The Oregon Trail to AI Defensibility
The Oregon Trail to AI Defensibility
5 VC firms share their survival guides
This post is brought to you by Spiky.ai, a video analysis platform for revenue teams. Their AI models add emotional context to better understand your customers & coach your sales team.
Can AI Startups Get to Defensibility?
AI is the next platform gold rush. But, will AI pioneers survive the Oregon Trail to defensibility?
The journey will be challenging with plagues of high valuations, limited defenses against the elements, and perpetual attacks by new technology developments.
The space is moving so fast and many startups look like wrappers around OpenAI.
What are the keys to survival?
What makes an AI company defensible and durable in the long run?
Where will value accumulate?
NFX, a VC firm that focuses on investing in companies that have network efforts, recently wrote a post, “AI is Like Water.”
Water is three things. Necessary. Ubiquitous. And the same f***ing. thing. inside every bottle.
It’s strange to compare the hottest, world-changing technology to something as simple as water. However, the ubiquity of it is valid.
The article argues that tech differentiation is harder to find and moats are shrinking, or “heading to zero.”
If the underlying tech is no longer a differentiator, then you need to start figuring out how you can build a durable business.
To answer that question, below are 5 must-read articles and key takeaways:
AI Defensibility Musings from 5 VC Firms
1. CRV - How to Build a Defensible AI Startup in 2023
We believe the most delightful innovations in AI are coming from the bottom up, not the top down, as companies like Midjourney, Runway, and Stable Diffusion have captured the public imagination. Open source horizontal models like ChatGPT will form the foundation of the AI boom, but many more venture-scale businesses will be built on the application layer.
Key Takeaways:
Ship quickly and prioritize ruthlessly
Build novel, best-in-class interfaces
Build community
“Successful products in the AI space can start out looking like consumer companies.”
Go vertical
Startups can win initially by moving quickly, but long-term moats are built by going where big tech and incumbents won’t. In a world where anyone can access LLMs, verticalized solutions based on a deep understanding of a target persona is one of the best defensible ways to build in AI.
The only truly defensible items in the application layer are workflow, product, community, and rapid iteration.
2. Greylock - The New New Moats
Why does it feel like there are “no more moats” to build? In an era of cloud and open source, deep technology attacking hard problems is becoming a shallower moat. The use of open source is making it harder to monetize technology advances while the use of cloud to deliver technology is moving defensibility to different parts of the product. Companies that focus too much on technology without putting it in context of a customer problem will be caught between a rock and a hard place — or as I like to say, “between open source and a cloud place.”
Key Takeaways:
Defensible business models can still be built around IP (these are foundation models)
There are “Systems of Record” (CRM, HCM, ERP), “Systems of Engagement” (Slack, Amazon Alexa), and “Systems of Intelligence” (combining data sets and providing insights)
3 major areas where you can build a “System of Intelligence”
Customer-facing applications
Employee-facing applications (HCM, ITSM, etc.)
Infrastructure systems (security, compute, monitoring, etc.)
The battle is moving from the old moats (the sources of data) to the new moats (what you do with the data).
What makes a system of intelligence valuable is that it typically crosses multiple datasets and multiple systems of record. One example is an application that combines web analytics with customer data and social data to predict end user behavior, churn, LTV, or just serve more timely content. You can build intelligence on a single data source or single system of record, but that position becomes harder to defend against the vendor that owns the data.
For a startup to thrive around incumbents like Oracle and SAP, you need to combine their data with other data sources (public or private) to create value for your customer. Incumbents will be advantaged on their own data. For example, Salesforce is building a system of intelligence, Einstein, starting with their own system of record, CRM.
3. NFX - The AI Startup Litmus Test
We were recently at a demo day where three companies sounded promising – then we looked at the open source libraries that they had just built skins on top of. Right now it can be tricky to know what is hard, unique, and defensible – versus a weekend project built on top of an open source project. Founders and investors, especially during a moment of exponential growth and consensus, need to run through a few litmus test questions when they are evaluating the next generative AI company.
Key Takeaways:
If you took out references to AI in your pitch deck, is this still a good business?
Businesses heavily dependent on generative AI are easily copyable. An unparalleled user experience; however, can help you here.
Unless you have billions of dollars, it’s going to be incredibly hard to stay on the cutting edge of AI research. This means you need to focus your time on what can be built on top of the foundation models.
The Litmus Test Questions:
4. Sapphire - Vertical(ai) is the New Horizontal
As we consider where to invest, we’re having debates internally and with founders about how companies build moats and defensibility when many are using the same underlying foundation models.
Key Takeaways:
Defensibility comes from having access to proprietary data sets that are difficult to replicate. Flywheels can help accumulate more & better data from customers that can be used to fine-tune those specific LLMs. It’s easiest to do this with vertical software.
Law and healthcare AI startups are good examples of companies leveraging “hard-to-reach” data sources. Sector-specific data is key.
Vertical software in general faces less competition
5. Wing - Leveraging Large Language Models to build a defensible startup
If you’re a startup building AI-native applications with LLMs, there are a few considerations that broadly relate to product approach. In general, you’ll want to think about how to build defensibility in relationship to the models and incumbent applications.
In this article, we’ll take a deep dive into these primary concerns, and discuss how you can strategically build your competitive advantage.
Key Takeaways:
Need to provide enough value on top of the model layer so that you aren’t commoditized.
At a high level, the range of options for startups is (from easiest to hardest):
Prompt engineering only: Focus on improving the model output by engineering the prompts used in the models, and by potentially selecting different vendors for various prompts.
Fine-tuning: Improve the model by fine-tuning it with feedback and input/output data from a dataset or real usage.
Train your own models: Train highly specialized models for specific use cases using all the data collected from the application in production.
Incorporating private data (or simply unique data from customers) will help to increase switching costs
Most incumbents will be adding AI, so your solution needs to be more than than just a feature. It’s gotta go deeper and use AI better in that category
4 Approaches to building defensibility:
Add AI value on top of the LLMs for your use case through prompt engineering and fine-tuning — but think about whether that value will still matter once the models improve.
Incorporate private data/customer data in the model context to improve outputs.
Assume that incumbents in your space will adopt surface-level generative AI features and think about how you can go beyond their ideas.
Think about the right insertion point for your product and try to go deep into workflows while minimizing disruptions — while still focusing on bringing out the full value of AI.
Summary of Strategies
👉 Go vertical - know the domain better than anyone else
👉 Private/customer/hard-to-reach data + data capture flywheels for unique data
👉 Build community + network effects
👉 Build systems of intelligence (incorporating multiple datasets from different sources for value delivery)
👉 Killer UI/UX experience
👉 Train your own models
You can also Follow Me on Linkedin.