#35 - Against the VC PWERM
Epistemology and uncertainty in venture capital
In the world of venture capital, uncertainty plays a large role, given the long 7-10 year time horizon of VC investments. One way to resolve some of that uncertainty empirically is to do what’s called a probability-weighted expected return multiple (PWERM) analysis. A sample, simplified PWERM for a company might look something like this:
10% of 0x return: The company blows up
25% of 2x return: The company is acquired in the next 3 years
25% of 5x return: The company can’t grow outside of America due to regulations but does remarkably well in the American market
30% of 10x return: The company continues along its current growth trajectory
10% of 50x return: The company explodes in growth, old incumbents go underwater
PWERM = 9.75x return (.1*0 + .25*2 + .25*5 + .3*10 + .1*50)
As you can see, the PWERM is basically a rough estimate of the return for an investment, helping VCs to decide whether a startup has enough certainty of financial upside to be worth the risk. As such, many VC firms often have a PWERM threshold for their investments (i.e., if this company doesn’t have a PWERM of ≥ 10x, we’re not investing).
But this begs an interesting epistemological question: How do VCs know or determine the probabilities and magnitudes of outcomes? After all, the hard thing about evaluating startups is that a startup is a grand experiment with a sample size of 1. We can’t run a billion simulations of Mark Zuckerberg starting Facebook back in 2004 to evaluate the probability of success at each step along the way. You can even ask meta-questions about the PWERM: What’s the probability that a particular probability is right? It’s all a game of uncertainty.
Of course, VCs use all sorts of heuristics to evaluate an investment, from company financials, to the strength of the team, to all sorts of business strategy theories. Those heuristics are all well and good. But translating these factors into probabilities in a PWERM can often be not a science, nor an art, but honestly some form of magic.
There are many dangers of relying heavily on PWERM. Here are 3:
(1) PWERM is amorphous. Two investors could arrive at very different PWERMs for the same company at the same round of funding. To make matters worse, the PWERM usually doesn’t come with any confidence thresholds or intervals.
(2) PWERM is easily game-able. By the time VCs begin calculating the PWERM, they may already have an inclination of whether they want to invest in the company, and they can intentionally, or perhaps even unconsciously, fiddle with the percentages and the valuations in a PWERM analysis to ensure that the final PWERM meets (or doesn’t meet) the firm’s threshold. Under this lens, the PWERM is merely the dressing that covers the black box of VC decision-making to give it an air of empirical legitimacy.
(3) PWERM is opaque. Standing alone, the PWERM is just some numbers attached to some outcomes without explanation of why those numbers correspond to those outcomes. In other words, it’s not very communicative. Even worse, once an investor has arrived at a PWERM, others on the investment team may be tempted to glance quickly at the number, form judgments about the company, and pull their attention away from fully digging into the company’s underlying risk factors.
There are, however, some ways to introduce more empirical rigor into a PWERM. For instance, even though there’s only been and only ever will be exactly one Shopify, there are dozens of other successful software-as-a-service (SaaS) companies. Studying those SaaS companies can reveal what a healthy, venture-returnable SaaS company looks like at each stage of fundraising, and indeed, a ton of ink has already been spilled on what financial metrics VCs are looking for in SaaS. A VC can therefore do some sophisticated financial modeling to arrive at probabilities of future success of a SaaS company. This appears to be exactly what Tribe Capital has done with its aptly-named Magic 8-Ball software, which uses a big-data-driven approach to evaluate a startup’s product-market fit and growth trajectory. Similarly, PWERM can be useful in the world of later-stage venture capital, where there’s more data on the company in question, and where investors might have an easier time comparing the company’s metrics to those of similar public companies.
In short, relying heavily on a PWERM-like metric makes sense where the investors have enough data to know empirically and precisely what factors went into the PWERM and how those factors were weighted. But in early-stage deep tech investing—investing in the world of atoms—where there’s hardly a uniform playbook or timeline for success metrics, it doesn’t seem right to me to lead with the PWERM. There are way too many moving parts that vary from company to company. As a rule of thumb, a top-level metric that is ill-defined will at best confuse the decision-makers and at worst lead them to make the wrong decisions.
In my opinion, early-stage non-SaaS venture firms shouldn’t have a hard-and-fast rule on PWERM requirements. VC firms also shouldn’t be waving their hands and, without context, throwing around sentences like “The PWERM is great” or “The PWERM is too low” unless everyone in the firm knows and agrees that that’s code for “I like / don’t like the investment for some reasons I’m not explicitly disclosing to you right now, but feel free to pick my brain on it if you’d like.” Instead, the PWERM, if it is to be used at all, should be a minor footnote in a broader investment memo that actually analyzes all the reasons to be excited and hesitant on a particular investment. This forces the VC to articulate their thoughts about the company and helps those around the VC to also learn from their analysis. In the same vein, investors who see a PWERM, standing alone, should always question the assumptions of the investor who arrived at that PWERM.
📚 What I’m reading
The rich kids who want to tear down capitalism. (The New York Times)
An interview with Toby Lütke, CEO of Shopify. (The Observer Effect)
What comes after smartphones? (Benedict Evans)
7 rejections. Brian Chesky, CEO and founder of Airbnb, shared a few rejection emails from VCs who chose not to invest in Airbnb in 2008 at a $1.5m valuation. Airbnb today is worth ~$100bn. (Medium)
An interview with Chris Best, CEO of Substack. (The Verge)
Framing a privacy right: Legislative findings for federal privacy legislation. (Brookings)
The left is now the right. (Matt Taibbi)
Voters say those on the other side “don’t get them.” Here’s what they want them to know. (Pew Research Center)
Facebook’s end-of-year internal product demos. Three interesting announcements here: (1) An AI tool to summarize news articles; (2) a wrist sensor to decode neural signals; and (3) a new virtual reality social network where users will be able to hang out with their avatars. (BuzzFeed News)
John Roberts’s self-defeating attempt to make the court appear nonpolitical. I’ve previously railed against the politicization of the Supreme Court, so this article was particularly fascinating for me. Key quote: “By striving so conspicuously to depoliticize the Supreme Court, [Chief Justice John Roberts] has brought about the very thing he hoped to prevent: No one has done more to politicize the court than the chief justice.” (Washington Post)