Why won't anyone define AI?
Why won't anyone define AI?
Beware the message that artificial intelligence can do practically anything
A month ago Citywire Selector published a story with the irresistible title "CEO Summit: Efficiency is not stupid". I mean, the jokes just write themselves, right?
Incidentally, the story was about stuff that writes itself, in that it was covering a discussion by six asset management CEOs on artificial intelligence (AI):
"For example, one of the CEOs talks about how AI is helping their asset management group answer RFPs (Requests for Proposals) which sometimes can have as many as 1,600 questions."
We'll leave the obvious question, of who designs an RFP with 1600 questions, for another day. Today is all about AI, and the extent to which it can fix the efficiency problem for asset management.
A big report came out this week from the Boston Consulting Group (BCG), titled "AI and the Next Wave of Transformation". Really, it's a report in two parts: the first being about the state of flows for asset management; the second about whether AI will be the industry's great efficiency savior. The thread connecting the two parts comes in the form of a simple stat: costs have gone up up up.
In 2010, costs as a percent of revenue stood at 66%. By 2023, they had risen to 70%. All the while, fees continue to shrink. (I cover more commercial insights from this report below, in Treasure Corner.) The solution, declares the report, is clearly AI:
"AI is being built into a variety of tools that asset managers can use to improve their operations. The power of such tools comes from AI’s ability to rapidly collect, synthesize, and analyze vast amounts of data from internal and external sources and then generate information on the basis of patterns found in the data."
Now, the crucial point is this: this report - and many of its kind – never defines what AI actually is. This is no accident. We talked about a Mercer report, which asked fund managers to what extent different technologies are "AI". The more managers were using a technology, the less they considered it as “AI”.
The same thing is happening here. There's Generative AI, but the report makes it clear that's different from AI:
"The subset of AI known as generative artificial intelligence (GenAI) has the ability to interpret and analyze unstructured data from a wide range of sources and create original content... Both AI and GenAI are becoming critical to asset managers."
The vague language allows for baroque overstatements of what technology can deliver:
" An AI model, however, can compare portfolio holdings and their risk levels, optimize runoff profiles, and establish new risk thresholds all in one step." (highlights mine)
You'll notice a reoccurring theme, where “AI” is suddenly and magically able to pull off complex tasks with no effort and in no time. The problem here is that while Generative AI (or large language models) are indeed new tech, the rest of it is just...the stuff we had before? Remember, for example, when "data science" was all the rage? When "data was the new oil" and data scientist was the sexiest job of the century?
The result of this obfuscation is genuine confusion. This week ETF Stream published a viewpoint by Banca Patrimoni Sella & C, an Italian asset manager, on the subject of AI in finance: An ETF buyer's perspective.
In it, the writers enthuse that "AI is capable of data processing on a scale that would have been unthinkable just a short time ago." and that "the decision-making process of an AI is not affected by biases and noise, as is the case with humans". Yet one paragraph down, they also point out that:
"AI algorithms, in turn, can be fooled by the poor quality of the data they ingest and come to the wrong conclusions. Indeed, their weakness is that they always generate an answer even when there is none."
In its report, BCG is speaking from position. The company recently stated it expects 20% of its revenues this year to come from AI consulting. The other thing is that I, too, speak from position. FinText has been delivering AI training to financial services for over a year. But this isn't really about squabbling over clients:
When kids are taught to read and write, it's done when they're young, so they can make the most of these skills as they move through life. We don't obfuscate reading and writing, because we know it's not going to help them learn.
AI is so poorly defined because, so long as it stays that way, companies can promise it can do anything and everything. But you don't have to fall into this trap:
Generative AI is new. There are some things it makes radically easier, but its results are fuzzy. There are ways to do quality control and integrate data and reap benefits. Some are easy, others are harder. Fiddling with data never stopped being hard — and it was never really that sexy. It will probably continue to stay that way.
Treasure Corner: The long view
Let's dig into the asset management side of the BCG report. This is really where the study shines, because BCG has been tracking flows data for over twenty years. If anyone from BCG is reading this: Guys, you have good charts, next time make them bigger! These were some of the tiniest charts I've seen in a while.
The analysis is based on a global benchmarking study of 80 asset managers, representing $69 trillion in AuM, or about 60% of global AuM. By and large, this is US data. Notice this is all mutual funds - not ETFs.
The first striking finding, which I've never really seen featured quite like this, is just how much passive has grown in appeal. Usually, charts show the steady climb of of passive AUM, to the point where it's overtaken active. But in this report, they break down net flows to passives by decade (roughly):
Between 1990 and 1999, 8% of net flows went to passive. Between 2000 and 2009, 26% of net flows went to passive. Between 2010 and 2023 84% of net flows went to passive. That is a big number.
Then there is industry concentration. In passives, it's super concentrated: in 2023 positive net flows to the ten largest passive mutual funds was 95%. They made sure to pop in the chart the other 5% for "Rest of the industry", just to bring home the message. For actives, concentration is also growing, but the situation is better: in 2010 55% of positive net flows went to the top-10, compared with 67% in 2023.
Finally, and I quote, "fewer new products are surviving despite attempts at innovation": in 2010 60% of mutual funds reached the ten-year mark. In 2023, that number is 37%.
These are grim stats. Come to think of it, maybe that's why the charts were so small.
But the context is bigger. This week, the Financial Times published a story about ETF flows being positive for 58 consecutive months. We've talked about it on the very first week of the year, and said at the time:
"(ETFs and mutual funds) aren’t two different industries, just one industry in a state of flux. Also, an industry that is, on the whole, growing. When the focus changes from choosing a side to the flux between sides, an altogether more interesting story emerges.
Overall, demand for investment products is up. Demand for marketing investment products is up. Things are changing, but they’re not looking bad."
They're not looking great for mutual funds, though.
Also Happening:
Jim Simons, founder of Renaissance Capital, passed away this week. Lots has been written about the man who summed up his own life with: "I did a lot of Math. I made a lot of money. And I gave most of it away."
Most of the stories focused on the money part. But my favorite coverage was by Institutional Investor, where a reporter recounts interviewing Simons twenty-five years ago. And the most important thing they wanted to say was that the guy was a nice man:
"Witty and down-to-earth, he was actually very open about the basic structure of the firm and its operating principles. In terms of the details of the trading strategies though, he deflected all questions. He did try to keep me from writing anything really stupid...
Very often when you write an extremely positive story, you will get a call from someone who feels the need to convey that you’re an idiot who got suckered. In the case of the Renaissance story, one of Simons’ relatives got in touch and said what the story missed was how nice and generous he was to his extended family. You don’t get too many of those calls."
Before large models became AI, they were just a part of a field called Natural Language Processing, basically how machines understand language. Seconds before LLMs burst into the world, we published a report, How Finance Uses Natural Language Processing, made of eight case studies of investment banks and asset managers, and the problems they were trying to solve with the tech. While the tech has evolved, the problems are mostly the same.
The FCA is comically shocked at the industry's response to its proposed 'name and shame' policy. We talked about the regulator's plan to disclose its investigations earlier. (Currently, it's only after an investigation has been completed.) As expected, the plan sparked a whole lot of backlash, under the guise of hurting London's competitiveness among international markets. Yet FCA Chair Ashley Alder insists they weren’t expecting such a “stern reaction”.