put 95% of tech-sector employees already using these tools in their daily work.
Your people are already there. They draft emails with it, summarize meetings with it, write code with it. The organization is not there. That gap is not technological, it is managerial, which is why it never closes with one more tool, one more license or one more inspiring lecture.
The gap carries a quiet cost as well. When the organization sets no rules, employees set their own: company data, open tools, no oversight of any kind. The rest of the world has already worked this out. According to IBM, 76% of large organizations appointed a chief AI officer in 2025, up from 26% a year earlier. The question is no longer whether someone owns this in your company. It is who. And if the answer is nobody, then every employee owns it alone, with no boundaries and no accountability.
The four failures behind every stalled pilot
MIT's NANDA project examined hundreds of enterprise GenAI deployments in 2025 and found that 95% of pilots moved the P&L by nothing at all. Set that against what I see inside Israeli organizations, and the same four failures keep surfacing.
- The project is chosen by hype, not by money. MIT found most budgets flowing to marketing and sales, while the highest measured ROI sits in operations and finance. A chatbot on the homepage is glamorous. Cutting invoice processing from 4 days to an hour is money.
- The tool does not fit the real work. Anything that makes an employee open one more screen and copy-paste between windows is dead within a month. AI works when it sits inside the process that already exists: in the inbox, in WhatsApp, in the system already open on the employee's desk.
- Built in-house when it should have been built outside. The same MIT study found that tools built with an external partner reach adoption at twice the rate of internal builds. I have written about this separately, but the short version is that your IT team is buried, and this is not their profession.
- No signed success metric and no business owner. If nobody on the executive team can finish the sentence "this pilot succeeds if X drops by Y% by date Z", the pilot will never end. Not because it failed. Because nobody ever defined what winning looks like.
What you can fix tomorrow morning
You do not need a steering committee or an 80-page strategy document. You need four decisions, and all four are yours to make.
- Pick one process with a price tag on it. Not "improve the customer experience" but "a report that eats three days of manual work every month", or "300 service calls a day, half of them the same call twice". A process you can write a shekel figure beside.
- Name one owner from the executive team. Not the head of innovation. The person whose budget carries that process and who will feel the outcome in their own KPI.
- Sign the metric before anything starts. A baseline measured today, a numeric target, a decision date. If someone will not put their name to a number, do not start with them.
- Ask for a small production system instead of a big pilot. One narrow system running on real data with real users inside 30 days beats a six-month POC in the lab. Over my career I have watched this from both sides: as a CEO waiting on vendors, and as the person who now builds these systems himself, from a voice bot that answers real calls to a management report that assembles itself overnight at a large medical institution in the north. The distance between "works in the demo" and "works at eight on a Tuesday morning with the system under load" is the whole of it.
One more word about expectations. You do not have to invent anything to know where the money is. Published industry benchmarks put savings from AI-supported customer service at roughly 30%; invoice processing automation removes 70% to 90% of handling time; and across the five proven agent use cases, customer service, contract review, supply chain, code modernization and fraud detection, the measured average return on investment is 171%. Start there, in the boring places where the money already sits, rather than in the exotic one that photographs well on a slide.
The State Comptroller's 68% is not a verdict. Those projects are stuck because nobody ran them the way a business runs a project, and that is exactly what a CEO knows how to fix. Sometimes all it takes is someone who speaks the language of the executive team and builds the thing with his own hands.
If you want to see what that looks like inside your organization, write to me. I will show you a demo built on a real process of yours.
