Why 95% of companies fail with Generative AI: MIT NANDA Analysis 2025

Implementing Generative AI in enterprises has become a strategic priority, but the results are alarming. According to the report
“The GenAI Divide”

published by MIT NANDA, 95% of organisations investing $30-40 billion in GenAI get zero return on investment. This study, based on more than 300 public AI initiatives, interviews with 52 organisations and surveys of 153 senior leaders, reveals a critical gap: while 80% of companies have explored tools such as ChatGPT or Copilot, only 5% of customised enterprise solutions make it to production.

The implementation of Generative AI in companies is not a technological problem, but a strategic one. Successful organisations are not those that invest the most, but those that evaluate correctly before implementing.

The harsh reality: from pilot to failure

The MIT NANDA report documents a devastating pattern in the implementation of Generative AI in enterprises: of the 60% of organisations evaluating enterprise AI tools, only 20% reach the pilot phase, and only 5% reach production with measurable impact on results. This means that 95% of projects fail before they generate real value.

Why AI projects fail

The main barriers identified in the study are:

1. Resistance to adopting new tools (score 8.5/10 on frequency)
2. Model quality concerns (7.5/10)
3. Poor user experience (6/10)

Paradoxically, the same professionals who use ChatGPT on a daily basis for personal tasks reject enterprise AI tools. One corporate lawyer interviewed in the study sums up this contradiction: “Our contract analysis tool cost $50,000, but I still use ChatGPT because it produces better results, even when the vendor claims to use the same underlying technology.

The fundamental problem is not the technology available, but that business tools do not learn from organisation-specific processes or retain the context needed to generate sustainable value.

Build vs Buy: the critical factor in implementing Generative AI in enterprises

One of the most relevant findings of the MIT NANDA report for the implementation of Generative AI in enterprises is the abysmal difference between internal developments and external strategic partnerships:

• Internal developments: 33% success rate
• External strategic partnerships: 66% success rate

Solutions with external partners are twice as likely to go into production. This contradicts the widely held belief that companies must build their own AI solutions to maintain control.

Why external partnerships work best

External experts who lead successful implementations share three characteristics:

1. Accumulated knowledge: They have seen what works and what doesn’t in multiple organisations.
2. Deep integration: Customise solutions for specific workflows rather than offering generic tools
3. Learning capacity: They build systems that adapt and improve over time.

Medium-sized companies working with specialised partners achieve full implementations in 90 days, while large corporations with in-house developments remain stuck for 9 months or more in the pilot phase.

Where the ROI really is: the investment paradox

MIT NANDA analysis reveals that approximately 50% of the Generative AI budget is spent on sales and marketing. However, the documented return on investment is somewhere else entirely.

ROI real en back-office

The 5% of organisations that have successfully crossed “the Generative AI gap” report:

• $2-10 million saved annually by eliminating BPO (Business Process Outsourcing) contracts
• 30% reduction in external creative and content agency spend
• 1 million+ per year in automated risk checks for financial services

ROI moderado en front-office

In comparison, improvements in customer-facing areas show more modest results:

• 40% faster lead qualification speed
• 10% improvement in customer retention through automated follow-ups

The paradox is clear: companies invest where AI is most visible (sales and marketing), but the real return is where it is least evident (operations, finance, procurement). Without prior strategic assessment, most organisations waste resources in areas of low impact.

What executives really demand for Generative AI implementation in enterprises

The MIT NANDA report coded the priorities that executives consistently mention when evaluating AI solutions:

• ~80% want a trusted vendor
• ~70% demand deep understanding of their workflows
• ~70% need minimal disruption with current tools
• ~66% requires capacity to improve over time
• ~63% require the tool to retain context

The problem of tools that don’t learn

66% of executives demand systems that improve over time, and 63% need tools that retain context. These figures reveal the fundamental problem: most enterprise AI solutions are static.

A lawyer interviewed for the study explains: “ChatGPT is great for initial drafts, but it does not retain knowledge of client preferences or learn from previous edits. It repeats the same mistakes and requires extensive context in each session. For high-stakes work, I need a system that accumulates knowledge and improves over time”.

This gap between what current tools offer and what users demand explains why 90% of employees use personal AI (ChatGPT, Claude) for work, while only 40% of companies have official subscriptions. There is an “AI shadow economy” where workers get more value from $20/month personal tools than from enterprise systems costing tens of thousands.

The window of opportunity: 18 critical months

The MIT NANDA report warns that the window to strategically implement Generative AI is closing. According to 17 IT procurement and sourcing leaders interviewed, enterprise RFP-to-deployment cycles range from 2 to 18 months.

The cost of inaction

Organisations that act now will establish competitive advantages that are difficult to reverse:

• Systems trained with own data and specific context
• Workflows optimised and running in production
• Prohibitive switching costs for competitors who act late

As one financial services CIO with $5 billion in assets explains: “We currently evaluate five different GenAI solutions, but the system that learns and adapts best to our specific processes will win our business. Once we have invested time in training a system to understand our workflows, switching costs become prohibitive”.

Successful medium-sized companies implement complete solutions in 90 days. Large corporations are stuck in the pilot phase for 9 months or more. This difference is not one of resources, but of approach: the former assess strategically and act decisively; the latter improvise and stall in committees.

How to cross the divide: methodology of the 5% that succeeds

The MIT NANDA analysis identifies four steps shared by the 5% organisations that generate real value with Generative AI:

1. Assessment of infrastructure and processes

Successful organisations start with a full audit of their technology infrastructure, workflows and available data. They don’t buy tools and then look for where to apply them; they identify specific problems and then look for tailored solutions.

2. Identifying use cases with measurable ROI

Instead of generic pilots, the successful 5% select use cases where they can measure measurable impact: reduction of BPO costs, reduction of agency spend, automation of manual processes with clear monetary value.

3. Implementation with specialised external partners

As the data shows (66% vs. 33% success rate), working with external experts who know what works doubles the chances of success. The best vendors don’t sell generic software, but customise deeply for specific workflows.

4. Scaling of learning systems

Successful implementations do not end at deployment. Systems must incorporate feedback, adapt to process changes and continuously improve. Emerging infrastructure such as Model Context Protocol (MCP) and frameworks such as NANDA will enable interoperability of AI agents that learn collaboratively.

Most and least disruptive sectors

The MIT NANDA report developed an AI Market Disruption Composite Index, scoring sectors from 0 to 5 on five indicators: market share volatility, growth of AI-native companies, new business models, changes in user behaviour and frequency of executive restructurings attributable to AI.

Most disruptive sectors

1. Technology (Score 2.0): New competitors gaining ground (Cursor vs Copilot), clear changes in development workflows
2. Media and Telecommunications (1.5): Rise of AI-native content, changing advertising dynamics.

Sectors with less disruption

• Professional Services (0.5): Efficiency gains, but delivery models to clients without structural changes.
• Healthcare and Pharma (0.5): Documentation and transcription pilots, clinical models unchanged.
• Financial Services (0.5): Backend automation, stable customer relationships.
• Energy and Materials (0): Almost zero adoption, minimal experimentation

Seven out of nine major sectors show significant pilot activity but little or no structural change. This gap between investment and disruption demonstrates the “Generative AI gap” at scale: widespread experimentation without real transformation.

Real employment impact: data vs. perceptions

Contrary to popular perceptions, the MIT NANDA report does not document massive layoffs due to Generative AI. The impact on employment is concentrated in historically outsourced functions:

• 5-20% reduction in customer support and back office processing operations
• Focus on disruptive sectors: More than 80% of tech and media executives anticipate hiring cuts in 24 months
• No impact on traditional sectors: Healthcare, Energy and Advanced Industries report no expectations of job cuts

The real impact is in eliminating external spend rather than reducing internal headcount. Companies that successfully cross the Generative AI gap cancel BPO contracts, reduce reliance on agencies and automate processes that previously required external consultants.

AI Assessment
: the missing assessment in the 95% that fails

The conclusion of the MIT NANDA report is clear: the problem of implementing Generative AI in companies is not technological, it is methodological. Ninety-five percent fail because they buy tools without assessing whether they fit with their infrastructure, processes and real needs.

What an IA strategic assessment includes

An effective assessment should include:

1. Assessment Workshop: Sessions with experts to discover real opportunities aligned with business objectives.
2. Technical audit: Analysis of infrastructure, data quality and workflows to detect areas for optimisation
3. Identifying use cases: Designing proof-of-concepts with measurable ROI
4. Implementation roadmap: Strategic plan with the best LLMs on the market

At Pasiona we have developed our
AI Assessment

service precisely to respond to this critical need: helping organisations to correctly assess before investing, identifying where their real ROI is and designing implementation strategies with a high probability of success.

Conclusion: from 95% that fail to 5% that create value

The implementation of Generative AI in enterprises is at a critical inflection point. With 30-40 billion invested and a 95% failure rate, the difference between success and failure is not in the technology available, but in the strategic approach.

The data in the MIT NANDA 2025 report are unequivocal:

• Solutions with external partnerships are twice as successful (66% vs. 33%).
• The real ROI is in the back-office, not where 50% of the budget is spent.
• Systems that learn and retain context are in demand by 66% of executives.
• The window of opportunity is 18 months before competitive advantages are consolidated.

El 95% de empresas fracasa porque improvisa: compra herramientas de moda, lanza pilotos sin estrategia, construye soluciones internas sin experiencia y se queda estancado meses sin resultados.

El 5% que triunfa evalúa primero: audita infraestructura, identifica casos de uso con ROI medible, trabaja con partners especializados e implementa sistemas que aprenden continuamente.

Which group does your organisation want to be in?

If you are considering implementing Generative AI in your business, the first step is not to buy technology, but to strategically assess. Contact our team at Pasiona to find out how our AI Assessment can help you cross the Generative AI gap and put you in the 5% that generate real value.