157 - Startup Valuation

One method that Daniel Kahneman discovered to be useful for reducing cognitive bias in decision-making processes is the direct comparison between 2 or more options across a predetermined set of criteria. This process is further aided by putting each evaluation into words, as well as a score. Applying this method, I built out an example of how to do this back in 2022 when I was first trying to calculate what our team's startup was actually worth, as we began approaching many investors.

*Note: All methods investors use for such value estimates are ultimately subjective, whether due to factors such as survivorship bias, substitution biases, confirmation bias, selective attention bias, or several dozen other relevant biases. What differs between the methods is whether or not they serve to reduce biases hanging over the subject matter, or if they simply pick their favorite biases and mix them into a cocktail. Getting investors "drunk" on cognitive biases is generally what frauds aim for.

As I mentioned previously, weak correlative proxies for trust, like publishing in prestigious journals, being an Ex-employee from a major firm, and gaming a few benchmarks are all strongly correlated with bad actors today, making them inversely correlated with whom anyone should trust in the specific condition of startup founders. So this begs the question, what actually matters?

The factors I selected for comparison between startups were:

  • Hype

  • Potential Impact: How can they change the world?

  • Value Proposition: What value do they offer?

  • Cost-to-Duplicate: Do they have a "moat"?

  • Milestones Achieved: What have they accomplished?

  • Progress Since Last Raise: What did they do with the funding?

  • Efficacy: How effective is it?

  • Profitability: Profit margins

  • Established Methods: Degree of research remaining

For this particular comparison, I kept it simple, on a 0-100% score, in increments of 5%, with no negative numbers. For some comparisons, like Efficacy, a more literal comparison might have required several orders of magnitude to be faithfully represented. For example, the 100x hardware efficiency and 10,000x data efficiency of Norn couldn't be represented in this format. However, this comparison must remain fairly simple to serve the intended purpose of reducing cognitive biases.

If investment firms established similar processes they could also more easily compare between successful and failed startups, discerning treasure from trash in the deluge of bad ideas they receive today. The methodologies I've observed VCs, angels, and other investors utilizing these past two years have been extremely underwhelming, and frequently quite idiotic. None of the firms I've met could claim to be making "data-driven" investment decisions.

While I had never intended to share this publicly, another change of tactics is in order. Perhaps it will be better received by some of the investors who follow me than by those who saw the method being applied in the pitch materials sent to them directly.

In running the updated numbers for 2024 I also noted that fraud by AI companies I was comparing against accounted for roughly a 700% increase in valuations and 477% increase in funds raised relative to AI companies with no documented instances of fraud.