The push to adopt artificial intelligence has left many organisations weighing a familiar question: should they build AI capabilities internally or work with external partners? Recent research from MIT NANDA offers a clear perspective, particularly for leadership teams under pressure to deliver results quickly and responsibly.

According to the State of AI in Business 2025 report, organisations that form strategic partnerships with external vendors achieve AI deployment success rates that are twice as high as those relying solely on internal development. These partnerships also tend to deliver faster implementation, lower overall cost, and closer alignment with day-to-day operations.

The findings challenge a long-standing assumption that in-house development always produces better outcomes. While building internally offers a high degree of control, the data suggests that, in practice, partnerships often lead to more reliable and scalable results.

Speed Over Scale

Mid-market organisations are currently leading in AI deployment speed. When working with the right partners, they move from pilot to full implementation in an average of 90 days. By contrast, enterprise organisations building internally typically require nine months or more to reach the same point. This gap is driven less by organisational size and more by execution approach.

Evaluating AI solutions requires time, technical understanding, and experience that most organisations cannot easily build internally.

“We’ve seen dozens of demos this year. Maybe one or two are genuinely useful. The rest are wrappers or science projects.” Chief Information Officer, State of AI in Business 2025 Report, MIT

External partners bring that experience with them. They arrive having already worked through technical challenges, implementation hurdles, and operational integration across multiple clients. This allows organisations to adopt solutions that have already been tested in real environments, rather than starting from a blank page.

Where The Value Lies

The research shows that some of the most meaningful returns on AI investment are coming not from customer-facing functions, but from back-office operations. Strategic partnerships focused on customer service, document processing, and internal workflows are delivering annual savings of between £2 million and £10 million.

Much of this value comes from reducing external spend. Organisations are eliminating business process outsourcing contracts, cutting agency fees by around 30%, and reducing reliance on expensive consultants by replacing manual processes with AI-supported internal capability.

In financial services, the report highlights organisations saving around £1 million per year by automating outsourced risk management activities. These are realised savings, not projections, and they flow directly to operating margins.

Closing the learning gap

A core challenge identified in the research is that many AI systems fail to improve over time. They require ongoing manual input and do not adapt meaningfully based on use or feedback. This “learning gap” helps explain why 95% of custom enterprise AI tools never make it into production, despite significant investment.

Strategic partners address this by designing systems that learn continuously. These platforms retain context, adapt to specific workflows, and improve through repeated use. Rather than offering generic tools with limited flexibility, the strongest vendors focus on deep integration with how organisations actually work.

A corporate lawyer interviewed in the study described the difference simply: “Our purchased AI tool produced rigid summaries with little room for adjustment. With better systems, I can guide the process and refine outputs until they meet our needs.”

Why Trust Matters

The research also sheds light on how successful AI partnerships are formed. Technical features alone rarely drive adoption. Instead, trust plays a decisive role. 20% of successful implementations began through existing vendor relationships while a further 15% came via partner referrals, and 13% through informal peer recommendations. By comparison, conferences and industry publications accounted for just 15% combined.

For organisations at an early stage, this points to a sensible starting point. Look first to vendors already trusted in other areas, speak with peers who have deployed AI successfully, and work with partners who understand your industry context. Technical capability matters, but it is rarely the sole factor in long-term success.

Employees Lead the Way

One of the most encouraging findings in the report concerns employee behaviour. While formal enterprise AI initiatives often move slowly, individual employees are already using AI tools effectively. More than 90% of employees surveyed reported regular AI use, even though only 40% of their organisations had purchased official subscriptions.

These employees, often described as “prosumers,” have already learned what works through everyday use. Forward-looking organisations are recognising this and allowing these users to help identify practical use cases and suitable tools.

Rather than relying entirely on centralised AI teams, successful organisations empower budget holders and domain leaders to surface problems and lead implementation, while maintaining executive oversight. This approach supports faster adoption without sacrificing control or alignment.

What successful partnerships have in common

The research identifies three shared characteristics among organisations that deploy AI successfully through external partnerships.

First, they insist on meaningful customisation. Solutions are adapted to internal data and workflows rather than forced into generic templates.

Second, they assess success based on operational outcomes, not technical specifications. Improvements in efficiency, cost reduction, and service quality matter more than model performance metrics.

Third, they treat deployment as a collaborative process. Early challenges are expected and addressed jointly. These organisations recognise that implementing AI is closer to an ongoing advisory relationship than a one-off software purchase.

Why Timing Matters

While the findings are encouraging, the report also highlights a growing urgency. AI systems that learn over time create increasing switching costs. Once a platform has absorbed an organisation’s data, processes, and preferences, replacing it becomes complex and costly.

Over the next 18 months, many organisations will commit to partnerships that shape their AI capabilities for years. Early movers have the opportunity to build durable advantages, while those who delay may find themselves competing against systems with far deeper institutional knowledge.

Frameworks such as Model Context Protocol, Agent-to-Agent coordination, and NANDA are already enabling this next generation of adaptive systems. These capabilities are available now, not theoretical.

Where to Start

For leadership teams uncertain where to begin, the research offers a clear approach. Start with narrow, high-value use cases in back-office functions where returns are easiest to measure. Seek partners who demonstrate genuine understanding of your workflows. Rely on trusted referrals. Measure success by business impact rather than technical sophistication.

Above all, involve employees who are already using AI tools effectively. They often see opportunities that formal programmes overlook.

The aim is not to build the most advanced AI infrastructure in the market. It is to deliver measurable value quickly, then build from there. Mid-market organisations are already showing that this approach works, often moving faster than larger peers.

Building and buying do not need to be mutually exclusive. Many organisations will use both over time. But the evidence is clear: for most, strategic partnerships provide a faster and more reliable route to successful AI adoption.