March 4, 2024


with the rise Open source AI models, commoditization of this breakthrough technology is coming. It’s easy to fall into the trap of targeting a newly released model at the ideal tech crowd and hoping it catches on.

Creating a moat when so many models are easily accessible presents a dilemma for early-stage AI startups, but leveraging deep relationships with customers in your field is a simple and effective strategy.

The real moat is the combination of AI models trained on proprietary data, and a deep understanding of how experts approach everyday tasks to solve subtle workflow problems.

In a highly regulated industry where outcomes have real-world impact, data storage must pass high standards of compliance checks. Often, customers prefer companies with a track record over startups, promoting an industry with fragmented data sets where no single player has access to all the data. Today we have a multimodal reality where players of all sizes hold datasets behind highly compatible walled garden servers.

This creates an opportunity for startups with existing relationships to reach potential customers who would typically outsource their technology to launch test pilots with their software to solve specific customer problems. These relationships may have arisen through co-founders, investors, advisors, or even previous professional networks.

The real moat is the combination of AI models trained on proprietary data, and a deep understanding of how experts approach everyday tasks to solve subtle workflow problems.

Presenting irrelevant credentials to customers is an effective way to build trust: Positive indicators include team members from a university known for AI experts, a robust presentation where a prototype enables potential customers to visualize results, or an understanding of how your solution will A clear business case analysis of how you can help them save money or make money.

A common mistake founders make at this stage is assuming that building a customer data model is sufficient for product-market fit and differentiation. In reality, finding a PMF is much more complicated: simply using AI to solve problems can lead to problems with accuracy and customer acceptance.

Adding experienced specialists with complex knowledge of day-to-day changes in a highly regulated industry often proves to be a daunting task. Even AI models trained on data may lack the accuracy and nuance of expert domain knowledge, or more importantly, lack any connection to reality.

A risk detection system trained on a decade of data may not be aware of industry expert conversations or recent news that could render widgets previously considered “risky” completely harmless. Another example might be a coding assistant suggesting code completion for a previous version of a front-end framework that has separately benefited from a series of high-frequency breaking feature releases.

In these types of situations, startups are best off relying on a model of launching and iterating, even on a pilot basis.

Three key strategies for managing pilots: