Why 70% of AI investments fail

Reading time: 10 minutes

Introduction: The €30-a-Month Disillusionment

Picture the scene. It's Monday morning at a large company. The CEO has just proudly announced during the Executive Committee meeting: "We have deployed Microsoft Copilot for the entire company. We are now an AI-first organization."

Applause erupts. The budget is approved: €30 per month, per employee. For 500 employees, that's €180,000 per year. A significant investment, but a necessary one to remain competitive.

The licenses are rolled out within a week. A two-hour training session is organized. The announcement email is sent. The "AI transformation" box on the strategic plan is checked.

Fast-forward three months later.

The same CEO receives an alarming report from the IT department: the effective usage rate of Copilot is stagnating at 12%. The majority of employees opened the tool once or twice, then went back to their old ways of working.

€180,000 per year for a 12% usage rate. The ROI is catastrophic.

I hear this story almost every week. The numbers vary, the companies change, but the scenario remains the same. And you know what? It's not the fault of the technology, the employees, or even the budget.

It's the fault of a fundamental design error.

---

Part 1: The Anatomy of a Programmed Failure

The "Software" Mindset Trap

For decades, companies have learned to deploy software using a well-oiled process:

1. Needs Analysis → Meetings with stakeholders
2. Vendor Selection → Comparison of solutions
3. Contract Negotiation → Cost optimization
4. Technical Deployment → Installation and configuration
5. Initial Training → 2-4 hour session
6. IT Support → Helpdesk for technical issues

This process works perfectly for tools like a CRM, an ERP, or a classic office suite. Why? Because these tools integrate into existing processes without fundamentally disrupting them.

But generative AI is different.

Radically different.

Why AI Is Not Just Another Software

Let's take the example of a CRM. When you deploy Salesforce, you ask your sales team to:

• Enter customer information in a structured format
• Follow a defined sales process
• Generate standardized reports

The change is procedural: "Instead of doing X in Excel, do X in Salesforce."

Now, let's take Microsoft Copilot or ChatGPT in the enterprise. You are asking your employees to:

• Rethink how they write (no longer writing from scratch, but co-creating)
• Change their relationship with information (no longer searching, but querying)
• Modify their time management (delegating repetitive tasks)
• Transform their role (moving from producer to supervisor)

The change isn't procedural. It's cognitive and cultural.

It's like asking someone who learned to write by hand to switch to a keyboard. Technically, it's simple. Psychologically, it's an upheaval.

The Typical Lifecycle of a Failure

Let me tell you the story of Sophie, an HR manager in a company of 800 people.

• Day 1: Enthusiasm Sophie attends the Copilot training. The trainer shows impressive demos: summarizing emails, generating presentations, analyzing documents. Sophie is excited. She immediately sees 10 use cases for her daily work.

• Days 2-3: Experimentation Back at her desk, Sophie tries Copilot. She asks it to summarize her inbox. It's convenient, but not revolutionary. She tries to generate a memo. The result is decent, but requires as much proofreading as if she had written it herself. She makes a few other attempts.

• Days 4-7: Frustration Sophie runs into difficulties:

  • She doesn't know how to formulate her prompts well
  • The results are sometimes off-target
  • No one around her is using the tool, so she can't exchange tips
  • Her manager hasn't given her clear objectives for using Copilot
  • She feels alone in her experimentation

• Days 8-30: Reverting to Old Habits Facing an important deadline, Sophie falls back on her old methods. They are faster, safer, more comfortable. Copilot requires too much effort for uncertain results. Gradually, she stops opening the tool.

• 3 Months Later Sophie has completely abandoned Copilot. When asked why, she replies, "I didn't really see the point. For my job, it doesn't work very well."

The problem? It's not Sophie. It's the system that let her down.

---

Part 2: Where Adoption REALLY Happens (and Why You're Not Seeing It)

The Illusion of Training

Here's an uncomfortable truth: training does not create adoption.

Training creates knowledge. And knowledge is not enough.

Imagine I give you a 2-hour training session on how to play the piano. By the end, you know:

• The position of the notes on the keyboard
• How to read a music sheet
• The basics of music theory
• A few simple exercises

Do you know how to play the piano? No. You know what you need to do to learn how to play the piano.

To actually play, you need:

• Hours of daily practice
• Regular feedback from a teacher
• Real-time correction of your mistakes
• Motivation to continue despite difficulties
• Exposure to different styles and techniques

AI adoption follows the exact same logic.

The Invisible Moments That Make the Difference

Adoption doesn't happen in the training auditorium. It happens in these micro-moments that no one measures:

• Moment 1: The Hallway (Day 5) Julie bumps into Marc at the coffee machine.

  Julie: "Are you using Copilot?" Marc: "I tried it, but I don't think it's that great." Julie: "Same here, I'm giving up on it." Result: Two potential users lost.

• Moment 2: The Team Meeting (Week 2) The manager asks, "Does anyone have any questions about Copilot?" Awkward silence. No one dares to admit they're not using it. The manager moves on to the next item. Result: Zero experience sharing, zero collective learning.

• Moment 3: The Desk (Week 3) Thomas spent 2 hours writing a report. His colleague David walks past.

  David: "Did you know Copilot could save you 45 minutes on this type of report?" Thomas: "Really? How?" David shows him in 5 minutes. Result: One user won over, who will go on to convert others.

• Moment 4: The Internal Teams Channel (Week 4) Sarah, an enthusiast, posts in the general channel: "Copilot tip of the day: I used [specific prompt] to prepare my client brief in 10 minutes instead of an hour. Here's how..." 15 people try it the next day. 3 share their own tips. Result: The creation of a positive viral dynamic.

These moments don't appear in any deployment plan. No KPI measures them. No budget funds them.

And yet, they are what makes the difference between success and failure.

The Real Adoption Curve

Companies think adoption looks like this: Training → Immediate Use → ROI ✅ ✅ ✅

But reality looks more like this: Training → Initial Enthusiasm → Trough of Disillusionment → Plateau of Frustration → Abandonment ✅ 📈 📉 😫 ❌

Or, in organizations that succeed: Training → Enthusiasm → Difficulties → Support → First Wins → Sharing → Snowball Effect → Integration ✅ 📈 📉 🤝 💡 🗣️ 🚀 ✅

The difference? Support through the "trough of disillusionment."

---

Part 3: AI Is Not an IT Problem, It's a Leadership Challenge

The False Logic of Assigning It to IT

In 80% of organizations, the conversation goes like this:

Exec Committee: "We need to deploy AI." CIO: "I'm on it." Exec Committee: "Great, keep us updated."

The CIO does what they know how to do: evaluate solutions, negotiate contracts, deploy technology, manage security, and train users.

All of this is necessary. But it is absolutely not sufficient.

Why? Because the CIO cannot change the culture.

They cannot:

• Force managers to use the tool in meetings
• Create a psychologically safe environment for experimentation
• Transform the working methods of every department
• Get HR to rethink their recruitment processes
• Convince legal to revise their drafting processes

AI transforms how people think and work. It's not an IT project. It's an organizational transformation project.

The 4 Fatal Leadership Mistakes

• Mistake #1: Delegating without getting involved The CEO announces the AI deployment with enthusiasm... and then never uses the tool themselves. Message sent to the teams: "This is important for you, but not for me."

• Mistake #2: Not allocating time Teams are asked to adopt AI... without being given dedicated time to experiment, learn, fail, and try again. Message sent: "Figure it out on top of your normal workload."

• Mistake #3: Not celebrating wins Peter automated a process that saves his team 5 hours a week. No one knows, no one values it, no one talks about it. Message sent: "Your innovation efforts don't count."

• Mistake #4: Punishing failure Marie tries Copilot to write an internal memo. The result requires a lot of corrections. Her manager comments, "You should have just written it yourself, it would have been faster." Message sent: "Don't try new things, it's risky."

What Successful Leaders Do

I've observed dozens of organizations. The ones that succeed have leaders who:

1. Use AI publicly and daily The CEO starts meetings by sharing how they used Copilot to prepare the agenda. They share their screen, their prompts, their results. They don't hesitate to show their failures, too. Impact: Teams see that it's serious, that it's normal to experiment, that even the boss is figuring it out.

2. Create risk-free spaces for experimentation "On Friday afternoons, you can spend 2 hours experimenting with AI on your projects. No reporting, no objectives, just testing." Impact: People have the time and permission to learn without pressure.

3. Value and amplify successes Every week during the team meeting, 10 minutes are dedicated to "AI Wins": who used AI this week to achieve a concrete result? Impact: Creates a dynamic of sharing and positive emulation.

4. Protect the right to make mistakes "We encourage experimentation. If it doesn't work, that's okay. If you don't try, that's a problem." Impact: People dare to step out of their comfort zones.

---

Part 4: The Complete Framework for Successful Adoption

The 4 Non-Negotiable Pillars

After supporting dozens of organizations, I've identified 4 absolutely essential pillars:

Pillar 1: Visible and Involved Leadership

• What doesn't work:
  • Announcing the deployment and then disappearing
  • Delegating adoption to IT or HR
  • Using AI only in private
• What works:
  • Executives use AI in their visible, daily work
  • Managers model the expected behaviors
  • Leadership publicly celebrates successes
  • Leaders talk about their own learning struggles

Concrete Example: An HR Director I work with starts every team meeting with a 5-minute "demo time": she shows how she used AI that week, shares her prompt, and explains what worked and what didn't. The result: her team has the highest adoption rate in the company (78% active users).

Pillar 2: Culture of Experimentation

• What doesn't work:
  • "Use Copilot" with no further guidance
  • Punishing unsuccessful attempts
  • Maintaining a culture of perfectionism
• What works:
  • Explicit permission to experiment
  • A clearly established right to fail
  • Time allocated for learning
  • Sharing failures as well as successes

Concrete Example: A law firm established "Legal AI Labs": every Friday afternoon, lawyers can spend 2 hours testing AI on their cases, with no obligation to produce results. The best use cases are shared the following Monday. In 3 months, they identified 15 concrete applications that collectively save 40 hours per week.

Pillar 3: Structured, Continuous Support

• What doesn't work:
  • A one-off 2-hour training, then "figure it out"
  • Technical support only
  • Lack of post-training follow-up
• What works:
  • A support program of at least 90 days
  • Internal champions are identified and trained
  • Specific, job-related workshops (not generic ones)
  • Weekly sharing rituals
  • Continuous monitoring and adjustment

The typical structure of successful support:

• Week 0: Initial training (2-3h) + champion identification
• Weeks 1-4: Weekly "Office Hours" sessions (30 min) + dedicated Teams channel
• Weeks 5-8: Department-specific workshops (to identify specific use cases)
• Weeks 9-12: Consolidation + creation of internal playbooks
• After 12 weeks: Rituals are maintained, champions become self-sufficient

Pillar 4: Measurement and Iteration

• What doesn't work:
  • Measuring only the technical deployment
  • Looking only at login rates
  • A complete absence of KPIs
• What works:
  • Qualitative AND quantitative KPIs
  • Weekly measurement
  • Regular user interviews
  • Data-driven adjustments

The Metrics That Really Matter:

• Usage Metrics:
  • % of weekly active users (not just "logged in once")
  • Average frequency of use
  • Types of prompts used
  • Time spent in the tool
• Impact Metrics:
  • Time saved on specific tasks (measured by survey)
  • Number of business use cases identified
  • User satisfaction (AI-specific NPS)
  • ROI calculated by department
• Cultural Metrics:
  • Number of "best practices" shared
  • Participation in sharing sessions
  • Number of emerging champions
  • Evolution of sentiment (qualitative analysis)

---

Part 5: The Critical 90 Days (The Concrete Action Plan)

Phase 1: The First 30 Days - Building Momentum

Objective: Prevent the fall into the "trough of disillusionment"

• Week 1: Immersion
  • Days 1-2: Engaging initial training (not a boring PowerPoint)
  • Days 3-5: Simple daily challenges sent by email ("Use Copilot to summarize this memo in 3 bullet points")
  • Identification of 5-10 champions per department
• Week 2: First Wins
  • "Office Hours" session #1: FAQ and troubleshooting
  • Creation of the "#AI-tips" Teams channel
  • Champions share their first use cases
• Week 3: Sharing
  • "AI Success Stories" meeting (30 min): 3-4 people present a concrete case
  • Begin user interviews (to understand roadblocks)
  • Adjust based on initial feedback
• Week 4: Consolidation
  • "Office Hours" session #2
  • Creation of the first internal "Playbook" (10 validated use cases)
  • Public recognition of top contributors

Phase 2: Days 31-60 - Deepening and Diversifying

Objective: Move from experimentation to integration

• Weeks 5-6: Business Unit Workshops
  • HR Workshop: Use cases for recruitment, training, communications
  • Legal Workshop: Contract drafting, legal research, summarization
  • Sales Workshop: Proposals, client emails, meeting preparation
• Weeks 7-8: Creating Routines
  • Integration into existing processes
  • Established weekly rituals
  • Training managers to become sponsors

Phase 3: Days 61-90 - Sustaining and Scaling

Objective: Make adoption autonomous and viral

• Weeks 9-10: Documentation
  • Creation of detailed, department-specific playbooks
  • Short use-case videos (2-3 min)
  • Enriched FAQ
• Weeks 11-12: Empowerment
  • Champions lead the sessions themselves
  • Implementation of a peer-to-peer mentoring system
  • Planning for what comes next

---

Conclusion: Adoption Is Not a Destination, It's a Journey

Let's go back to our CEO from the beginning, the one worried about a 12% usage rate.

Now, let's imagine an alternative scenario.

Instead of treating the Copilot deployment as an IT project, he treated it as an organizational transformation. He invested as much in support and coaching as he did in licenses. He used the tool himself, publicly, sharing his mistakes and successes.

90 days later, the report is different:

• 68% weekly active users
• 23 business use cases identified and documented
• Estimated savings of 2.5 hours per week per user
• ROI: 340%

Same investment in licenses. Radically different result.

The difference? The human support.

The 3 Truths to Remember

1. AI isn't a tool, it's a transformation As long as you treat it like software, you will fail. It transforms how people think, work, and collaborate. That requires time, patience, and human guidance.

2. Adoption doesn't happen in the training room It happens in the hallways, in meetings, on Teams, in informal moments. Create the conditions for these moments to be positive and constructive.

3. Leadership makes all the difference This is not an IT project. It's a leadership project. Executives must be the first users, the first role models, and the first sponsors.

The Real Question

Don't ask yourselves, "Have we deployed AI?"

Ask yourselves:

• Who is supporting our teams during the critical 90 days?
• What kind of culture are we creating around experimentation?
• What are we really measuring?
• Where are we in our transformation?

AI adoption isn't a box to be checked. It's an organizational muscle to be developed.

The organizations that win are not the ones with the best tools.

They are the ones that invest in the humans who use those tools.

---

Over to You

Where is your organization on its AI adoption journey?

• 🔴 Phase "We just deployed, no one is using it"
• 🟠 Phase "A few champions are using it, the rest are watching"
• 🟡 Phase "Usage is growing but not yet integrated"
• 🟢 Phase "Massive adoption and measurable ROI"

Share your experience, your roadblocks, your successes. Collective learning is the key to success.

At AI Tomorrow Solutions, we help non-tech departments (Legal, HR, Operations) turn their AI investments into measurable results, with a 90-day support program that covers the 4 essential pillars. Because adoption isn't an event. It's a process. And we are here to guide you every step of the way.

Does this article resonate with your experience? Feel free to:

• 💬 Comment with your own story
• 🔄 Share to help other organizations
• 📧 Contact me to discuss

#AI #DigitalTransformation #Leadership #Adoption #Copilot #ChangeManagement

---

Note: This article is approximately 2500 words and should take about 10 minutes to read. It significantly expands on each point from your posts with concrete examples, anecdotes, and a detailed action plan.