Lean Startup and Generative AI: Mistakes to Avoid

The Limited Impact of Massive Investments in Generative AI

For several years, companies have been making substantial investments in generative artificial intelligence, hoping for significant advancements. However, a 2025 study conducted by MIT's NANDA initiative revealed a concerning finding: approximately ninety-five percent of pilot projects in this field failed to produce measurable results. Although AI models technically function, the programs in which they are integrated often fail to meet their objectives. This raises questions about the effectiveness of current project management methods.

An Outdated Approach to Project Management

The persistent failure of generative AI projects has its roots in an outdated project management method. Historically, many companies have adopted a linear approach: they fund a large-scale project, define the specifications in advance, and then hope that the final product will meet expectations after a long development period. Eric Ries, in his book The Lean Startup published in 2011, had already warned against this practice following the burst of the dot-com bubble, which saw many companies fail due to products that the market did not want. According to Ries, the key to success lies in the speed of learning rather than the speed of delivery.

The Evolution of Tools and the Need for a New Discipline

Since the publication of The Lean Startup, technological tools have evolved significantly, reducing execution costs and making the Lean approach more relevant than ever. In the design field, a similar method of making large initial bets has long prevailed, but it is now being questioned. Traditional design thinking, with its lengthy and costly processes, is no longer suited to a world where prototypes can be created in a matter of hours. This approach, often billed at $300 an hour, is no longer viable in an environment where speed and efficiency are essential.

The Importance of Continuous Learning

I have personally experienced this transitional period, where we began delivering quickly by adopting short development and learning cycles. This approach, which foreshadows the principles of Lean Startup, allowed me to acquire skills that I still apply today. In an environment where AI is rapidly evolving, continuous learning has become an essential skill for adapting to relentless changes.

Genchi Genbutsu: Go to the Gemba

One of the fundamental principles that The Lean Startup continues to teach is the concept of genchi genbutsu, which means "go and see." This principle, borrowed from Toyota, emphasizes the importance of going to the field to understand real problems rather than relying solely on discussions in the conference room. Companies that succeed in navigating the challenges of AI are those that stay close to their end users and adapt their solutions accordingly.

Understand the Problem Before Proposing a Solution

According to CB Insights, the primary cause of startup failure is poor product-market fit. Companies often fail because they develop products that the market is not ready to adopt. Generative AI projects suffer from the same issue, as they are often designed to impress rather than to solve concrete problems. It is crucial to understand the real needs of users before developing solutions.

MIT researchers have pointed out that failures were not due to weak models, but to tools designed to signal innovation rather than to solve a real problem. A demonstration that impresses a boardroom does not guarantee a tool that will stand the test of time. Toyota already had a solution to this problem with the concept of genchi genbutsu — go and see in the field.

Concrete Actions to Avoid Mistakes

• Observe real work: Spend time with potential users to understand their needs before starting development.
• Clearly define the work: Identify the specific task that the product must accomplish for the user.
• Prioritize viable solutions: Focus on funding projects that provide real added value, rather than those that are merely impressive demonstrations.

Resources for Further Exploration

• CB Insights: Study on the main reasons for startup failure, highlighting poor product-market fit.
• Teresa Torres: Approach to continuous discovery to stay connected to real customer needs.

The Rapid Learning Loop: Build, Measure, Learn

The core of the Lean Startup method rests on a continuous learning loop: build, measure, learn. This approach emphasizes the importance of evaluation and learning, often overlooked when deadlines tighten. The book Sprint by Google Ventures offers a method for quickly testing ideas in five days, allowing for validation of an idea before committing to long-term development.

The five-day method developed by the Google Ventures team in Sprint allows for moving from a real problem to a tested prototype in one week. It starts with mapping the problem on Monday, building a realistic prototype by Thursday, and testing it in front of five customers on Friday. This approach allows one to know if an idea is viable before spending a quarter on its construction.

The Importance of Rapid Experimentation

The principles of Lean Startup, inspired by Toyota's lean manufacturing, encourage rapid learning and evidence-based decision-making. Generative AI, by reducing the costs and timelines of experiments, enables quick and inexpensive testing. This offers the opportunity to learn more and faster, provided these experiments are well-directed.

Actions to Optimize Learning

• Conduct targeted experiments: Identify the riskiest hypothesis and test it as a priority.
• Evaluate before delivering: Implement automated and human checks to ensure quality before delivery.
• Measure real impact: Focus on the impact of features on users, rather than the number of features developed.

Resources for Further Exploration

• Sprint, Jake Knapp et al.: Guide to testing an idea in five days.
• Lean Software Development, Mary and Tom Poppendieck: Adapting lean principles to software development.

Speed Through a Targeted Approach

True speed in development does not come from excessive ambition but from a focused and well-defined approach. MIT data shows that companies that succeed in deploying their projects quickly are those that concentrate on specific and achievable goals. Large companies, despite their resources, often fail due to their excessive ambition.

MIT data highlights that internal builds succeed about one-third of the time, while companies that opt for partnerships with specialized vendors achieve better results. The fastest players are not necessarily the large companies with numerous pilots, but rather mid-sized companies that have a narrow focus and ship efficiently.

The Importance of Specialization and Partnerships

Companies that succeed in rapidly deploying their AI projects are those that prioritize partnerships with specialized vendors. This approach allows them to focus on the essentials and avoid costly and often ineffective internal builds. Internal builds have succeeded only about one-third of the time, while mid-sized companies, with a narrow focus, have demonstrated superior efficiency.

Actions for Rapid Execution

• Prioritize partnerships: Seek specialized vendors before developing internally.
• Define a narrow scope: Focus on achievable short-term goals.
• Treat AI as a tool, not a final solution: Use AI to accelerate the process, but ensure a human validates the final result.

Resources for Further Exploration

• Fortune: Analysis of the MIT NANDA report on the importance of partnerships in AI deployment.
• Patrick Neeman: Experience writing a book with generative AI.

Documentation: A Support, Not an End in Itself

In Toyota's production system, documentation that serves only itself is considered waste. It should support the work, not replace it. Companies must ensure that documentation remains a useful tool and not an unnecessary administrative burden.