This post delves into the world of large language models (LLMs) and Generative AI, exploring their impact on the startup ecosystem. As the Founder and CEO of Zendog Labs, I have witnessed the escalating hype surrounding LLMs and Generative AI, and the crucial questions that arise for founders and investors alike. In this post, we aim to discern whether LLMs are mere overhyped fads or genuine game-changers for businesses. So, let’s dive in!
The Power of Large Language Models
LLMs, including OpenAI’s ChatGPT, have transformed the startup ecosystem by providing a new foundational infrastructure. Existing startups increasingly incorporate them into their operations. And, we’ve witnessed the emergence of new startups solely focused on LLM-based products.
The allure of LLMs lies in their ability to streamline business processes and optimize resource utilization. Startups can achieve more with fewer resources. Startups can also enhance their product offerings, providing solutions like chatbots and image generators to target markets.
The Startup Perspective
Significant Acceleration for Startups
Startups that embrace LLM-based capabilities by pivoting from their legacy business models often experience remarkable acceleration. We have witnessed this with multiple startups across various use cases, such as travel planning, mental health, and content generation, that rapidly grew after integrating LLM technology. We hypothesize that both B2B and B2C customers are actively exploring different LLM-based applications, leading to organic growth and reduced customer acquisition costs for startups and other businesses undergoing this transition.
The Rise of LLM-Based Startups
Additionally, there has been a surge in startups that are built entirely around LLMs. We have observed the launch announcements of these startups go viral on social media, generating widespread adoption. Remarkably, some of our customers have achieved millions of views, tens of thousands in revenue, and hundreds of thousands of users from a single post on a social media platform. This success is further amplified by encouraging engagement, such as providing discount codes for the first 1,000 commenters.
The Dark Side of LLMs
However, it’s crucial to acknowledge that not all LLM-based ventures have positive implications. One trend we’ve noticed is influencers creating digital avatars for chat interactions, where individuals pay by the minute to engage in conversations with a fake persona. We have seen instances of this model being particularly popular with the OnlyFans community. While these ventures may generate substantial profits, we believe there is a need for regulatory intervention to address excessive value extraction. As startup enthusiasts, we firmly believe that ventures should strive to serve a greater purpose.
Investor Perspectives and Key Questions
With backgrounds in venture capital, we are adept at navigating both the founder and investor landscapes. This enables us to guide founders in decoding the investor psyche. Investors interested in Generative AI products and startups capitalizing on LLMs need answers to three crucial questions: discerning “must-haves” from “nice-to-haves”, bolstering unit economics, and distinguishing enduring, defensible products from commodities. For startups to grab the attention of investors, it’s important that they find good responses to these three questions.
Nice-to-Have vs. Must-Have
Investors meticulously assess the long-term viability of LLM-based products, particularly evaluating whether a product falls into the category of “nice-to-have” or “must-have” solutions. Nice-to-have products are at risk of being easily replaced or losing value rapidly, resulting in high churn rates and low customer lifetime values. Conversely, must-have solutions are deeply integrated into critical business processes, making them challenging or even impossible to replace. This is exemplified by companies like SAP, which wield considerable pricing power due to the indispensable nature of their offerings. For founders, it is crucial to identify avenues for building must-have solutions that are not easily replaceable. While other products may generate short-term cash influxes, they lack the potential to become the enduring revenue engines that investors seek.
Unit Economics: The Challenge of High Costs
LLM-based products pose a significant challenge due to their high computational costs. Startups often rely on third-party infrastructure providers like OpenAI, which adds to the expenses. As a result, unit economics suffer, and many startups in the Generative AI space struggle with gross margins between 20-30%. This is particularly concerning since software businesses typically enjoy gross margins of 70-90% or even higher. The responsibility of optimizing computational resource utilization lies with the CTOs and technical teams, and investors are keenly interested in the startup’s technical roadmap for improving unit economics.
Defensibility: Navigating Vulnerabilities
Building a product on a third-party infrastructure, like OpenAI, gives rise to concerns regarding the defensibility of the offering. The accessibility of OpenAI’s API has significantly simplified the process of developing LLM-based products, to the point where freelancers on platforms like Fiverr can offer to build such products for as little as $5,000. This ease of entry raises questions about LLM-based ventures’ uniqueness and long-term viability in an increasingly competitive landscape.
Strategies for Success
Having explored the appeal of LLM-based products for startups and the key questions investors pose regarding their viability, it’s time to delve into initial strategies that startups have adopted to build more defensible businesses with healthy gross margins and long-term customer loyalty.
Collecting Proprietary Data
First, startups prioritize collecting proprietary data from consumers and customers in a manner that aligns with data privacy regulations. This data serves as a unique and coveted resource for the business, appealing to investors seeking defensible ventures. By owning, leveraging, and potentially selling this data, startups can optimize models and unlock new revenue streams.
Building Proprietary Models
Another key strategy observed among startups is the development of proprietary models tailored to specific use cases, particularly in the transition from text-based to visual content. By optimizing existing LLM models to suit their unique products, startups can reduce data requirements and computational costs effectively. The ability to fine-tune LLM models for specific use cases empowers startups to deliver differentiated and value-added offerings to their customers, setting them apart from competitors and reinforcing their position in the market.
Establishing Feedback Loops
Establishing robust feedback loops with users is a vital strategy for startups to continuously collect data and optimize their models. Startups that actively solicit and track user feedback create powerful data circuits that can prove a significant competitive advantage, deterring late entrants. This doesn’t need to be an intricate process. It can be as straightforward as encouraging users to rate the effectiveness of your product’s results with a thumbs up or down.
Embracing Usage-Based Pricing Models
A fourth and increasingly prevalent strategy is the shift towards “pay as you go” or usage-based pricing models. As computational costs continue to pose significant challenges, it becomes less viable to rely on the fixed monthly payments typical of traditional SaaS pricing models. The allure of SaaS models, namely the stability of cash flows, is certainly enticing for investors. However, in the absence of substantial technological innovations that reduce computational costs, it appears inevitable that we will witness a broader shift towards usage-based pricing, which would “end” the current world as we know it.
In conclusion, LLMs have the potential to be game changers for startups. Startups have experienced significant acceleration by leveraging LLMs. However, founders must carefully evaluate whether their product is a must-have and address challenges such as unit economics and defensibility. Strategies like collecting proprietary data, building proprietary models, establishing feedback loops, and embracing usage-based pricing models can enhance a startup’s chances of success.
About the Author
Rafael is an experienced operator and investor with a background in building startups and scaling businesses. With over 10 years of experience, Rafael has built seven startups, including as a Principal at Roivant and Co-Founder at Eatearnity and Maison Baum. Rafael holds degrees from prestigious institutions such as the LSE, Duke, St. Gallen, and Boğaziçi.
About Zendog Labs
At Zendog Labs, we are passionate about supercharging growth for startups and enterprises. We help startups achieve product-market fit, accelerate revenue, and secure capital, and our Venture as a Service leverages the startup game for enterprises.