Why 95% of enterprise AI pilots fail — and what MIT's research actually shows.

MIT's State of AI in Business 2025 report made headlines for a single figure: 95% of enterprise generative-AI pilots fail to produce measurable results. The headline is eye-catching. The detail is what matters for anyone actually trying to integrate AI into a business.

The headline finding

From the report: "…for 95% of companies in the dataset, generative AI implementation is falling short. The core issue? Not the quality of the AI models, but the 'learning gap' for both tools and organizations."

Executives often blame regulation or model quality. MIT's data points somewhere else — at enterprise integration. Generic consumer tools like ChatGPT excel for individuals because of their flexibility, but they stall in enterprise use because they don't learn from or adapt to specific workflows.

Budgets are going to the wrong places

The report also found a misalignment in resource allocation. More than half of generative-AI budgets are devoted to sales and marketing tools — yet MIT found the biggest ROI in back-office automation: eliminating business-process outsourcing, cutting external agency costs, and streamlining internal operations.

This is consistent with what we see in client engagements. The highest-leverage applications are rarely the flashy ones. They're the ones that replace recurring, expensive, externally-procured work with AI-assisted processes the business controls directly.

Build vs. buy — the single most important variable

The finding most relevant to this business: purchasing AI tools from specialized vendors and building partnerships succeed about 67% of the time, while internal builds succeed only one-third as often.

This is particularly relevant in financial services and other highly regulated sectors, where many firms are attempting to build proprietary generative-AI systems in-house from first principles. MIT's research suggests that going solo leads to far more failures.

The nuance worth flagging: "specialized vendor" is not the same as "public-cloud AI provider." The vendors MIT is describing are integrators — organizations that understand the client's domain, bring a reference architecture, and own the integration work. That's the posture that succeeds, not the posture of handing the problem to a generic platform.

What we take from this

Three things, consistent with how we already work:

• Integration is the product. The model is a component. The value is in how that component is wired into the organization's actual workflows, data, and governance.
• Back-office first. Where the numbers move, the wins are usually unglamorous — document processing, routine correspondence, knowledge retrieval, internal research.
• Specialization beats DIY. The 67%-vs-33% gap is not small. Firms that try to build AI capability from scratch, without someone in the room who has done it before, do markedly worse than firms that partner.

That last point is, frankly, why we exist. We're a small, regional integrator focused on private, in-house AI for SMBs in Southern California. We're not a platform, we're not a model provider — we're the people who show up, understand the business, and wire the capability into what the organization already runs.

Source: MIT NANDA, The GenAI Divide: State of AI in Business 2025. Reporting attributed to MIT researcher Aditya Challapally.