Unlock the Editor’s Digest for free
Roula Khalaf, Editor of the FT, selects her favourite stories in this weekly newsletter.
Investment banking is a people business. Clients pay generous fees in the hope of getting smart advice from analytical thinkers; relationships are managed deftly by silver-tongued advisers. An experiment by UBS raises the question of whether the humanity in high finance is overrated.
The Swiss bank has digitally cloned about three dozen of its equity analysts, generating short videos presented by lifelike avatars. They perform scripts based on the analyst’s research notes, complete with hand gestures and eyebrow raises. The AI is good but not perfect, so the result is slightly unheimlich. Still, the bank says the videos perform as well with clients as the old-fashioned kind.
Replicating living employees might be new, but the use of AI in banks is not. Mostly, employees rather than clients engage with it. Morgan Stanley has a note-taking assistant it uses during meetings; Goldman Sachs runs “co-pilots” to help with everything from coding to translation. Keeping the bots on the inside is rational: for brands boasting supernormal smarts, so-called hallucinations or communication gaffes can be corrosive.
A common catchphrase on Wall Street, therefore, is “human in the loop”. That means a pair of biological eyeballs scanning everything destined for a client. UBS follows that principle too: in the case of its avatars, analysts review both script and final video. As they should, because in highly regulated industries, humans bear the brunt of mistakes, as well as the risk of clawed-back pay or professional bans.
More financial firms will inevitably follow UBS in wearing their AI on the outside, and equity analysis is a good place to start. Commonly available models can now do the work of a second-year associate, according to one head of research. In time, AI ought to be able to spot patterns and corral data in ways a human cannot. It would be naive to think only drudge work will be phased out, as fewer humans supervise broader loops.
As “agentic” AI gets more sophisticated, with ever more data available for rapid retrieval, the sky is the limit. Updating price targets in real time? Sure. Putting pointy questions to a company CEO on the quarterly earnings call? Why not? At the very least, it would pull the plug on the dismal habit of congratulating the boss on a fantastic set of results — as half the analysts on Microsoft’s last earnings call did.
That moment is a ways off. For now, UBS says its goal is merely to increase productivity and spare camera-shy analysts. But as any spreadsheet jockey knows, the point of productivity gains is that they make revenue go up, or costs go down. Since it is notoriously hard to charge for sell-side research, more attention will turn to the potential for cost savings. Expect the avatars to develop a professional life of their own.
