Top 3 Mistakes Gen AI Companies Will Make

Intro

In the past 9 months, ChatGPT has inspired a new generation of companies focused on applying AI to new work streams and industries. However, companies like Cresta, Deepgram, and Gong have been building around GPT-2 and BERT since 2018. I joined Cresta as an early engineer and experienced the initial challenges and successes of building in this space. In this essay, we examine the mistakes that previous generations of applied generative AI companies made and that the new wave of applied generative AI companies should be on guard against.

Over-customization for Individual Customers

One of the most common mistakes made by applied generative AI companies is succumbing to the allure of initial deals, leading to an overemphasis on building custom-requested features for specific customers. It's easy to understand why teams might fall into this trap. These enterprise deals often range from $100,000 to well over a million dollars, and given the considerable sums at stake, they're often hotly contested, with specific requests made by the customer.

So what could these requests entail?

• Custom model training based on each customer’s individual data.

• Custom tooling to accommodate each customer’s unique workflow.

• Entire new lines of product that are tangentially adjacent to the company's existing product but not part of the core product/vision.

But what are the implications of accommodating these requests?

• Training or fine-tuning models based on each customer’s individual data can be costly, depending on the volume of data and the complexity of the desired goals. These costs can soar upwards of $10,000. This over-customization can inadvertently transform a company with SaaS multiples into a boutique, white-glove consulting service, which isn't attractive from a venture capital perspective.

• Spending cycles on building custom tooling/features can stretch the core team thin. This constant shift of focus can lead to half-baked products and continuous context-switching away from the core product, stifling innovation and growth.

It's crucial for companies to understand that while customer-centric solutions are essential, there's a delicate balance between building something for your customers and building for one customer. Instead of falling into the trap of over-customization, teams should aim for scalable solutions. These solutions should be designed to serve a broad range of customers while maintaining the flexibility to cater to specific needs, ensuring the company's sustainable growth and market competitiveness.

Misplaced Investment in Research Roles

Another common mistake is attempting both research and industry application in the early stages. The task of applying AI to various fields and industries is already an uphill battle. Often, these sectors are resistant to change, even when existing tools and methods are inefficient or suboptimal. To infiltrate into these industries, teams must invest focused effort into understanding these entrenched workflows.

Moreover, the rapid pace of AI development presents another layer of challenge. With tech giants and dedicated AI research companies like Anthropic and StabilityAI, keeping up with the latest advancements can be a Sisyphean task for a startup.

Attempting to divide attention between market application and research can lead to strategic ambiguities, ineffective resource allocation, and, ultimately, a slower growth trajectory. An alternative strategy is focus on application in the early stages. Once this foundation is laid, the company can incorporate a research-oriented aspect.

Need for Vertically Integrated Home Experience

Finally, many generative AI companies overlook the importance of creating a competitive moat. One way to do this is building a vertically integrated product experience, with applied AI. Every doctor navigates patient care through Epic. Every designer brings ideas to life in Figma. And every trader navigates markets with Bloomberg. Similarly, the goal of every applied AI company should be creating the go-to platform for that industry.

Teams fall into the trap of building into the existing workflows of their customer. For example, a team building a content generation tool for marketing teams might integrate their product with Microsoft Word. This is a good start, but it doesn't go far enough. To create a truly competitive moat, companies need to build vertically integrated product experiences that encompass the entire customer journey. Thus, the goal should be taking the customer out of Microsoft Word.

Conclusion:

In conclusion, the road to success for generative AI companies is fraught with challenges. These can be navigated effectively by avoiding the trap of over-customization for individual customers, striking a balance between market application and research, and focusing on creating a vertically integrated home experience. The future of applied AI is incredibly exciting, and with a thoughtful and strategic approach, these companies can lead the charge towards that future.

If you have any feedback, questions, or are looking to connect, please comment below or my Twitter DMs are open.