AI Entrepreneurs: Strategies for Fortifying Your Competitive Moat
How to build a generational AI company that will survive the test of time
“How will you build an economic moat?” – investors prod AI entrepreneurs with this question incessantly. Economic moats became a theme at the New York AI Summit and throughout the pitching competition. Many investors and entrepreneurs had different perspectives; to be honest, I was humbled by the quality of insights. The concern is valid. In an increasingly competitive world where the barrier to entry is getting lower and lower, how will your company survive the test of time?
When OpenAI released ChatGPT, there was an outpouring of AI builders who piggybacked off it. Coined as “wrappers”, they built their own GPTs with specific training and some smart prompt engineering. OpenAI flipped the script completely when they released the ChatGPT App Store. Now anyone with zero coding or ML experience can make an AI enabled application. This new feature made several wrappers obsolete, highlighting their lack of moat. How do you justify your value proposition to customers when they can get a similar copy for free? Even for OpenAI, the moat is dubious. Open-source foundational models are catching up quickly, at a fraction of parameters and thus cost. In the chart below, you can see that several foundational models are performing similarly to GPT. Alpaca (developed by Stanford) cost only $600 to develop compared to GPT-4 at $100MM+.
Relative Response Quality Assessed by GPT-4
However, I argue that it is possible to build a sustainable economic moat that makes it difficult for competitors to erode a company’s market share, profitability, and overall success. The good news is that moats have not changed, even for companies building in AI. They can take on various forms, such as but not limited to: intellectual property, trade secrets, pace of innovation, brand, and quality. However, in the AI space, I’ve found that value accrues in favour of the top 1-3 players. I dive into this in detail below.
Only Players of Scale Eats
As you can see from the chart above, the top 5ish foundational models are the highest in price and are generally correlated with parameters. Premiums rapidly fall off on a dollar per token basis, as there is a long tail of other models that are low cost or free. Customers in the OpenAI era have abundant choice, but value only accrues to the top. I personally use ChatGPT, Perplexity and Google for search, but I still don't know ANYONE that uses AskJeeves or Bing. This type of “Winner Takes All” market dynamics can be seen in:
Social Media: Facebook (including its subsidiary Instagram) and YouTube (owned by Google) due to network effects
eCommerce: Amazon is the dominant player due to economies of scale
Ride Share: Uber and Lyft due to both economies of scale and network effects.
Operating Systems: split between Microsoft, Android and iOS
Winner takes all markets are particularly important in trendy spaces such as AI applications, where new startups emerge every week and run the risk of commoditizing each other. If it is so important to be one of the top players that stands out in a noisy environment, then how do you get there?
How to Become #1
Many investors I’ve spoken to have come to consensus that that a data flywheel is the only defensible path to create a sustainable competitive advantage. A data flywheel is when proprietary data allow entrepreneurs to build the best possible product, attract more users than the competition, and thus earn even more valuable data to further fine tune their product. This is obviously a strong proposition, but it is not the only way to build a moat. Data is actually becoming less of a competitive advantage. Companies with similar or incomplete data can build a similar product. It is not difficult (or illegal) to reverse engineer a model with synthetic data and achieve similar outcomes as we’ve seen with Alpaca.
Take the example of all the value that was created in the golden age of software beginning in the dot-com boom of the late 1990s. Over a span of several decades, software development was suddenly accessible to anyone with proficient coding skills. However, the availability of skills and tools didn’t mean that there would be thousands of CRMs with similar capabilities. Today, there are only a few widely used CRMs such as Salesforce, and some industry specific ones such as Affinity (CRM for VCs). Fast forward to the AI era now, anyone with a proficient team may set up a company that leverages a foundational model, and builds a customized solution for a particular end market. Just because the tools are there and it is easier to build, doesn’t mean the market will be saturated with free and highly functional tools.
Back to the Basics: Delighting the Customer
Build products that customers can’t live without. Affinity didn’t become a huge success because they were the most affordable or the most functional. Over time, they’ve built a great brand and that became their competitive advantage. Affinity of course can be copied, but it would take significant resources and time. Affinity won the market for investment managers because they focused on building products that customers LOVE. This is the basics of a software product, and of most successful companies in general. You should be:
Customer Focused: most of the decisions developers make regarding the technology stack are unlikely to be apparent to end users. What will significantly shape the company's path to success is the effort to enhance customers' experiences, making them relevant, rewarding, and enjoyable.
Responsive to Feedback: incorporate information captured from customer service and marketing/sales that feeds into product design
Showcasing ROI: clear, instant, and positive return on investment for your customer. Whether that means time saved, cost saved, or additional revenue generated
Interchangeable: given the rapid pace of evolution, it is important to make your application platform agnostic. GPT may not be the winner in a few years, and it is important that your company can adapt.
Sticky: increase switching costs through customization, integrations with other tools and data lock-ins
This list above is coincidentally, the SAME playbook for software companies looking to build a moat. This is because generative AI applications ARE software. Unicorn companies such as Slack simply built a better version of their competition which in Slack’s case, they built a better version of a messenger application. The fundamentals of business building apply in the same way to AI applications. Just because the barrier to entry is lower, does not mean that the industry is any less valuable. We are in the early innings of a new AI era, and I’m excited about what’s to come. I genuinely would like to hear from you on the topic of economic moats for AI and I hope to meet more entrepreneurs building in the space.