Every deal is underwritten twice. Once with money, and once with a relationship, and only one of the two appears in the documents.
The second underwriting is often what makes the transaction possible. Someone vouched. Someone made an introduction that carried their own standing with it. By the time anything formal begins, that has already happened, and nobody has recorded it anywhere.
It is also the one that comes due, and it comes due at the worst possible moment.
The arithmetic
Two venture capitalists who graduated from the same university are about thirty-four percent more likely to invest in a deal together than two who did not. If they belong to the same ethnic minority group, the likelihood of them partnering rises by roughly thirty-nine percent. Having once worked at the same firm produces a similar pull. These figures come from a study by Paul Gompers, Vladimir Mukharlyamov and Yuhai Xuan, published in the Journal of Financial Economics, covering 3,510 individual investors and 12,577 portfolio companies.
Findings of this kind are usually read as an access problem, and that reading is fair as far as it goes. If capital moves along lines of shared background, then people outside those lines encounter a market that is formally open and practically closed. That argument has been made carefully by many people and I do not intend to repeat it.
The same study contains a second finding that is discussed far less, and it does not concern who receives the money. It concerns what happens to the money afterwards.
The second finding
The authors traced the performance of these partnerships. Investors who had previously worked at the same firm were, when they co-invested, associated with a seventeen percent lower probability of a successful exit. Where the pair had attended the same undergraduate university, the likelihood of success fell by nineteen percent. Where they shared an ethnic minority background, the reduction was twenty percent.
The characteristics that most reliably brought two investors together were the same characteristics associated with the largest declines in investment success.
The obvious inference is that similarity itself is the problem. The study points to something more specific.
Investors also sorted on demonstrated ability: two investors with degrees from top universities were about sixteen percent more likely to co-invest, and that kind of similarity was associated with better outcomes. What damaged performance was similarity of the other sort, the kind grounded in shared origin rather than shared capability. Investors were selecting on both, and only one of them was working.
The mechanism matters more than the magnitudes, and here the authors are unusually direct. They tested whether high-affinity pairs were simply choosing worse companies, and concluded they were not. The cost, in their words, is most likely attributable to poor decision-making by high-affinity syndicates after the investment is made.
The problem did not appear to be the quality of the deals. It appeared afterwards.
Where the debt falls due
The relationship that makes the deal possible is the same relationship that makes the difficult conversation afterwards unlikely. Familiarity opens the door and then stands in the doorway.
Call it affinity debt. The relationship is spent to originate the transaction, and correcting the transaction later requires spending it again, out of an account that has already been drawn down.
When something goes wrong in a portfolio company, somebody has to say the uncomfortable thing: that the strategy is not working, that the founder is the constraint, that the plan presented last quarter was optimistic. Whether that gets said depends on the cost of saying it. Between two people who met through work, that cost is a professional disagreement. Between two people who have known each other since university, it is something closer to a personal breach. The second pair does not consciously decide to avoid the subject. They simply find, repeatedly, that this is not quite the meeting for it.
There is a compounding effect that makes this worse. The study also found that affinity predicts not only whether two investors collaborate once, but how often they collaborate again. Groups selected for mutual comfort are also groups that keep re-forming, so the composition problem deepens without anyone choosing it. Nobody excludes anybody. The room simply keeps producing itself.
Beyond venture capital
Nothing in this mechanism is specific to venture capital. It appears wherever origination and correction are assigned to the same relationship.
An audit relationship that has run for fifteen years. A supplier who has been on the approved list since before the current head of procurement arrived. A family business where the directors were appointed by the promoter and one of them now has to tell him the succession plan is not working. A relationship manager who has covered the same borrower group for a decade and must now classify the exposure.
In each case the relationship did real work at the start. It made a stranger legible, shortened diligence, and got something done that would otherwise have stalled. In each case the same relationship is the instrument that must be used to raise the problem later.
Commercial banking offers one structural answer to this problem, and it did not solve it through character. It separated origination from credit approval, giving the second function its own reporting line, its own incentives and, ideally, no relationship to protect. Whatever one thinks of how well that works in practice, the design intent is unambiguous. The person who brought the business in is not the person who decides when it has gone wrong.
Venture capital does not generally have an equivalent separation built into the model. Most boards do not either.
What the evidence does not say
Several qualifications apply. The data run from 1973 to 2003, which is a long window but an old one, and the industry has changed in ways that may cut either direction. Success is measured by exit, which is a crude proxy for value created. The post-investment mechanism is inferred from the pattern of results rather than observed directly, and the researchers say so. More recent work by Gompers and colleagues continues in this direction, but anyone citing a single study as settled should be treated with suspicion, including me.
The larger qualification is that networks are not a defect in the system. They are doing necessary work. Under real uncertainty, where the outcome cannot be computed and the founder's character matters more than the model, trust is a reasonable substitute for information, and trust is built socially or not at all. A capital market with no relationships in it would not be more rational. It would be paralysed, or it would fall back on collateral and credentials, which have their own exclusions.
The point is not that capital should stop being social. It is that the social layer is doing more work than we account for, and some of that work is being done badly.
What follows
What I notice, sitting on the receiving end of investment discussions in several countries, is how much is settled before anything formal begins. By the time a proposal reaches a committee, the question of whether these are serious people has usually been answered somewhere else, by someone who knew someone. The committee then examines the numbers with real care. It is examining a decision that has already been substantially made, and its scrutiny is directed at the part that was never really in doubt.
The argument for mixed rooms is usually made on grounds of fairness and resisted on grounds of returns. This evidence points the other way. Affinity should not be assumed to be a concession made at the expense of performance. In this evidence, it is a cost carried by it, and the cost is largest exactly where the stakes are highest, in early positions where governance matters most.
The moment to worry is therefore not the investment committee. It is the second year, when the company is struggling and the people obliged to hold an unwelcome conversation are the people who introduced each other. Diligence is built almost entirely around the decision to invest. Very little institutional attention is paid to the conditions under which the follow-up judgment gets made, which is where this study locates the damage.
And it has to be handled structurally rather than by disposition. Everyone believes themselves capable of telling a friend an unwelcome truth, and most people are, once. What erodes is not courage but frequency, across years of shared meetings and mutual obligation. The answer is composition and process: who is in the room, whether anyone in it has no relationship to protect, and whether the difficult question is somebody's assigned job rather than an act of individual nerve.
The question worth asking
Capital is conventionally described as cold, and the description flatters it. It moves through friendship, shared classrooms and former colleagues, and it does so for reasons that are largely sensible, because that is how strangers become legible to one another.
The warmth is simply not free. It buys entry to the deal, and it collects payment later, in the conversations that nobody quite got round to having.
Which leaves a question that applies well outside venture capital, and is worth asking of any long relationship that has been good for business.
Who in this arrangement is still able to say the difficult thing?
And what would it cost them?
Sources
Paul Gompers, Vladimir Mukharlyamov and Yuhai Xuan, "The cost of friendship," Journal of Financial Economics. Sample of 3,510 individual investors and 12,577 portfolio companies, 1973 to 2003.
