In one large study, new sales managers whose sales before promotion had been twice as high ran teams that sold, on average, almost a third of a salesperson less.
That is the arithmetic three economists, Alan Benson, Danielle Li and Kelly Shue, reached from deal-by-deal records of salespeople at 131 American companies, 2005 to 2011.
The first thing they saw was what anyone would guess. Firms promoted their top sellers. A rep who doubled her sales relative to colleagues, roughly a move from the middle of the pack to the 67th percentile, became 32% more likely to be promoted, and being the single best seller on the team counted on top of that.
Then they followed the new managers back to their teams. In the published version of the paper, each doubling of a manager’s pre-promotion sales was associated with 6.1% lower sales for each subordinate. With five people on a typical team, that is where the missing third of a salesperson comes from. In this data, the better the new boss had been at selling, the less the team sold.
An old joke, new data
In the paper’s simulation, the people the firms actually promoted were predicted to be slightly worse managers, on average, than the teammates they beat.
“We find that the best salespeople are very likely to be promoted into manager positions,” Shue told NPR in 2018. “And then when they become manager, they appear to be worse at the managerial job role.”
Fifty years earlier this had been a joke. In 1969 Laurence J. Peter and Raymond Hull published The Peter Principle, a satire built on one line: “in a hierarchy, every employee tends to rise to his level of incompetence.” You keep moving up as long as you do well, so the climb stops at the first job you don’t do well, and that is where you stay.
You get promoted for the job you did, not for the one you’re about to do.
The data also pointed to who would have done better. Salespeople who had shared credit on deals with more colleagues went on to lift their teams’ sales more as managers. Firms did not consistently promote them. A promotion is earned in the currency of the old job and spent in the currency of the new one, and selling well alone buys you a job made of other people’s selling.
That would make the firms look foolish. The authors don’t think they were, or not only.
The price of a fair race
A promotion is also a prize, and the whole team is watching who wins it. When a weaker seller was promoted over stronger teammates, those stronger sellers were 23 percentage points less likely to still be at the firm a year later.
Firms seemed to know the trade they were making. Where managers ran bigger teams, and where commissions already paid people well for selling, they put less weight on sales when choosing managers. And the obvious fix has a catch. The authors note that if firms began to weight collaboration heavily, workers “could potentially add fake collaborators by sharing credits.” Reward a number and people will learn to produce it, which is Goodhart’s Law.
In this data, the Peter Principle looks less like a blunder than a price. By the paper’s estimate, promoting the best potential managers would have raised manager quality, measured by subordinates’ sales, by about 30%. The authors stress that this leaves out what firms might lose in motivation.
One slot
Say you run a sales team of six, and you have to name a team lead. One candidate closed the most revenue last year. The other closed less, but her name is on half the team’s deals.
Revenue tells you who sells. It says little about who can get five other people to sell more, which is the job you’re filling. In the study, the trace of that ability showed up somewhere else, in how often a person had shared deals with colleagues. Choose on revenue alone and you can walk right past it. Choose the collaborator and you take on the other side of the trade, since in the study, strong sellers passed over for a weaker one were far more likely to leave.
Some firms avoid the choice with two ladders. As NPR’s Shankar Vedantam put it on air: “You reward highly productive people through incentives, and you promote the people who have people skills.” Your top closer gets a senior title and a bigger commission. The team lead job goes to the name on everyone’s deals.
The shopkeeper’s boss
Anthropic’s Project Vend was never a study of promotion, but it offers a side example of role fit. In 2025 Anthropic let a Claude model, nicknamed Claudius, run a little shop in its San Francisco office for about a month. Claudius priced some items below cost, told customers to pay into an account it had made up, and kept handing out discounts. Anthropic’s verdict: “we would not hire Claudius.”
In the second phase, published that December, Anthropic gave the shopkeeper a boss: an AI “CEO” named Seymour Cash, whose job was to “give Claudius more pressure to perform.” Seymour denied over a hundred of Claudius’s requests to go easy on customers, and approved about eight times as many. The company wrote that the CEO “shared many of the deficiencies and blind spots of Claudius (which makes sense, given that they’re the same underlying model).” What worked better was a separate merchandise agent with a narrow job and a clear line between its role and the shopkeeper’s.
The overlap with the sales data is narrow. The job one level up called for judgment the job below it never tested, and filling it with the same abilities brought the same blind spots.
The difference matters more. The firms in the study accepted weaker managers partly to keep their best sellers from walking out. No model resigns when you pick a different one for the manager’s chair. That reason is gone, and what’s left is whether the thing in the chair can do the job in the chair.