Defensibility in AI Applications

I have seen many AI startups in the past few months, and we need to talk more about defensibility. While it is great to see rapid innovation in ML and AI, I urge entrepreneurs and startups to consider defensibility further.

What is defensibility?

Defensibility is demonstrating high barrier to entry against other upstarts and incumbents for a clear and valuable use case. In the past, having a capable team of data scientists for an interesting problem is a good start in building the high barrier to entry, but this is not enough anymore.

Here are some ways to consider defensibility when building modern ML/AI applications:

• The use case should not be a feature that a big incumbent can build as a natural extension of a core product.
• The team of data scientists bring domain expertise and business acumen to their roles and have the the capacity to contribute to pre-sale, field engineering, and post-sale.
• The company has ownership and/or clear rights to use proprietary data and has differentiated insights into alternative data that can augment existing proprietary and open data.
• The company can demonstrate clear technical model evaluation that both technical and business stakeholders can understand and appreciate against a widely acceptable baseline.
• The human evaluation process is robust in most scenarios where value is accrued.
• The feedback loop can provide highly valuable insights back to the business.

Admittedly, it is hard to create defensibility in the application layer but it can be done, especially now as maturity in considering the use of ML/AI progresses. The AI middleware layer is another exciting and emerging area, and the large incumbents have an opportunity to lead now through organic product innovation or platform acquisition plays.