Research by MIT NANDA and McKinsey covering more than a thousand AI projects reached the same conclusion: 95% of AI investment produces no result. Only 5% of projects hit their planned ROI. Three quarters never leave the pilot stage.
This is a practical guide for managers on why that happens and how to end up in the other 5%. It draws on my own experience and on research into both successful and failed projects, and it focuses on the thing that actually decides the outcome — your management system. Because not every company culture is ready to get value from AI.
You will get the six traps that swallow AI budgets, the patterns behind the projects that work, and step-by-step plans sized for a small, medium or large company.
Inside the Book
1. The Problem: Why 95% of AI Investment Produces Nothing
The numbers first, from MIT NANDA and McKinsey research covering more than a thousand projects: only 5% hit their planned ROI within 18 months, 75% never leave pilot mode, and companies without a systems approach averaged −2% EBIT. This chapter explains the GenAI Divide between the leaders and everyone else, pinpoints exactly where projects die on the way from pilot to production, and describes the Shadow AI Economy — the AI already running inside your company that nobody approved and nobody manages.
2. Six Traps — and the Eight Patterns of the 5% Who Succeed
The traps, each with its way out: systems that cannot learn or remember, investment flowing to the wrong places, treating AI as a plaster over a broken process, people resisting and quietly sabotaging, dirty data producing dirty results, and building a pilot with no plan for scale. Then the mirror image — what the winners consistently do, backed by numbers: buying rather than building (67% success rate), fixing the process before adding the technology, the 70/30 rule of chasing the optimum instead of the maximum, rewarding usage rather than outcomes, and treating rework as normal rather than as failure.
3. The Management System
The core framework: eight components of the systems approach, three pillars that must be built at the same time — governance, technology and people — and three levels of AI management. Practically, this means tying AI to your business strategy through five questions and a goal-cascading model, choosing between a centralized, distributed or hybrid organizational model, standing up a Chief AI Officer and an AI steering committee, and honestly assessing whether you are ready — with two ready-to-use questionnaires, one for pilot readiness and one for scaling. The whole path is laid out as four waves over roughly 18 months: foundation, quick wins, scaling, transformation.
4. Process First, Then Technology
Why lean comes before AI: automating a wasteful process just makes waste faster. The eight types of waste and the tools to remove them, then the decision that shapes everything downstream — buy, build, or adapt — with an honest comparison of all three and when each one wins. Finally the infrastructure: on-premise versus cloud versus hybrid, how to choose, data masking and anonymization for sensitive workloads, compute requirements, avoiding vendor lock-in, securing your models, and a selection checklist.
5. From Idea to Portfolio to Delivery
How to find the right projects and run them: the typology of projects in a portfolio, three phases of identifying and prioritizing ideas, and ongoing portfolio monitoring. Then the AI project lifecycle in seven stages with explicit decision gates, the tables and tools to run each one, and the mistakes that most often derail projects. Plus a full chapter on risk across six categories — technical, legal and regulatory, ethical and reputational, operational, financial, and human — with a four-step process for managing them.
6. Proving It Paid: Metrics and ROI
Technical metrics are not enough to justify an AI budget. This chapter builds a balanced scorecard across four perspectives — financial, customer, process, and learning and growth — shows which metrics belong in each, how to build the dashboards and reporting around them, and gives a practical, step-by-step method for calculating return on investment that will survive a conversation with your CFO.
7. People: The Part That Actually Decides It
The problems that sink AI projects are rarely technical. This chapter gives you a competency model for AI adoption — who you need and what each of them must be able to do — and an honest look at how organizational culture either carries the change or quietly kills it. Then the practical half: managing resistance, a three-level communication model that reaches executives, middle managers and frontline staff in the terms each of them actually cares about, and three specific techniques for the person leading the project.
8. Generative AI and LLMs in Practice
The hands-on chapter for the technology everyone is buying right now. How to architect a GenAI solution, how to write instructions that produce reliable output rather than impressive demos, and how to control hallucinations instead of hoping they go away. Plus the part most teams discover too late — the economics of GenAI: where the costs actually accumulate, how they scale with usage, and how to keep them from quietly outgrowing the value. Closes with worked practical cases.
9. Playbooks by Company Size
The same system, sized to fit. Small companies: your real profile and constraints, the four problems you will hit, why your strategy is "buy," a portfolio you can afford, cases from companies that made it work, and a 3–6–9 month plan with a readiness checklist and the five mistakes to avoid. Medium: buy plus adapt plus build, standing up a center of excellence, portfolio management, and a 12–18 month plan. Large: build plus buy plus adapt, corporate-level AI governance, managing change across 1,000+ people, and a 24–36 month roadmap. It closes with the lesson from China's national AI strategy — plus appendices you can use directly: the ten biggest project risks, the red flags that mean stop, data quality requirements, quality metrics for each type of AI, and a regulatory compliance table by region.