The world of AI investing was recently sent into a minor panic, and for good reason. It’s not just the disappointing launch of GPT-5, Mark Zuckerberg’s AI hiring freeze or the talks about an impending AI bubble about to burst.
A new report from MIT Media Lab’s NANDA Initiative, “The GenAI Divide: State of AI in Business 2025“, delivered a stark statistic: 95% of enterprise AI pilot projects are failing to yield measurable financial returns.
This finding spooked investors, driving down shares of major tech companies linked to AI. But while the headlines screamed about a potential “AI bubble,” the true implications of this report should be making C-suite executives even more anxious than investors. But for AI startup founders, this outcome is not primarily an indictment of AI technology itself, but rather an indication of how enterprises are currently implementing AI.
Why are AI pilots failing in the Enterprise?
The MIT report identifies key reasons for 95% of AI pilots failures. In a nutshell, the reasons stem from a fundamental mismatch between generic AI tools and complex enterprise environments, rather than a lack of technological capability.
For startup founders, it would be important to understand these issues when approaching enterprise clients:
- The “Learning Gap” is Significant. Enterprises and their employees often lack the understanding of how to effectively use AI tools or design workflows to capture benefits while minimizing risks. Generic tools like ChatGPT, while effective for individual productivity, fall short in enterprise settings because they do not learn from specific workflows, retain feedback, or adapt over time. Your startup needs to build systems that do learn, remember, and evolve, addressing this fundamental gap.
- Misaligned Investment: Prioritising Visibility Over Core Value. Over half of generative AI budgets (50-70%) are still allocated to sales and marketing tools. However, the report indicates that the most significant ROI emerges from foundational back-office functions like procurement, finance, and operations. Enterprises are prioritising visible, top-line functions over high-ROI back-office opportunities. Your opportunity involves solving these critical, often less visible, problems with measurable cost savings.
- The “Buy vs. Build” Dilemma: Purchasing Solutions Yields Higher Success. This is a significant finding for your ventures. Organisations that purchased AI models and solutions experienced nearly twice the success rate (67%) compared to those that attempted to build their own internal systems (only a 33% success rate). Internal teams, despite deep business knowledge, often lack the extensive applied experience gained from numerous implementations across various industries. Your startup, as a specialized vendor, can provide this depth of experience.
- Lack of Integration & Cultural Friction. AI solutions are frequently implemented as “novelty add-ons” rather than being deeply integrated into core business systems, leading to fragmented data and conflicting signals. Furthermore, “shadow AI” is prevalent, with over 90% of employees using personal AI tools for work tasks, often outperforming formal initiatives. This underscores the critical need for solutions that are both effective and seamlessly integrated into existing workflows. Your product should function as a core operating system component, not a separate application.
- Poor Strategy & Ownership. Many projects are driven by “trend-chasing” instead of addressing clearly defined business problems. Success rates improve when organizations decentralise authority but maintain accountability, empowering line managers and front-line teams to drive adoption. Your solutions need to facilitate this distributed ownership.
The Winning Playbook for AI Startups
Assuming that the challenges are clear, is there anything startup founders can do to succeed with enterprise customers? Absolutely. But success requires more than technological innovation and it is more likely to work for some products and less for others. It requires a strategic approach to building, selling and deploying AI products with enterprise customers.
- Solve Specific, Measurable Business Problems: Shift focus from generic “AI initiatives” to defined business problems with measurable outcomes.
- Action: Target areas such as back-office automation (e.g., procurement, finance, operations) where the potential ROI is substantial, including reported savings of $2-10M annually in BPO elimination or a 30% reduction in agency spend.
- Build Learning-Capable, Adaptive Systems: Address the core “learning gap” by developing AI systems that retain feedback, adapt to context, and improve over time.
- Action: Your product must be “agentic” inherently embedding persistent memory and iterative learning. It should accumulate knowledge from interactions and autonomously orchestrate complex workflows.
- Prioritise Deep Integration, Not Just Add-ons: Ensure your AI solution is an integral part of the business’s operating system, not merely a superficial addition.
- Action: Design for deep integration into core enterprise systems such as ERP, CRM, and supply chain, to avoid fragmented data and enable meaningful decision influence.
- Embrace the “Buy” Mentality (as the Seller): Recognize that enterprises are more successful when purchasing solutions than when building them internally.
- Action: Position your startup as the specialised vendor possessing “10,000-hour knowledge”. Demonstrate a profound understanding of industry-specific workflows and offer tailored solutions.
- Empower Front-Line Managers & Address Cultural Change: Acknowledge that technological change involves cultural adaptation.
- Action: Design your product to be intuitive and flexible for “prosumers” (power users) on the front lines, who often drive bottom-up adoption. Provide clear training and demonstrate how your tool fits into or improves existing processes.
- Start Narrow, Scale Deep: Avoid attempting to solve all problems simultaneously.
- Action: Focus on achieving small, visible wins in narrow, non-critical workflows, then expand into core processes. Low setup burden and rapid time-to-value are crucial for initial adoption.
- Leverage Trust and Networks: Enterprise buyers are often skeptical of new vendors.
- Action: Emphasise channel partnerships, procurement referrals (e.g., from board members or advisors), and distribution through established enterprise marketplaces. Informal peer recommendations and existing vendor partnerships are powerful discovery mechanisms.
- Position for the Agentic Web: The future entails an “Agentic Web” of interoperable, autonomously coordinating systems.
- Action: Align with foundational frameworks like NANDA, Model Context Protocol (MCP), and Agent-to-Agent (A2A). This enables your agents to discover, negotiate, and coordinate across the internet infrastructure, fundamentally transforming business processes.
- Act Decisively – The Window is Closing: Enterprises are rapidly establishing vendor relationships for learning-capable tools.
- Action: Develop adaptive agents that learn from feedback, usage, and outcomes now to build durable product moats through data and integration depth. Once an enterprise trains a system to understand its workflows, switching costs become prohibitive.
Overall, the findings in the MIT report are more of a strategic roadmap rather than solely an indictment of AI. The observed 95% failure rate signals significant opportunities for your startups to build true, transformative value. By focusing on learning-capable systems, deep integration, and strategic partnerships, you can guide enterprises toward successful AI implementation.
I'm a former general partner at google ventures, head of Google for Entrepreneurs in Europe and founding head of Campus London, Google's first physical hub for startups.
I'm also the founder of Techbikers, a non-profit bringing together the startup ecosystem on cycling challenges in support of Room to Read. Since inception in 2012 we've built 11 schools and 50 libraries in the developing world.
- Weekly Firgun Newsletter – March 6 2026 - March 6, 2026
- Weekly Firgun Newsletter – February 27, 2026 - February 27, 2026
- The Rise of the Personal AI Agent - February 26, 2026
