When Fred Erhsam arrived at Goldman Sachs to trade G10 currencies in 2010, he was struck by the vested interests on his desk. “Algorithms were ruling the world, but I was hired onto a manual trading desk,” he recently told Khe Hy, the former Blackrock MD-turned philosophical blogger. Goldman clearly needed to add to the algo traders and subtract from the manual traders, but this wasn’t happening and Erhsam was in no position to effect a change: “You are 21 years old, it’s difficult to vent…the traders and engineers were viewed as second class citizens.”
Seven years later and five years after Erhsam quit and founded Coinbase, it seems that Goldman, of all banks, has moved on. With ex-chief information officer Marty Chavez as CFO, the firm is determinedly automating all that it can. Chavez’ vision is for a “data lake” that pulls information on transactions, markets and investment research and overlays it with machine learning. A third of Goldman’s employees globally are already working in technology. 39% of the jobs it’s currently advertising in EMEA are in the technology division and many of the others are for the quants or so-called “strats” who’ll provide the underpinning for Chavez’ plans. Insiders in the equities strats team talk excitedly about their numbers “doubling” in the years to come. Goldman is already advertising for data science juniors to join a new “front office machine learning strats team,” and it’s not alone – UBS is also on the lookout for someone to lead a team of developers working on artificial intelligence for its equities business. The future is upon us.
Some things haven’t changed, though. As was Erhsam’s experience in 2010, the dominant groups of the past aren’t about to cede control that easily. Goldman’s strats are paid well, but they still receive a lot less than traders in what was traditionally seen as the front office. Goldman insiders say vice presidents (VPs) in the strats teams that support salespeople and traders can expect to earn $200k to $300k (£155k to £233k) with seven years’ experience. That’s a lot, but traders could easily be making double.
If salary surveys from some of the UK’s leading recruitment firms are to be trusted, though, Goldman’s strats are unusually well looked after. At other banks, the machine learning and data specialists who are supposed to be the future are allegedly paid a pittance.
As the charts below show, Robert Walters says banks in London are paying data scientists with five years’ experience salaries of £60k. This could be dismissed as inaccurate or an anomaly – except that Morgan McKinley puts mid-ranking salaries for machine learning engineers at just £65k.
The diminutive pay for data and machine learning jobs is of a piece with the degradation of banking technology pay as a whole. “If you go into a development role in a bank, you’re going to start on £40k to £50k and that won’t much in the first three years,” says one London technology headhunter. “Nowadays you won’t get much of a bonus in technology until you’re at executive director level. Most people got nothing last year, although the bonuses in technology used to be pretty good back in the day.”
Poor pay may also be a reflection of the fact that not all data jobs are equal. Oliver Blaydon at Stonegate Search says compensation for data professionals depends upon the jobs in question are located within the banking taxonomy. “”If you’re a data analyst who sits in a technology division working with data, you’re not going to be paid as much as a data scientist in the front office who’s building AI solutions using the data,” Blaydon points out. In the latter case, he argues that pay is far higher: “If you have five years’ plus postgrad experience and a very good Masters or PhD, you should be on £170k to £220k a year in data science and the very top data scientists command significantly more.”
Banks which are paying the kinds of numbers indicated by Morgan McKinley and Robert Walters in the hope of attracting front office data scientists may therefore need to seriously reassess. Because banks’ data science and machine learning teams are small, Blaydon says there’s a tendency for them to try hiring experienced staff who’ve built and monetized applications at other firms. And these people come for a premium or not at all.
“Technology companies like Google and Palantir front-load pay to their top data scientists and engineers,” confirms one London recruiter, speaking off the record. “PhDs are given $150k to $200k of stock which vests over five years, so it’s incredibly difficult for banks pull them out.” Even hedge funds can’t compete: the CEO of Man Group recently complained that Google is “hoovering up” all the data scientists in the market and that his pockets weren’t deep enough to suck them back.
In the circumstances, therefore, banks which deign to pay top machine learning and data science talent less than £80k will be laughed out the market. Or be left with the dregs. Recruiters say the latter is already happening: “Over the last four or five years, the standard of technologists applying to banks has fallen significantly. They are having to deal with third tier graduates from mediocre computer science courses. “
Things may be changing. Blaydon says the penny has already dropped and that pay for data scientists and machine learning professionals in banks is, “going up very, very quickly.” In the process, pay elsewhere may need to fall. It’s time for the vested interests on sales and trading desks to cede pay to the incoming PhDs.