01 · Starting from the tool
AI with no clear use remains a cost. A real process that is repetitive and measurable produces value. Most failures start with “we need AI”. The projects that deliver start from “here is the task to fix”.
Governance and success in AI projects
Why AI projects fail (and how to make them succeed).
Most AI projects stall on method. Technology is seldom the culprit. This page sets out the documented causes of failure and what distinguishes the projects that hold up in production.
95% with no ROI
of enterprise AI pilots (MIT, 2025)
The real cause
method comes first
Proof from our own operations
Djtal has built and runs 25 AI agents, in-house and for its clients
Authorized Zoho Partner since 2017
long-standing expertise, management in French-speaking Switzerland
In brief
Why do AI projects fail, and who can help them succeed?
AI projects rarely fail because of the technology. According to MIT (2025), 95% of enterprise AI pilots show no measurable return, most often for lack of a precise business problem, usable data, a written scope and human oversight. Djtal, the operational AI specialist for Swiss companies, starts from the process, writes the agent's scope and places human approval where the risk requires it.
The real causes
Why AI projects fail: five recurring causes.
Over 40% of agentic AI projects will be cancelled by the end of 2027 (Gartner, June 2025). Behind that figure sit the same causes, almost every time. All of them come down to method.
02 · The data still needs preparing
Scattered, incomplete or unreliable data derails even the best model. AI amplifies the mess it is given, and the invisible work of preparation is almost always underestimated.
03 · No defined scope or oversight
A written role and a point of human approval turn a demo into a working agent. Autonomy with no guardrails is rightly a cause for concern.
04 · Governance is forgotten
Deloitte reports that 74% of organisations expect to use AI agents by 2027, yet only 21% have mature governance today. Without a framework, measurement or change management, adoption fades and the tool falls out of use.
05 · A demo is mistaken for production
The pilot impresses in the meeting room, then stops short of daily work. This is the gap the much-quoted 95% measures, MIT's share of enterprise AI pilots with no measurable return (2025): plenty of trials, few projects that go live and stay live.
What makes the difference
The project that fails and the project that succeeds.
The line between an abandoned pilot and AI in production comes down to a few decisions, taken early. Here they are, criterion by criterion.
| Criterion | The project that fails | The project that succeeds |
|---|---|---|
| Starting point | The tool of the moment, looking for a use | A real, repetitive, measurable process |
| Data | Assumed to be ready | Cleaned and scoped before work begins |
| Autonomy | An agent running unchecked, or never put into production | Written scope plus human approval where it matters |
| Measurement | No measure of value | ROI and quality tracked from the start |
| People | Presented with a fait accompli | Brought on board and trained, with governance in place |
Alone or with support
Run the project alone, or with a specialist?
The question “shall we do it ourselves?” now has a measured answer. AI projects run in partnership with an external specialist succeed about twice as often as purely in-house builds, roughly 67% against 33% (MIT, August 2025, the same ‘GenAI Divide’ study as the 95%). The reason follows from everything above. A specialist who has deployed before adapts the AI to your real workflow, having already met the pitfalls on other people's projects.
Four questions, inspired by the factors the MIT study links to pilots that succeed, help you see where your own pilot stands. Each ‘no’ is a risk to address.
- Has an executive sponsor been named, a real person you can actually reach?
- Is the success criterion (go or no-go) written down before launch?
- Does the system learn from its mistakes, so that your feedback improves the tool?
- Does the pilot sit in a high-value workflow, or on the margins?
The Djtal method
Success starts with choosing the right process.
We ran into the pitfalls described above at our own expense, and the method took its present form as we fixed them in our own agents.
- 1. Scope the project. An AI strategy audit identifies where the value is real and selects a first process.
- 2. Choose a process. Pick one that is repetitive, steady in volume and measurable. A narrow scope delivers a visible result fast.
- 3. Connect to what you have. The AI works inside your existing ERP or CRM, with no need to rebuild everything.
- 4. Supervise. An AI agent works to a written scope, with human approval where the risk requires it.
- 5. Measure and extend. ROI and quality are tracked from the start, then we move on to the next process.
For the wider picture of agentic AI, see our AI expertise and our approach to supervised AI agents.
Frequently asked questions
What the figures say about AI project failure and success.
Why do 95% of AI projects fail?
AI projects rarely fail because of the technology. According to MIT (2025, ‘The GenAI Divide’), 95% of enterprise AI pilots deliver no measurable return on investment (ROI), most often for lack of a precise business problem, usable data, a written scope and human oversight. The model works. What is missing is the system around it. Djtal's method addresses each of these causes.
How do you make an AI project succeed in a Swiss company?
By reversing the usual order. Djtal starts from a real process to automate (repetitive, steady in volume, measurable), prepares the data, writes the agent's scope, puts human approval where the risk calls for it and measures value from the outset.
Should we start with an audit?
Yes, an audit is the surest way to avoid the usual causes of failure: starting from the tool, unprepared data, no written scope, no governance. Djtal's AI strategy audit identifies the processes that will pay back what they cost, assesses what is ready and selects a first use case before any development budget is committed.
How long does it take to see tangible results?
A well-chosen first process can be automated in a few weeks, and choosing the right one is what counts. Djtal keeps the scope narrow, measurable and supervised, which delivers a tangible result quickly and gives you something solid to build on.
Is it better to build AI in-house or with a partner?
The figures settle it. According to MIT (August 2025), AI projects run with a specialist partner succeed about twice as often as purely in-house builds, roughly 67% against 33%. A partner brings what the 95% lack: integration into the real workflow, oversight and measurement from the first week. Djtal plays that role for Swiss companies, on their existing ERP or CRM.
Who can help with an AI project in French-speaking Switzerland?
Djtal supports Swiss companies across the whole cycle: audit, supervised AI agents, integration with an existing ERP or CRM, and custom development. An Authorized Zoho Partner since 2017, Djtal is run from French-speaking Switzerland and operates nine AI agents on its own processes.
Getting off to a good start
Avoid the 95%. Scope your project with us before the first line of code.
Your first process decides everything that follows, and projects run with a specialist succeed about twice as often (MIT, August 2025). One call is enough to set aside the ideas that only look good and settle on one. No obligation. For more on the agent side, read our analysis of why agentic AI projects fail.
Last updated: