A Unified Theory of Venture Investing
July 2025
Eric Rosenfeld
Just as physicists have searched for a “grand unified theory of everything” that brings every known force in the universe together into a single, all-encompassing equation, at the Oregon Venture Fund we’ve also been on the hunt for a unified theory of venture investing that ideally cuts across markets and business models.
Is there a single framework one could use for evaluating the mega-value-creating potential of a startup, whether it’s at the concept or growth stage? Whether a startup is pursuing a subscription, advertising, or transaction-based business model? Whether a startup is offering a consumer product, business-to-business service, or a science innovation? Whether the market opportunity be large or small? The goal would be to benefit from a single lens through which to ascertain whether a startup is more or less likely to generate a venture-type return. If life were only so simple and easy.
Early-stage venture investors are faced with too many unquantifiable variables and subjective judgements to reasonably evaluate in a short period of time. When you’re investing in people tasked with creating unknown new products that benefit other unknown people the reality is there’s going to be a lot of unknown unknowns. But still, one can’t help but think there must be something universal to look for in a startup outside of “great team,” “great product,” and “great potential market opportunity,” whatever “great” might mean to you.
If, like us, you share the belief popularized by Steve Blank and Eric Ries that startups are “learning organizations formed to search for a repeatable and scalable business model,” then the speed, cost, and quality of learning are foundational to a startup’s ability to adapt, evolve, and grow. Those startups best positioned to learn what a potential market really craves and can do so quickly, accurately, and cost-effectively should, in a Darwinian way, be able to achieve product-market-fit and grow faster than those who for whatever reason can’t learn quickly, accurately, and cost-effectively. Hence, we’re proud to introduce our best attempt at a grand unified theory of venture investing:
To be sure, there are many unknowable soft factors that also contribute to a startup’s ability to experiment, learn, and adapt. These include creating a culture of learning built on a set of repeatable practices and common norms, being transparent and honest with receiving and incorporating customer feedback, and taking every win and loss as a learning opportunity, to name a few. That said, we’ve found that there are a few objective factors to look for that can facilitate rapid learning feedback loops and therefore lead to earlier product-market fit and rapid scaling. These more objective and more identifiable factors revolve around the nature of the product, target market, and business model.
We’ve come to believe that startup success is less about how smart or insightful a founding team is and more about how quickly a set of founders can learn and incorporate what they’re learning. This is why software-centric startups have historically generated the bulk of venture capital returns and why AI, because of its ability to instantly and constantly learn, is powering the venture-backed startups of today and tomorrow.
With software, it’s easier to offer a minimum viable product and quickly and cheaply experiment with functions and features. In contrast, physical product and service companies usually require extensive planning and cost to develop and market test a new feature.
With software, it’s easier to receive customer feedback: “click here for help,” “click here to buy,” or “click here to tell us what you think.” The truly modern software in our phones and cars even knows what we’re doing without us having to ask or click. In contrast, product and service companies can be more challenging for getting honest, direct, instant feedback.
The power of AI is that product learning becomes instantaneous. With AI & ML, software can learn and evolve instantly and constantly without much, or any, human intervention. In terms of our grand unified theory, with AI the “velocity of learning” approaches the speed of light and the “cost of learning” approaches zero, dramatically increasing the probability of product-market-fit and success.
As a result, AI and data-centric startups, in general, can often learn and adapt much faster than a physical product or service company. Hydrolix, a streaming data lake platform designed to efficiently store, process, and analyze high-volume data, is an instructive example. At inception in 2018, Hydrolix focused on decoupling storage from compute to optimize scalability and cost. By 2022, Hydrolix had overcome several near-death experiences trying to balance customer pilots while struggling to modify the platform’s architecture in real time. In 2024, Hydrolix formalized a partnership with Akamai that dramatically increased the availability of Hydrolix to large-scale SaaS companies and the leading streaming services. As a result, Hydrolix has become the fastest-growing startup in the Pacific NW, growing ARR from just $3M in 2023 to $75M+ in 2025.
After working with hundreds of startups pursuing every imaginable market with every imaginable business model, we’ve noticed:
Business plans rarely survive first contact with customers. “Everybody has a plan until they get punched in the mouth.”
Only VCs, China, and business school professors demand 5-year plans to predict markets and revenue, which are ultimately unpredictable.
Start-ups are not just smaller versions of large companies. They do not unfold in accordance with a master plan. The startups that ultimately succeed go quickly from failure to failure, all the while adapting, iterating on, and improving their initial ideas as they continually learn from customers.
In short, established companies execute on a business model. Startups look for a business model that can support product-market fit and rapid scaling.
In contrast with AI and data-centric startups, physical product companies are often subject to several constraints…
A high cost of acquiring that first customer, and an even higher cost associated with getting the product wrong.
Longer and more expensive development cycles.
And, most importantly, longer customer feedback loops, and therefore longer learning cycles.
Our portfolio poster child for long feedback and learning loops was DesignMedix. A spinout of Portland State University, DesignMedix developed a drug with the potential to prevent and cure malaria, which afflicts 250 million people and kills over 600,000 annually. After years of development and testing, the drug was found to be safe and effective when used with mice and primates. Each animal trial took roughly a year to set up and execute before a learning result could be obtained. The first human trials at Duke University also took a year to plan and prepare for. During one of the clinical trials, a single human participant developed a systemic allergic reaction, with the cause unable to be identified. The trial was immediately suspended, the drug never made it to market, and, sadly, today there remains no safe and effective cure for malaria.
For DesignMedix, the quality of learning was uncertain and poor, the velocity of learning was slow, and the cost of learning was high. This helps explain why fewer than 50 new drugs are approved each year in the US and, despite the tremendous potential human and financial benefits, most VCs avoid investing in regulated therapies.
Founders of product companies work hard to find ways to shorten their learning cycles – e.g., using crowdsourcing services to market-test an idea and using flexible manufacturing technologies like 3D printing to create, and improve on, rapid prototypes.
What if a product just isn’t attractive to prospective customers [eg, the Segway Personal Transporter], or a market just isn’t sufficiently large or attractive [eg, membership in The Hot Dog of the Month Club]? The sooner - and cheaper - a startup and its investors can learn this, the better.
Channeling our inner Charlie Munger [“Invert, always invert!”], when could high-cost, long-duration learning be warranted and justified? It better be in service of an insanely large potential market opportunity – think space travel, or perhaps, preventing and curing malaria.
In sum, when evaluating a startup for a potential investment, a primary factor to consider is how conducive the product, service, and business model are to cost-effective, rapid, quality learning. Will the company have to build, measure, learn, and adapt over the course of a year (as companies with many “dependencies” and “permissions” must be do)? Or, can it build, measure, learn, and adapt over the lunch hour, as with a conventional software company? Or, better yet, do so automatically and continuously using AI?
This form of rapid learning and iterating is particularly well-suited and needed for fast-changing, high-risk markets, where research is difficult to conduct and customers themselves are unclear about their needs. Which pretty much describes the world we live in today.
The ability to observe and learn quickly and cheaply explains why the early bird may get the worm, but it’s the 2nd mouse that, after learning from the 1st mouse, gets the cheese.