Most Hiring and Investment Decisions are Terrible Due to Pattern Matching
Posted by Bob Warfield on November 27, 2012
One often hears venture capitalists refer to their decision making process as “pattern matching“. This reference to machine learning (and a fuzzy match really ought to be pattern recognition, but hey, the techies aren’t using this language so much) is intended to sound smart. One envisions coupling years of hard-earned experienced with the biggest neural net found between two ears. Unfortunately, all too often, pattern matching involves making inferences from a statistically insignificant amount of data and calling that a formula for success or wisdom. It’s just noise, yet we see it so very often at work in making the most important of decisions.
Consider the decision of who to hire. I recently read two articles that smacked heavily of pattern matching. First was a write-up in the San Jose Mercury News on Silicon Valley’s Age Bias. While cultural biases are certainly at work (people of similar ages simply relate to one another better), those making decisions from any greater gap in years are doing so based on pattern matching. A careful reading of the article yields the best advice for job applicants is to:
– Ditch your wrist watch, they’re Old School–kidz these days use their phone.
– Make sure you have fruit on any electronic devices you’re toting. No Dell or Blackberries please, if Steve Jobs didn’t envision it, it is a boat anchor holding your career back.
– Shave your head or dye your hair, and invest in as much cosmetic surgery as you can afford. Wrinkles are Verboten!
– Lose the Bally Executive Loafers grandpa, and don some Converse High Tops. While you’re at it, forget Faconnable dude, and get a hip tee-shirt. Maybe I will wear my “There’s no place like 127.0.0.1” to interviews.
It’s possible those are useful attributes for executive success, but they sound to me like noise and affectation. In other words, they’re trying to game the pattern matchers.
Jason Lemkin has his hiring pattern matcher running at full song when he advises not to hire CEO’s, Architects, Gamers, or Dualies. Dang, I’m shot down on two before I even get started since CEO was the very first job I ever had and I have been a “Dualie” (VP of both Engineering and Product Management) on a number of occassions. I can’t agree with a single one of Jason’s pattern matching rules and the reason is simple–I know too many exceptions for each of them, not even including my own personal work experience. The reasons why they’re exceptions begins to lay the groundwork for why pattern matching results in so many bad decisions:
When hiring or investing, we want the Exceptions, not the norms.
But, Pattern Matching weeds out the Exceptions before we even get a chance to consider them.
That’s worth repeating again: when hiring or investing, we are explicitly trying to find the exceptions. If asked, we will react very negatively to the idea we’re seeking the mean. Yet, if it’s exceptions we want, why do we pursue a decision making methodology that is so geared to not adding value because it depends on statistically insignificant noise and often eliminates the Exceptions before we know they’re there? These ideas, conscious (I won’t hire a CEO) or unconscious (that guy has no gray hair on his bald head so he must be a better specimen than the last guy I talked to), when laid bare on the table, start to sound pretty silly when you think of it in the light of finding Truly Exceptional People or Deals. If there really was a formula, everyone would be Exceptional and Rich, but they’re not. Most great success stories are Contrarian. We can rationalize Google after the fact all we want, but back when it started, it was a contrarian play. The pattern matchers would’ve argued search was done and over and companies like Excite and Alta Vista had won, much as folks today are arguing the Consumer Internet is done and companies like Facebook and Twitter own it.
Before I go much further, let’s not confuse Selling with Buying. If you are selling, knowing the Buyer is a pattern matcher and what patterns they’re looking for is a huge advantage. It tells you exactly how to pitch what you’re selling to win the business. Always go to school on your Buyer’s tendencies in this regard. If you can’t get much information, ask questions that will lead to your own pattern matching model of what that Buyer expects. The mere fact that you can go to school on pattern matchers shows they have an additional weakness that is endemic to that form of decision making: not only does it give the wrong answers but it is too easy to game.
What I want to talk about is not advice for the Seller, the path there is clear–lean into the pattern matching with gusto. Rather, I want to talk about how to be a better Buyer. How to avoid the Pattern Matching Trap in your own decision making.
Step 1: Write Down Your Objective Criteria
Like I said before, nothing like laying the rules down on the table and shining a bright light on them.
Step 2: Identify Rules that Are Hopelessly Anecdotal, Perishable, or Symptoms Rather Than the Disease and Eliminate or Rewrite them
Anyone who has worked in machine learning or big data (I’ve done both, having a Genetic Algorithms and an Unstructured Search startup under my belt) will be familiar with the notion of curve fitting, or to use a better term, overfitting. Testing a predictive algorithm involves feeding it historical data and seeing how well it’s answer matches the historical outcomes. It is important to always hold out a significant body of historical data that is not used to train the algorithm, but only to evaluate its accuracy. If you fail to do so, you run the risk that your algorithm embodies no essential understanding of what it is predicting. Instead it is simply implementing a mathematical algorithm that happens to produce a result similar to the historical data it was trained on. You have an overfitted algorithm in that case, and it is worse than useless because it provides misleading answers under the cloak of way too much confidence.
For human pattern matchers, it is almost impossible to get enough data points to be statistically significant and to avoid overfitting. The consequence will be that you can point to several historical examples where some formula worked, but in the latest example at hand you have a bad hire or you’re losing on an investment because it strangely isn’t behaving according to the model. The grim reality is you’ve based your decision on purely anecdotal evidence and not enough of that. You overcome that shortcoming by going beyond the pattern to analyze the underlying causes that lead to the pattern.
Patterns are perishable too, because they’re driven by market forces that change over time. We can go back perhaps 3 or 4 VC cycles to the heyday of the 80’s or 90’s and find all sorts of patterns that just simply are no longer operative. It doesn’t matter how well they worked then. To determine whether you have perishable data, you need to have current data points. This won’t save you if you’re in the absolute midst of a transition, but most of the time the system is stable and not in a punctuated equilibrium. So make sure you have some current data points and are not just referring to something that happened years ago one time.
Lastly, on the question of Symptoms rather than Disease, it’s pretty simple:
Is that pattern there because it explains something in a deep and insightful manner, or because something hurt at the time, and you’re not sure why, but you want to make sure none of the conditions surrounding that hurt can be repeated?
The latter is a symptom while the former shows at least some understanding of the disease.
Jason, whom I have ribbed a bit for his article on not hiring CEO’s, actually provides his analysis of the underlying disease for each case. He doesn’t want to hire someone who had CEO on their resume for a small company because he is afraid they’ll be too full of themselves. He also goes on to further clarify that it is people who were CEO’s of insignificant companies that he particularly worries about, and if you were CEO of a real startup that didn’t make it, that’s okay. The thing is, having done that analysis, I would skip the whole “Don’t hire CEO’s” thing and instead think about how to objectively determine which candidates are too full of themselves. “Don’t hire CEO’s” was a proxy for the rule that should have been, because almost any candidate for any senior position might turn out to be too full of themselves. Avoiding such people is the real insight, and that insight is not captured simply by avoiding “CEO” on resumes.
Step 3: Focus on Rules that are Objective, Verifiable With Enough Data Points, and Lead to a Model that Makes Sense. Avoid the Proxies.
Avoid proxies. A proxy is a rule that sounds right and works for a while, but turns out to be treating the Symptom and not the Disease. I was in a very successful company one time that optimized all of its behavior around PR. Turns out that was a great proxy for a while but what Wall Street really wanted was Profits, not PR. PR was just a tool to be used to generate more Profits and focusing single-mindedly on it led us to take our eyes off a lot of balls that had nothing to do with our image but everything to do with our profitability. That company went from being a $500 million behemoth to nonexistent in just a few years after that time when PR started to fail.
Let’s consider the hiring question. Remember, we want to hire Exceptions and not regress to the mean. Does having some hugely successful company like Google or Facebook on the resume constitute a useful signal?
I would argue no, because there are lots of people that have worked at some big successful company. We’ve all had the experience of hiring one of them and then wondering why they were ever successful in those organizations (or even were they successful) because they aren’t accomplishing anything in our organization except telling endless useless stories about how things were done at their old company. Working for one of those Beacon companies is a proxy and what’s needed is some rule that far fewer people can qualify for. A rule that correctly identifies something more important than fogging a mirror at a Google or a Facebook.
What I like to look for is someone who has played a critical role in accomplishing something amazing at almost any company. Have you tapped into one of the lead developers on Google’s Gmail? Now you’re talking. But how do you know they were a lead developer? This is a critical role for references. Track down somebody on the Gmail team–it’s a targeted search so you should be able to find someone the candidate hasn’t given you pretty easily using LinkedIn or some similar tool. Call them up and be very straightforward about asking whether your candidate was really one of the top handful of developers that made it happen. If the story is true, your contact should be extremely enthusiastic about them. If they’re evasive, the story is not true. Move on. This is not the Droid you’re looking for.
Hiring the young is a particularly pernicious proxy for this kind of criteria. If you’re 22, you haven’t had much time to make a huge contribution to an impressive project. Even if you did, you have probably only been able to do so once. Perhaps you’re even still in the middle of finishing doing so. OTOH, if you have a 40 year old, you can start to look at statistically significant trends. By 40 you will have had a shot at at least 4 five-year stints. People say if you stay longer than 5 years, something is wrong, but that may be just another false proxy. Out of 4 five year stints, how many were successful? At age 40, I’d had 2 successes, 1 failure, and had started in a role for a company that would eventually be my 3rd success with an IPO. Now do you want to hire the youngster who may have done it once but mostly just sounds really smart, has done a lot of hacking, and is out of Stanford, or a guy who actually had a 67% success rate? Apparently, VC’s and others want the former. They just don’t realize that Old Age and Treachery Always Overcomes Youth and Enthusiasm. That and they probably don’t want to sit on a board with a CEO who has tons more operating experience than they do and is hence less “coachable”.
I’ll close with a final pattern matching example. In a great post, Fred Wilson recently exposed one of the big sources of the herd mentality among Silicon Valley VC’s. In “What Has Changed?“, Fred talks about what’s different with today’s deals, and is talking primarily about the growing sentiment there is a shift away from investing in Consumer Internet deals. He mentions three causes, but the most interesting is this one:
The momentum/late stage investors have moved from consumer to enterprise. there is a large pool of money in the venture capital asset class that is opportunistic, momentum driven, and thesis agnostic. this pool is driven largely by the public markets. this pool of capital was “all in” on consumer web/social web in the 2009-2011 time frame. it drove a lot of activity throughout the venture capital markets because each layer of the VC stack (angel, seed, Srs A, Srs B, Srs C, etc) needs to be aware of what the next layer up wants to fund. when the momentum/late stage wanted web/social, the layers below gave them web/social. now that the momentum/late stage wants enterprise, we should expect the layers below to give them enterprise.
To put that in the context of this article, the early stages are trying to game the pattern matching of the late stages to give those investors what they want. Dave McClure, a seed stage investor from 500Hats, reacts pretty negatively to that thesis by saying, “If true, this is a huge error.” He’s right and both Fred and Dave go on to suggest that those who have a clue (and both do) will continue to focus on their core investment thesis regardless of what these late stage funds see in their pattern matching crystal balls. Unfortunately, the nature of VC is that it is a universe whose physical laws are constructed to maximize the likelihood of bad pattern matching. After all, an industry where one deal in 20 or 30 makes a fund will focus on the statistically insignificant by design. They lose track of minimizing the failure of the other 19 to 29 deals and focus entirely on finding the next One. This precludes the model of a fund that is pretty good at minimizing loss and making solid singles and doubles and focuses entirely on home runs. Ironic that Moneyball says that’s exactly the wrong thing to do and VC returns are doing so poorly these days. Clearly separating the Cause and Effect of Successes and Failures is just one more way of saying avoid too much Pattern Matching.
Conclusion
Much of the world works through pattern matching. It’s a bad approach to decision making, but one you will have to deal with. Your skill will be in getting your signature to fit the pattern matching key of your buyer, and in avoiding the pattern matching trap for your own decisions. In the hiring world, it’s figuring out how to identify the truly exceptional before you’ve already weeded them out with banal pattern matching rules. In the investment world, it’s figuring out what the market really wants and needs while dressing that up to fit the pattern matching of your Board and Investors. Having to endure that task is one reason they say it is lonely at the top. If you’re pattern matching, you’re doing what many others are doing and you’re susceptible to being gamed. You’re not being Exceptional, so try another strategy.
The other approach for entrepreneurs is to keep the pattern matchers out of your life as much as possible. As an entrepreneur, you don’t need to get hired because you can create your own job. But, one of the important secrets is you don’t need investors either. There’s a big world of bootstrappers out there and today’s investors are going to insist that you take most of the bootstrapping journey before they’ll help you anyway. On the subject of entrepreneurs creating their own jobs and not needing investment, ironically, most potential entrepreneurs I talk to that are worried about getting investments are guilty of following another proxy. What they really want is the job and salary that they perceive as coming from the investment. If you are in that category and you’re honest with yourself, you will have realized you are not an entrepreneur after all. No harm in that and your decision making will be a lot clearer once you realize what you’re really after and quit trying to pattern match your way into being a founder.
People who see beyond the surface patterns to the essential truths (hey, maybe they’re all just increasingly successful patterns) are called Visionaries. They’re the ones that Win Big more than once.
5 Responses to “Most Hiring and Investment Decisions are Terrible Due to Pattern Matching”
You must log in to post a comment.
John Greathouse (@johngreathouse) said
Bob – thanks for the shout out re my definition of Pattern Matching. I just rewrote this entry and will soon publish it on Forbes.
Great content – keep it up!
Ryan Batterman said
“For human pattern matchers, it is almost impossible to get enough data points to be statistically significant and to avoid overfitting.”
People have done studies to determine the validity
of various predictors of job performance. e.g. see Hunter and Hunter (1984) (http://psycnet.apa.org/index.cfm?fa=fulltext.journal&jcode=bul&vol=96&issue=1&page=72&format=PDF) — in particular pg 90. Cognitive ability (i.e. IQ) has a correlation coefficient with job performance of .53 — huge! And it’s been verified by hundreds of studies. Unfortunately it’s illegal to use it though 😐
Bob Warfield said
Ryan, thanks, that’s a good observation.
Putting IQ tests aside, I think testing has a lot of potential, but it has to be used carefully. For example, some folks who are otherwise excellent developers may be lousy at some timed test that pops up and asks them to create an algorithm to do some crazy thing. I would look for a way to offer a test that more closely mimics the work environment. For example, the availability of tools like github could make it easy to actually give an applicant a real assignment over the course of one week to the next week’s interview. Give them access to some code (not all your source code!) and contacts with the other engineers to ask questions. See if they can crank up and accomplish something and then have them present what they did to the team and do a code review.
Another test I think is helpful are the personality profiles like Myers Briggs. I have had the experience of seeing how well I interact with the different profiles versus my own and the test has real value in determining how well you might get on with your team.
jasonlkn said
🙂
The Series A Crunch: One More Reason to Bootstrap and Skip Venture Capital « SmoothSpan Blog said
[…] jasonlkn on Most Hiring and Investment Dec… […]