Why 95% of GenAI Projects Fail and How Enterprises Are Getting It Right

According to a recent report from MIT’s Project NANDA (short for Networked AI Agents in Decentralized Architecture), enterprises have invested $30-40 billion in AI initiatives focused on generative AI. The only problem is, 95% of them aren’t getting any return. 

Is AI not as disruptive as we think? According to the MIT report, the answer is more nuanced. While 95% of generative AI initiatives are falling short, it’s mainly due to poor enterprise integration, training, and learning gaps – not the technology itself. 

The report from Project NANDA suggests that while platforms like ChatGPT and CoPilot have been widely adopted, they mostly increase productivity at the individual level. True enterprise integration (the 5% of initiatives that generated an outsized return, in terms of boosted revenue or reduced cost) required real learning, implemented on an organizational scale. So what does this mean for organizations? 

Build vs. Buy: Generative AI at Enterprise Scale

After conducting over 150 executive interviews from 300 public AI initiatives, the MIT researchers determined that, to be valuable, generative AI for enterprise needed to retain feedback, adapt to context, and improve with time – and according to the MIT report, “a small group of vendors and buyers are achieving faster progress by addressing these limitations directly.” Implementations with external vendors such as buying a product already developed had twice the success rate of internal builds. 

AI models are costly to build from scratch, and the sheer amount of training data required to create a model is out of reach for even some large organizations. Investment in Bloomberg’s AI, for example, came in well over $10M, despite having a large repository of financial documents and other training data already available (it’s Bloomberg after all!). For many organizations in claims, buying a generative AI model that is already built, with plenty of flexibility and custom components, is a much faster, more effective way. 

When enterprises use ready-built AI platforms designed for its industry, this feedback, context, and improvement is built right in. And, according to the Project NANDA team, this is how the highest performing organizations won. External tools help modernize back-office operations and bring workflows up to speed – resulting in a much more tangible return. 

Building on a Strong Foundation

According to Project NANDA, 60% of organizations evaluated enterprise-grade AI tools, but only 20% reached the pilot stage and only 5% reached production. Enterprises led the pack in pilot volume, but lagged at reaching scale. Of these enterprises, investment in visible, top-line functions took priority – but back office tasks had the highest ROI. 

For industries like insurance and legal tech, the stakes are often higher. New regulations (and a changing regulatory environment) mean extra hurdles when creating or using generative AI. Implementing AI successfully isn’t just about building from the ground up, especially in insurance, legal tech, or claims. Today’s generative AI platforms can be customized to suit the needs of your company and industry, but using a tool that is both compliant and human supervised can save you plenty of time. 

While generative AI hasn’t been the ‘magic pill’ some industries were hoping for, the 5% of generative AI adopters in the study saw outsized returns. These use cases present lots of interesting data on when, where, and why GenAI can be used. Partnerships with external vendors allow enterprises to make the most of ready-made AI tools: ones that are compliant, well integrated, and make a positive difference on your P&L. 

To learn how to join the top 5% of organizations successfully implementing AI at scale, download our Enterprise Claims Transformation Guide today and gain a complete playbook for achieving a competitive advantage through human-assisted intelligence.