The Cost of Keeping “That Guy”

The first time I heard that the firm could not manage without Bertram, I was the one saying it. Bertram is not his name: I have borrowed it from Bertram Gilfoyle of the television comedy Silicon Valley, a systems architect of real ability who treats everyone outside engineering as a fault to be worked around. Every firm that uses technology has one, and the staff call him “that guy”. Mine had built our telematics rating engine, which prices a motor policy from every journey the car makes and is an order of magnitude more complicated than the ordinary kind of rating engine. It was fast, accurate and understood by exactly one person. The documentation was a single file called README, which contained the word “No.” So I made him Technical Lead.

For two years he divided the firm into engineering and “the talking departments”, which were sales, claims, compliance, finance and me. Maureen, who ran sales in our wholesale broking division, told the product meeting that our largest broker was about to leave, on the evidence that they had stopped complaining. He asked her to come back with data. The broker left, which was data, and so in time did Maureen. He was also often right. He warned me that the service which receives the journeys would fail under the load, and asked for £140,000 to rebuild it. I said that was too expensive. It failed on a Saturday in December, the refunds, compensation and overtime came to a little over £300,000, and we gave him a bonus for mending it, as we did every time. Each time, I did a sum with two columns. The cost of losing him went in the first and was large. The cost of keeping him went in the second, and I entered it as nil. I have since read what is known about this arrangement, and very little of it lets the employer off.

What the laboratories say

Psychology has no diagnosis for a gifted man who despises the sales department. It has a set of findings about expert confidence, most of them from American laboratories, and they turn out to be about a wider class of person than he or I would have liked.

The nearest is the work of Victor Ottati and colleagues (2015). In six small experiments they made people feel expert, in some cases with nothing more than an easy quiz, and found that those people became less willing to consider other views. Their explanation was social. Expert is a role, the role carries a licence to stop listening, and people who feel entitled to the licence use it. The samples were thirty to sixty people each, and a direct replication in 2018 found the effect with some ways of inducing the feeling and not with others. I treat it as a description of a custom I recognise, and not yet as a law.

The firmer result is less flattering to everyone. Stanovich, West and Toplak (2013) reviewed several kinds of experiment on myside bias (that particularly pernicious subtype of confirmation bias: the habit of marking evidence up when it agrees with you) and found that it has almost no relation to intelligence. Kahan and colleagues (2017) gave American adults a table of results to interpret. When the table was said to describe a skin cream, the more numerate subjects read it correctly more often. When the same figures were said to describe a ban on carrying concealed handguns, the more numerate ended up further apart along party lines than the less numerate, because they were better at finding the reading they wanted. That experiment concerned political identity. Nobody, so far as I have found, has run it on engineers and sales directors, and the step from party loyalty to professional pride is my inference.

A third finding suggests why the contempt runs in one direction. Rozenblit and Keil (2002) asked people to rate how well they understood ordinary devices such as a zip or a flush toilet, then to explain one, and watched the ratings fall. An engineer has been made to explain his own work many times and knows what it takes. He has never been made to explain how a broker account is kept, so he goes on believing he understands it well enough to see that there is nothing in it. Again the application is mine.

What Maureen knew when she said the broker had stopped complaining is the kind of knowledge Michael Polanyi described in 1966: we can know more than we can tell. Bertram’s request for data was reasonable in form, and it ensured that the only knowledge admitted to the product meeting was the kind he produced. Hayek had met the attitude by 1945. Writing about the trader who lives on knowledge of local conditions, he called it “a curious fact that this sort of knowledge should today be generally regarded with a kind of contempt”.

None of this tells me what went on in Bertram’s head. The studies describe averages in groups of strangers, and I will not fit a man to them from memory. I am better placed to recognise myself. I felt expert in judging people, I had a conclusion I was fond of, which was that I had hired a genius, and I spent two years marking up the evidence for it.

What the sum left out

My first column assumed that the value of the engine was the value of Bertram. Economists have treated that kind of attribution as a problem since 1972, when Alchian and Demsetz described two men lifting cargo into a truck. You can weigh what was loaded in a day. You cannot work out from the weight what either man contributed. Michael Kremer’s model of production (1993), named after the seal that failed on Challenger, goes further and multiplies the quality of each task by the quality of every other, so that a first-rate engine attached to a third-rate sales operation is worth a fraction of itself. It is a stylised model. It is also how I build a rating model, in which no single factor can be credited with the premium, and I did not notice the resemblance.

What I had measured was dependency. A missing bolt can ground an aircraft, which tells you where the aircraft is exposed and nothing about what the bolt earned.

Dependency has been priced. Israelsen and Yonker (2017) identified American listed firms that had disclosed a life policy on a key employee, and found that when such a person’s departure was announced the most exposed firms lost 8% of their value. Their key employees were four times as likely as top managers to hold a doctorate, so the person the market worries about is more often a Bertram than a chief executive. I sell insurance and know what that policy does. It pays on his death. The study prices his leaving. Neither puts a figure on his staying, which was the peril I was running.

How does a firm come to depend on one person? Partly through him. I asked Bertram three times to document the engine. The first time it was on his list. The second time he did not understand what I meant by documentation. The third time he explained that a wrong document was more dangerous than none. Connelly and colleagues (2012), working from surveys and interviews with a few hundred employees and students, mostly North American, sorted what they call knowledge hiding into three forms: evasive hiding, playing dumb and rationalised hiding. I had collected the set. Software engineers have their own measure, the truck factor, which is the number of people who would have to be hit by a truck before a project stalled. Avelino and colleagues (2016) estimated it for 133 popular open-source projects, and in most of them the answer was one or two.

Mostly, though, the firm does it to itself, and here the evidence is against me. Benson, Li and Shue (2019) studied sales staff at 214 American firms. The firms promoted their best sellers, and a doubling of a new manager’s own previous sales went with a 7.5% fall in the sales of each person who then reported to them. That is sales, and I have found no equivalent study of engineers. I can report one case.

I also paid a bonus for every night the system was rescued and nothing for any night on which it did not need to be. A rescue is seen, and it belongs to one person. Prevention is invisible and its benefit is shared by everyone. A firm that pays for the first and not the second will get more rescues, and nobody has to behave badly for that to happen.

When he is right

The best argument for putting up with a Bertram is that he is sometimes the only person in the building who is right, about something that costs far more than a weekend of refunds.

On the evening of 27 January 1986, engineers at Morton Thiokol, which made the solid rocket boosters for the space shuttle, recommended against launching Challenger in the cold forecast for the next morning. One of them, Roger Boisjoly, had warned in writing the previous July that the seals in the booster joints could fail. Thiokol’s managers withdrew to confer. By Boisjoly’s account, the general manager told the vice-president of engineering to take off his engineering hat and put on his management hat. The recommendation was reversed, and seven people died the next morning.

Bertram’s opinion of the evidence used by the talking departments also has something behind it. Bergh and colleagues (2017) tried to reproduce the results of 88 papers in the Strategic Management Journal from the data printed in them. About 70% did not disclose enough to be checked at all. Of those that could be checked, almost a third reported significant results that did not hold.

Then there is the finding I would have preferred to come out the other way. Melwani and Barsade (2011) ran three experiments on what being treated with contempt does to the person on the receiving end. Their subjects became more aggressive towards the person who had shown it. They also did better at the task. These were short tasks in a laboratory, and I draw no conclusion from them about a sales team over two years. But I cannot cite the study as showing that contempt makes people worse at their jobs, which is what I had assumed it said.

So the useful question is how to tell a warning from a habit. Boisjoly named a joint, a temperature and a consequence. Bertram’s warning about the journey service was of the same kind: a component, the load it was built for, the load it was about to meet, and a price. I should have read it as underwriting information and I read it as a budget request. His view of Maureen was a different sort of statement. It named no mechanism, covered five departments and could not have been shown to be wrong. The same man made both, in the same voice, and I had been listening to the voice.

It is tempting to sort such people into types, with the brilliant and correct at one end and the merely obstructive at the other. My Bertram was both in the same week. The sorting has to be done claim by claim, and doing it was my job. I once asked him whether he could imagine a commercial decision of ours being correct. He said he could imagine it. I laughed, and wrote nothing in the second column.

What happened to Bertram

It ended over a number. I wanted claims frequency by driving score for a board paper, which at a telematics insurer is roughly a request for a glass of water. I raised a ticket, as required. On the Thursday it was closed with the status “Won’t fix: user error”. I was the user.

I dismissed him at his desk, in a voice I had last used on a car hire firm in Malaga, and went back to my office to wait for the answer.

It did not come. Priyanka, who was less experienced and less enamoured of her own abilities, asked whether she and two of the others could have the winter to read the engine. She reported that it was as complicated as he had always said, that most of it was very good, and that it could be written down. Then they rebuilt the journey service, for about what he had quoted.

For a while I told this as the story of how I was held hostage. Nobody kidnapped me. I hired him, promoted him out of the job he was good at, turned down his one costed warning, paid him by the rescue and told the board he was indispensable before he had thought to say so himself. He was a very good engineer doing what the firm rewarded. I was the firm.

Last month the board reviewed key person risk, and I reported that we no longer depended on any one individual. The chair asked whether I had ever written down how the pricing committee reaches its decisions.

I have started a file. It is called README, and I am finding it hard to get past the first word.

Principal sources: Ottati et al. (2015) on earned dogmatism; Stanovich, West and Toplak (2013) on myside bias; Kahan et al. (2017) on motivated numeracy; Rozenblit and Keil (2002) on the illusion of explanatory depth; Polanyi, The Tacit Dimension (1966); Hayek, “The Use of Knowledge in Society” (1945); Alchian and Demsetz (1972); Kremer (1993) on O-ring production; Israelsen and Yonker (2017) on key human capital; Connelly et al. (2012) on knowledge hiding; Avelino et al. (2016) on truck factors; Benson, Li and Shue (2019) on promoting the best salespeople; Bergh et al. (2017) on reproducibility in the Strategic Management Journal; Melwani and Barsade (2011) on contempt. The applications to engineers and sales directors are the author’s own.

Note: names and details have been altered to protect the guilty and the innocent.