95% of AI Pilots Fail? Why?

Despite $30 to 40 billion in enterprise investment into GenAI, 95% of organizations are getting zero return. Why is this the case? What is the short answer?

• It is not the underlying AI technology.
• It is not even adoption; many companies spend money attempting to apply AI.
• It is principally due to failure in the proper execution of AI implementation.

MIT recently published a report titled “The GenAI Divide: State of AI in Business 2025,” stating that 95% of enterprise AI pilots are failing to generate any economic impact. While this has created a number of catchy headlines, one needs to delve into the detail of the report and the nature of AI implementation to truly understand the reasons, and what they imply.

What is clear from the study, and from my personal experience, is that this is not a question of the potential of AI or, in many cases, of adoption, but rather the proper implementation and application of the technology. As a long time board member of several companies and the recent CEO of a large scale, process oriented company, I have witnessed countless transformation projects fail.

The following are my views on the key reasons for these early stage underwhelming results. This is not meant to replace or repeat the report, but merely to give you my experience based views.

Full institutional commitment required

Enterprise level tech implementations and transformation programs are complicated and need complete institutional buy in to succeed. Often even the best designed projects fail when sponsored only at the top, yet succeed in other contexts when the entire organization is fully committed. This deeper buy in is difficult to achieve due to the inherent threat that AI poses to the historical way of doing things and to large portions of the workforce.

Institutional buy in is only accomplished through the combination of:

• strong senior management leadership and sponsorship;
• clear and frequent communication of objectives and expected benefits;
• proper project design and management structure;
• clear communication of what is expected of local management (budgets, performance evaluations); and
• proper incentives (bonuses, promotions).

Dealing with the J curve effect

It is almost always the case that using AI in your service or product delivery is less effective at the outset than existing models. Companies need to be comfortable with an imperfect product at the start that will improve dramatically over time. This is the so called “J curve effect,” where performance may need to suffer now to provide superior benefits later. This requires a high degree of maturity in managing clients, employees, and other stakeholders that is not common in most companies.

Creating an AI learning culture

AI literacy is a fundamental capability requirement for the modern enterprise; AI proficiency represents a competitive advantage in workflow optimization. This capability needs to be inculcated across organizations through training of the existing workforce and in new hiring. Without this proficiency, no company can succeed over the long term.

Lack of correct focus on the application of AI

GenAI tools like ChatGPT and Copilot are widely adopted: over 80% of organizations have explored or piloted them, and nearly 40% report deployment. But these tools primarily enhance individual productivity, not P&L performance.

More than 50% of GenAI budgets go to sales and marketing, often adding visible features but lacking the clear ROI potential of back office and service delivery process automation.

Successful enterprise applications (and many AI start ups) focus on narrow but high value use cases, integrate deeply into workflows, and scale through continuous learning rather than broad feature sets, evaluating tools based on business outcomes rather than software benchmarks.

Failure to adopt an AI operating model, and to take hard decisions

The most common misinterpretation of AI’s role is that it is a tool to add to existing systems to improve performance. If you view AI as merely a tool, you are adding complexity onto your existing systems (typically a web of varied and aging systems built up over time), keeping humans at the core of the basic processes, and you are doomed to gain merely incremental benefits.

The reality is that AI allows us to create an entirely new operating model that recreates core business processes and service delivery holistically. The combination of (i) GenAI/Voice AI enabling customer interactions superior to human interaction and (ii) decision engines and process automation creates the potential for vast efficiency and effectiveness gains.

More importantly, this technology creates a self contained system that replaces almost your entire service/product delivery model and large parts of your system architecture. Clinging to legacy systems and viewing AI as a tool are the principal reasons start ups are dramatically more successful in AI implementation while large enterprises often fail. The challenge is that large enterprises typically possess the large commercial footprint necessary to maximize AI’s economic impact; the key is implementing at enterprise level the way a start up would. This is a great cultural and management challenge, and the key to what MIT calls the “AI Divide.”

The last and most obvious challenge: automation of processes and delivery models provides improved performance at lower unit cost, but that cost advantage is not scalable, and true ROI is not possible, unless implemented across the company. The largest factor in realizing true economic impact is the corresponding reduction in human resources. Reducing the workforce allows run rate cost reduction at scale while the AI improves your delivery and, combined with the learning culture described above, sets you up for success.

Not to end on a negative note: the progress to date is real and holds great potential. If I can identify and adapt to the factors above and the additional hurdles identified in the report, then any leader can do so. It doesn’t guarantee success, but it improves your probability of success.

Despite the headline failure this report highlights, the future is very bright in my opinion. Looking at the data, one can focus on the 90% of complex tasks not entrusted to AI, but I choose to focus on: (i) the fact that 70% of common tasks are entrusted; (ii) that AI will make complex tasks simpler to execute; and (iii) that the 10% of complex tasks currently entrusted to AI will inevitably increase dramatically.

When one considers that this type of broad discussion of AI didn’t even exist a few years ago, one has to think that a few years from now we will be in a much better place.

I hope this post is informative. The MIT report, “The GenAI Divide: State of AI in Business 2025,” is great reading. Have a great day. Andrés