ChatGPT brings big data center stage for managers

Money managers say last November’s launch of generative artificial intelligence bot ChatGPT was a game changer, even if they caution it could take years to figure out how the rules of the game are likely to change.

For now, asset managers and asset owners alike continue to boost the number of data scientists in their ranks while working to hammer out the best ways of integrating data science insights into their broader investment organizations.

The bot, and similar tools built on large language models, have given anyone with a smart phone the power to read through hundreds of earnings calls, summarize scores of analysts’ reports or fill out a handful of RFPs, all in the space of a few minutes.

By way of example, the six or seven hours it might have taken for an insurance company to write up a report on a property it was evaluating could be translated to six or seven minutes — a boost in efficiency with broad implications for the world of investing, said Lou D’Ambrosio, a New York-based partner with Goldman Sachs Asset Management and head of the GSAM Value Accelerator team working with the firm’s private equity portfolio companies to bolster their businesses.

It will “change the very being of the way organizations work — the way we invest, the types of investments we have, how we work with our companies for differentiation,” he said.

As of June 30, GSAM reported $2.46 trillion in assets under management.

With natural language processing, the time it takes AllianceBernstein’s operations team to check a 400-page offering memorandum for restrictions that might exclude ERISA clients drops to a few minutes from 15 to 20 minutes — a real cost savings when the company has to comb through about 50 of them a day, said Andrew Chin, AB’s New York-based head of investment solutions and sciences. AB reported AUM of $691.5 billion as of June 30.

Some money managers see the potential ripple effects in epochal terms.

Human evolution has been marked by a small number of critical turning points, such as the start of farming and the eclipse of manpower by machine power, noted Vlad Zdorovtsov, director of global equity research with Acadian Asset Management LLC, a Boston-based quant firm with AUM of $100 billion. Ten years from now, the breakthroughs being seen in “deep learning” may end up being recognized as a similarly important moment in the evolution of asset management, he said.

Money managers credit ChatGPT’s mastery of language with opening the floodgates for artificial intelligence over the past year.

The ease of accessing data just by asking the large language model a question is what’s led to a breakthrough, said Dana Jackson, chief information officer with Franklin Templeton, a San Mateo, Calif.-based manager with $1.43 trillion in AUM.

ChatGPT’s conversational interface effectively opened the door for “AI getting to everyone” — a watershed moment, agreed Jordan Vinarub, head of the New York-based Technology Development Center that T. Rowe Price Group Inc. launched over six years ago to build new capabilities to support continued growth.

As of June 30, T. Rowe reported AUM of $1.35 trillion.

For now, however, much remains unclear about how money managers will integrate those capabilities into their businesses.

Asset managers and wealth managers are already using it or will soon use it because with the advantages it provides — in terms of time savings and ultimately cost savings — they have no choice, said Christopher Geczy, a professor of finance at the University of Pennsylvania’s Wharton School, as well as founder, CEO and CIO of West Conshohocken, Pa.-based money manager Forefront Analytics LLC.

“You can’t run and hide … but there’s a fine balance between being bold and reckless,” GSAM’s Mr. D’Ambrosio said. “You want to be very deliberate with this technology. It’s powerful stuff … do it in contained places. You want to try, you want to learn, you want to repeat, you want to accelerate,” he said.

Roughly nine months since ChatGPT burst onto the scene, such exploration remains very much a work in progress.
It’s clear that “people like to play with it but they haven’t really mapped it into what’s useful for them yet,” Mr. Geczy said.

Not for lack of trying.

“We’re knee-deep in figuring out how to use (large language model-based tools) effectively,” even as their capabilities continue to evolve at an ever-more rapid pace, said Michael Masdea, co-head, risk and investment science with Wellington Management Co., a Boston-based manager with AUM of $1.2 trillion. At the same time, “there’s all kinds of other aspects that we need to think through … especially around the data, proprietary data, sharing our data,” he said.

T. Rowe Price’s Mr. Vinarub described a similarly frenetic environment. “The technology landscape is changing so rapidly” — more week to week now as opposed to month to month or quarter to quarter, he said. Amid that maelstrom, “everyone’s trying to learn what can you do and what are the implications,” he said.

Franklin Templeton, for its part, has “a number of pilots underway” putting large language model capabilities to work across investment management, distribution and operations, Mr. Jackson said.
In line with those pilots, the firm, with 25 data scientists in technology at present, is revising upward its “material” hiring plans, he said.

And the bull market for data scientist candidates isn’t confined to the U.S. Mr. Jackson, for example, noted that his firm has sought out Ph.D.s in that sector as far away as Hyderabad, India.

Large and midsize money managers are continuing the buildup of their data science teams that began in earnest five or six years ago, said Richard Dell, a London-based partner with Mercer and head of equity manager research for the firm globally. “Most, if not all, have some in-house data science capabilities which they use to augment their traditional bottom-up analytical capabilities,” he said.

A growing number of big asset owners are likewise joining the race for talent, even if they remain a few years behind money managers in that regard, Mr. Dell said.

“I would be amazed if most large asset owners are not building out data science capabilities,” he said. “The support they can provide to organizations, whether that’s in generating efficiencies or in generating insights,” is huge and hiring has picked up considerably over the past five years, he said.

Japan’s Government Pension Investment Fund became one of the latest heavyweights to join the fray over the past year, with the ¥219.2 trillion ($1.52 trillion) pension fund’s most recent annual report, out in July, noting that it had secured changes in regulations constraining the organization’s salary payouts in order to compete for highly specialized data scientists.

Eiji Ueda, GPIF’s chief investment officer, in his review of the year said the fund’s globe-spanning portfolio gives it advantages in collecting valuable data and GPIF will work to field a world class data science team capable of squeezing added returns out of its investments without taking on more risk.

Mr. Ueda, in an emailed response to questions, said he regards portfolio risk management as a “matter of science” and data science capabilities in particular as capable of improving returns for the fund’s traditional and private market exposures alike — “not only in manager selection but also in rebalancing, index selection and others.”

“We believe we can turn … precious data into better risk-return” for GPIF’s portfolio, he said.

Asset owners are effectively finding themselves in a similar position to money managers now — focusing on the applications that newly powerful AI tools can be put to use for, money managers say.

Work with clients helping them think about how to take advantage of new opportunities has picked up considerably over the past 18 months, reflecting the rapid maturation of available AI tools, noted Kristian West, a New York-based managing director and head of investment platform with J.P. Morgan Asset Management.

Even three years ago, it would have been unheard of to have the kind of capabilities offered by tools such as ChatGPT on your phone, Mr. West said. Asset owners aren’t looking to build their own large language models — “it’s about applications … how you apply your own proprietary data to that model and then how do you put it into your process,” he said.

Similar challenges are facing asset managers when it comes to integrating data science capabilities into their larger investment programs.

J.P. Morgan Asset Management, which reported $2.8 trillion in AUM as of June 30, has roughly 25 data scientists and another 20 focused on machine learning engineering but with access to a broader pool of 900 data scientists and 600 machine learning engineers across the broader J.P. Morgan Chase & Co. franchise.

With powerful machine learning tools becoming so much more accessible now, J.P. Morgan Asset Management wants to take advantage of opportunities to boost operational efficiency “as much as anyone else and it’s something which we are actually focusing on and trying to do,” Mr. West said.

Laying the groundwork for JPMAM to best leverage the firm’s data science capabilities requires creating an “ecosystem” or environment where the team can be used and utilized on very specific use cases but with enough risk controls in place so that a far broader set of individuals can apply themselves to the challenges at hand, Mr. West said.

“You don’t say, OK, those 40 people, you do that,” he said. “What you want is as many people as possible to do that …We are very strict around a process so that as many people take advantage as possible (on) multiple areas of opportunity around operational efficiency, investment capabilities and client enablement,” said Mr. West, adding “That’s how you get scale. That’s how you get the ability to take advantage of this as quickly as possible.”

Mr. Masdea said Wellington has likewise worked over the past five or six years to calibrate the best way to get data science insights into the hands of the firm’s more than 50 investment boutiques, as Wellington’s investment teams are called.

Wellington’s “investment science” team — launched roughly six years ago with 30 people after a periodic review of industry trends concluded that data science and analytics were poised to revolutionize the industry — has grown its head count to 70 now, he said.

Working toward the goal of “catalyzing” Wellington’s fundamental active investment teams, the firm’s leadership concluded that two prominent models — forcing teams to bring in data scientists or, alternately, having a satellite data science team that would lob in ideas for those fundamental teams to consider — were both a poor fit for Wellington’s culture.

Instead, “what we needed to do was create a partnership — one that created a cultural equivalency for scientists and investors,” Mr. Masdea said. “So, we established what we call partnerships where we won’t do work with someone unless they’re willing to kind of commit to treat it as a partnership,” he added.

“Creating an environment where data scientists and investors, or ‘artists,’ could come together as partners — that’s our goal, that’s our special sauce,” Mr. Masdea said. And one element of that was to create “embedded scientists” who would sit with fundamental teams for a year or more, coming to understand their philosophy and processes and better understand what the most impactful data science capabilities could be for them, he said.

At present, six investment boutiques at Wellington include embedded scientists, Mr. Masdea said.

T. Rowe’s technology team, by contrast, is located in New York, a few hours away from the firm’s headquarters in Baltimore. Despite that distance, a “long legacy of collaboration and collegiality and a team-based approach to everything” has worked to ensure T. Rowe’s discretionary, active investment teams continue to get timely access to the data technology tools they need, Mr. Vinarub said.

For example, a weekly “demo” meeting every Friday brings together a coalition of business and technology groups to discuss what’s going on, “what we’re learning ... what transpired this week in technology ... and the implications,” Mr. Vinarub said. “We work together actively all the time,” an effective means of keeping abreast of developments when things are changing this quickly, he said.

At the end of the day, however, most money managers remain confident that human beings will continue to have a future in money management.

AI tools will support portfolio managers, not replace them, predicted George N. Patterson, Boston-based managing director and chief investment officer with PGIM Quantitative Solutions. The greatest benefits may be found in the “non-sexy parts of the business,” he said, such as making back-office operations more efficient or improving data efficiency. Newark, N.J.-based PGIM Quantitative Solutions reported $98.9 billion in AUM as of June 30.

“When you think about the nature of the actual problem of investing, of choosing a security that is going to drive a certain amount of the return to fill a client’s needs … so far machines … haven’t shown the capability to do that very effectively,” Mr. Masdea agreed.

That, in part, reflects “the ever-changing nature of the relationships between security prices and the underlying data, the amount of data, the stationary aspects of data, the adversarial aspects of data” — which make machine learning and AI-based techniques not very effective at the actual process of investment, he said.

“Super effective, however, at all the surrounding pieces of investing,” Mr. Masdea said.