Vanguard once put a number on what a financial adviser is worth: about 300 basis points a year, across six or seven categories of work. Devin Ekeberg said AI is already doing six of those categories, and doing them better.
The usual response to that is a defense of the whole job. Ekeberg, who runs adviser education at PIMCO, gave up six-sevenths of it and argued the remaining category — behavioral coaching — is the one that pays for the relationship.
"Currently the AI models will spit out, a 35page response in what a client should do when the adviser might recognize it's like, hey, this client's having an emotional response before we make a decision about their portfolio. Maybe we should actually just take a walk around the block first, right?"
Ekeberg spends his time with advisers in the field; his co-guest Anastasia Buyalskaya sits on the other side of the same problem, presenting to PIMCO's quarterly investment forums and helping design how the firm's own investment committees make decisions.
The full interview is covered here so you can skip it. 51 minutes of audio, 20 minutes of reading.
Here are the 17 principles that matter.
👤 Guests: Anastasia Buyalskaya, Assistant Professor at HEC Paris and a behavioral finance consultant to PIMCO; and Devin Ekeberg, Senior Vice President on PIMCO's advisor education team
🎙️ Host: Greg Hall, who leads PIMCO's wealth management business in the United States
📰 Published: 14 September 2026 on YouTube
🔴 YouTube | 🟣 Apple Podcasts | ⏱️ 51 min | ✅ Time saved: 31 min
Key Takeaways
Vanguard put the value of an adviser at about 300 basis points a year, and Ekeberg says AI already covers six of the seven categories behind it
The one left is behavioral coaching, which he does not expect software to take
The fix for bias in an investment committee is meeting structure, not self-awareness
The most senior person speaks last, and views are collected by survey before anyone talks
Committees work at three, five or seven people and stop working above that
Thirteen people in a 45-minute meeting get two to three minutes each, and nobody is accountable for the call
Emotions are information, and the test is whether the emotion came from the market or from your commute
Debiasing people does not work; putting guardrails around the decision does
An adviser's best argument against a client's panic is the client's own earlier self
Large language models are built for engagement, which makes them agree with you
The counter is to prompt for a devil's advocate case against your own plan
Buyalskaya sees a new bias in AI use: a one-page answer that feels complete
She catches it by asking about her own field, where the model invents papers and biases
Behavioral awareness has not made another 2008 less likely, because the people who lived through crises leave the market
1. What Behavioral Finance Is
Greg Hall opened by saying PIMCO has spent the last few years deliberately building behavioral finance into its own investment process, then asked Anastasia Buyalskaya to explain the field.
She dated it to one paper. "And I would say that the field really started in the 1970s with the publication of prospect theory which is a paper from Daniel Kahneman and Amos Tversky"
It came out of a gap between the model and the behavior. Expected utility theory makes assumptions about how people treat gains and losses; when researchers asked real people to make simple versions of the decisions clients face, risk preferences changed with the context and the reference point changed the whole evaluation
Hall asked whether this was the debunking of what he had been taught — "the myth of the perfectly rational actor" from Econ 101, with total information and smooth curves
Buyalskaya said it was, and that the myth is still being taught. "It totally debunked that myth and I also was taught that and I think a lot of students are still taught that"
She read Kahneman's Thinking, Fast and Slow on publication in 2011, after his 2002 Nobel Prize, and said she did nothing else until she had finished it
Hall's aside was that the average adviser listening is unimpressed, since they could have said the same thing thirty or forty years ago
2. The Markowitz Blind Spot
Devin Ekeberg took the same history back two decades further, and made it worse.
"Well, it's actually a little bit worse than that," he said. Modern portfolio theory dates to Harry Markowitz in the 1950s — also a Nobel Prize, and the source of mean-variance optimization
The theory is not a historical artifact; it is still the plumbing. "Every asset management firm in the world uses some element of modern portfolio theory to construct their portfolios and so forth"
The 1970s work did not replace it, it added a second axis. There is now a balancing act between the mathematical side of investment management and the human side, and "if you're ignoring one or the other, you can end up with some pretty terrible outcomes"
3. What the GFC Taught Her
Hall asked what in Buyalskaya's own experience had primed her for Kahneman's book.
She started her career just before the global financial crisis, on a trading floor. She was not managing a profit-and-loss account and described herself as a fly on the wall
What the crisis destroyed was a belief about expertise. She had assumed that an expert with a clear financial model faced very little uncertainty; those illusions "were totally popped"
The interest that followed was specifically about experts, not amateurs — that people with excellent analytics and robust models are still deciding under uncertainty and still subject to bias
4. The Biases That Matter
Hall asked which biases actually show up in decision processes.
Buyalskaya said there are more than a hundred documented biases and that memorizing them is not the job. There is a core subset worth knowing, and she was curious whether it matched the subset Ekeberg teaches advisers
Overconfidence is the one PIMCO's own chief investment officer returns to. "I mean, one that you'll hear Dan Ivascyn talk about a lot is overconfidence"
The reason is structural: "So anybody that's an active manager by nature is out there trying to beat the market, right? It's almost the definition of overconfidence"
The remedy is an attribution system that separates skill from luck, so confidence can be reined in where the skill is not there
Confirmation bias has two forms, and she distinguished them. One is going out to find supporting data — "So I'm actually going to cherrypick data maybe to confirm my view and it's not going to be a representative subset of all the data available" — and the other is reading identical data differently because of the view you already hold
The tell is which research an investor forwards you. "So, if you talk to an investor, they're much more likely to send you the sellside research that supports their view, right?" The same selection operates on who gets hired onto a team or put on a committee
Ekeberg put the count higher and said the categories matter more than the list. "I think the last time I read I think academics have identified something like 150 different biases," all of them repeatable under test
5. Emotions Are Information
Ekeberg sorted the biases by which part of the brain produces them, then asked Buyalskaya whether the split still holds academically.
He used Kahneman's system one and system two. The automatic part produces the emotional biases that are hardest to overcome; the cognitive part, where information is received and processed, produces its own
His practical reason for the split is that the prescription differs. Identifying exactly which bias is operating is difficult, but working out which half of the brain it came from is easier, and an emotional response needs a different remedy from a cognitive one
Buyalskaya said emotions get a bad reputation they have not earned. Citing her former Caltech professor Ralph Adolphs and his book with David Anderson, she described emotions as functional states: "So rather than something kind of that's distracting you, emotions are information"
The test is the source, not the feeling. Anger about the debt crisis may be informative and may belong in a portfolio decision. "But anger because somebody cut you off on the highway on your way to work probably shouldn't inform the portfolio allocation"
6. Structuring the Committee
Hall asked what mechanisms actually work against bias inside an investment committee.
A group setting introduces its own biases, Buyalskaya said — authority bias, and groupthink where everyone is pulling from the same sources
The theories that promise better group decisions assume something groups rarely satisfy. Wisdom of crowds and collective intelligence both assume uncorrelated inputs, which an investment committee often does not have
She rejected the usual denial. Investment teams sometimes tell her authority bias is not a problem because the chief investment officer is not intimidating; her answer is that the fix is structural rather than personal
The structure is specific. "The solution is just to put some structure into the investment committee meeting such that the most senior person is the last one to speak," with protected air time for junior and contrarian voices
Simple pre-meeting surveys anchor people on their own views before the discussion starts
She named Tiffany Wilding's regional committee, which surveys the whole team on macroeconomic variables and looks at the distribution before anyone speaks, then deliberately pulls on the tails
The point of all of it is to avoid debiasing, which she says research shows does not work. Guardrails around the decision, not correction of the person
7. Three, Five or Seven
Hall said Buyalskaya has strong views on committee size and asked her to set them out.
"And so the magic numbers that I always talk about are three, five, and seven" — drawn from the governance literature
"Odd numbers work quite well because you have a tiebreaker," which matters once a committee uses surveys or formal votes
Above seven, accountability dissolves. Her example is a committee of thirteen people, all based in the United States and all fixed income experts, making a call on interest rate sensitivity — and nobody identifiable to answer for it if the call is wrong
The air-time arithmetic is the second reason. Thirteen people in a 45-minute meeting get two to three minutes each, which she said is not the best use of those people, and the views being dropped are usually close to views already in the room
8. Deciding in Advance
Ekeberg moved the same idea into the adviser's practice, where the decision is made before the emotion arrives.
Advisers are building systems that act as a default decision. He drew the comparison from his brother in law enforcement, who is trained to manage emotion in favor of a standing operating procedure, on the theory that a decision made in advance is a better decision
Clients arrive loaded with input they cannot act on — headlines, wars, inflation, tariffs — so "And so advisors are sort of acting almost a little bit as therapists"
The technique is reframing rather than suppression. Instead of fear about the portfolio, the adviser tries to find something the client can be interested in: asset classes that look more attractive than they did, and small changes rather than drastic ones
9. Speed Versus Quality
Hall noted that advisory firms are now mostly teams rather than sole practitioners, and asked whether smaller groups face a different dynamic.
Buyalskaya framed it as one trade-off, in both directions. "Well, I think just recognizing that there's always a trade-off between decision speed and decision quality"
Adding people slows the process down, deliberately. That usually buys a better decision, but markets move and clients lose patience, so the cost is real
Her diagnostic is whether the extra people change the answer. If three people's input already gets the group to the decision, the room is too big; if there are open questions and the decision still has to be made, it is too small
10. Premortems and Depletion
Buyalskaya endorsed Ekeberg's point and named the practice behind it.
The technique is deciding in what she called a cool head state. Before a drawdown, the client is asked what they would do if the portfolio fell 20%, and how they would change the allocation
She grounded it in professions that do this formally. Medicine, law enforcement and the military use checklists and premortems because they know nobody keeps 100% of their cognitive capacity in an emergency
Ekeberg added the physiological reason. Engaging the cognitive part of the brain depletes a finite resource: "It's very difficult to maintain that cognitive energy all day long"
He cited the research on when decisions get made: "And you've probably seen some of the studies on even judges in the courtroom will make certain decisions in the morning and very different decisions in the afternoon because their case load is so high and they're just completely drained"
The pairing he recommends is a premortem plus a red team. "The premortem exercises, the red teaming, which is sort of a sense of trying to challenge your assumptions when you're at your best" — followed by a deliberately high bar for changing the decision later
11. Recruiting the Client
Hall turned the discussion on himself, describing how his own advisers handle him.
A written policy statement recruits the client into the decision. When Hall changes his mind three weeks later on the strength of a headline, his advisers remind him of the plan he agreed to
He said the value is ownership rather than persuasion. It reminds him that a cooler version of himself signed off, and gives the client a sense of accountability for their own plan
Buyalskaya uses the same move with institutional investors. When an investor objects that she is telling them what to do, her answer is: "No, your former self is telling you to make this decision because you committed to this, 3 weeks ago"
She said it works better than arguing without a stake in the outcome
12. Curated Information
Hall picked up Ekeberg's word — clients being "fed" information — and asked how the choice of sources changes behavior even in a field as ostensibly objective as finance.
Ekeberg said the curation is heavier than people notice. "So everything you read on social media or through the news and things like that is pretty heavily curated," and the best practice is to know the source, how you are processing it, and why it reached you
His example was his 16-year-old daughter, whom he tells there is "a billion dollars worth of engineers" on the other side of the app working to keep her scrolling
Buyalskaya described three stages of the same problem. At a newsstand you made an active choice between two papers and knew which prior you were confirming: "I still remember when you'd go to a news stand and you would make an active choice between two newspapers"
Digitally intermediated news replaced that with an echo chamber of things similar to what you have already read
AI is a further step, because it does not surface its sources at all — the citations exist somewhere, but nobody opens them
Her conclusion was a question rather than an answer. "And so I think we're getting more and more removed from those primary sources," and whether that fuels overconfidence and confirmation bias or improves decisions is, she said, still open
13. The Platform's Imperative
Hall extended the point from media to the adviser's own employer.
He said consolidation in the registered investment adviser market has left few advisers genuinely independent. Most are affiliated with a larger platform
A platform has its own imperatives it is pushing through the adviser network, which sits alongside the adviser's own judgement
His description of the resulting tension was sympathetic. Most advisers he knows are loyal to their firm, happy on the platform, and still entrepreneurs and independent thinkers trying to balance the internal voice against the external ones
14. Gamification and Dopamine
Hall asked Buyalskaya to expand on how much of a decision is dictated by which button is in front of the person making it.
She has been researching gamification, which she said is now everywhere, financial decisions included. Digital broker-dealer apps increasingly use it
The mechanism is dopamine, and the rewards do not have to be real. Once you know how sensitive the brain is to a dopamine release, the effectiveness of vapid rewards stops being mysterious
She gave the case for it too. Health and fitness apps gamify well-being for good ends, and she said gamified brokerage apps have brought parts of the retail market into markets they had not entered before: "So I do think that gamification has led to some democratization if you will"
The limit is where the same mechanism turns into churn. "But I think understanding when that could lead to overtrading when that could lead to sort of excessive risktaking that you later regret that's really important"
Hall was blunter about the word. "It's used freely and sometimes it's employed by firms that they claim like Prometheus bringing fire from on high" — and his warning to advisers was that whoever brings the gift usually has something to gain from the client accepting it
Buyalskaya's answer was that this is the adviser's edge. The apps are good at reaching the reflexive part of the brain; a good adviser engages the reflective part, with checklists and premortems, so the client makes decisions they are unlikely to regret
Ekeberg read the same facts as an argument for professional advice. The adviser's job is to take abstract information a client cannot act on and connect it to what they actually care about: "It's connecting all of that performance to the stuff that client cares about" — liquidity for a purchase, college fees, a retirement lifestyle, a legacy. "So I would be very optimistic if I was a financial adviser in the face of all of that stuff that's been so disruptive"
15. Is AI Good for Decisions?
Hall noted that a story about an AI company shipping a tax-planning module had knocked wealth management stocks, and asked whether AI will improve decisions or prey on the weaknesses the conversation had been cataloguing.
Buyalskaya said behavioral scientists are on both sides of it, then described the design problem: the models are built for engagement, so they are trained to be agreeable
The flattery is the product working as intended. "So they do flatter you and they do have a tendency to find data and evidence to support what you want to hear" — which lands directly on confirmation bias
The counter is to write the prompt against yourself. She asks the model to play devil's advocate and explain why a financial plan is likely to fail, or to attack an investment thesis
"So there's a big area of prompt engineering and I think it takes a little bit of experimentation because some of these tools are very sensitive to exact nature of the prompt"
She has seen the tools used well. "But I've seen investors use these tools for good, to use them to question their own decision-making, use it to, run various counterfactuals that otherwise would be very time consuming for human analysts, let's say, to run"
The new bias she flagged is the illusion of explanatory depth, from a paper by Lisa Messeri and Molly Crockett that she recalled as being called AI and illusions of knowledge. "So things like this illusion of explanatory depth, which I love, which is the idea that you will ask AI, a question and the response will feel very sort of self-contained and very well written"
"And their point is you actually probably understand the tip of the iceberg and you might be a little bit overconfident because there's a whole bunch of stuff that you don't understand"
Her own test is to use it inside her own field. "So I'll use AI a lot and ask it questions about behavioral finance and it'll sort of make up papers, right?" — a failure she can catch there and would not catch elsewhere
Hall's observation was that the models answer by weight of evidence, not quality of it. "I mean the large language models I've noticed, they will return an answer based on the volume of writing about a given topic," so "And so if a viewpoint just has more writing about it, then it becomes the answer"
He compared it to a colleague who was the most articulate arguer he had met and frequently wrong: "It looked good. It sounded good, but it didn't end up being the right answer"
16. What AI Takes From Advisers
Hall asked Ekeberg whether advisers are using AI for meeting preparation, and the answer turned into the episode's sharpest claim.
Ekeberg said advisers have always chased efficiency and that the current wave has a precedent. "A lot of people were scared that the robo advisers were going to take over the real thing"
He reached for Vanguard's attempt to price the adviser. The Advisor's Alpha study quantified the value of a financial adviser across six or seven categories — portfolio allocation, withdrawal strategies, tax strategies, behavioral coaching and others. "They landed at around 300 basis points a year"
Then he conceded most of it. "Now it occurred to me when you look at that those lists of categories like six out of the seven things that they identified that advisers took credit for providing value AI is already doing and probably doing better than a human adviser is doing"
The exception is behavioral coaching, which he said he does not expect a model to take over
"Currently the AI models will spit out, a 35page response in what a client should do when the adviser might recognize it's like, hey, this client's having an emotional response before we make a decision about their portfolio. Maybe we should actually just take a walk around the block first, right?"
Hall said the robo-advisor era produced the same lesson. Consultants told the industry it was over; what followed was that clients crossing certain wealth thresholds turned to a human being to bridge the logical and the emotional
17. Why 2008 Could Recur
Hall closed by asking whether the spread of behavioral awareness has itself helped keep the economy out of a crisis like 2008.
Buyalskaya declined the compliment. "I think that's a very generous hypothesis"
Her objection is turnover, not ignorance. "But I do think memory can be short and obviously the composition of people in markets changes every decade"
She pointed to research on experience effects. Reading about the Great Depression does not carry the weight of having lived through it, and the work she cited from Malmendier shows what people lived through has a lasting influence on how they think about risk and allocation
The implication is a schedule rather than a forecast. "So, it's possible that as those of us that remember some of these big downturns kind of leave the market, right, and retire and more of those actors are leaving the market that we do get another crash"
What she would concede is awareness, not immunity. Far more market participants know about behavioral finance and their own biases today than twenty years ago
Hall's closing observation was the same one from the adviser's side. There is a palpable difference between advisers who lived through the Asia crisis, Russia, Long-Term Capital, the dot-com bust, the telecom bust, Enron and the financial crisis, and people extrapolating good times forward without a thought for what could go wrong
Bonus Insights
Buyalskaya has co-authored and ghost-authored PIMCO's macro signposts column, which is usually written by Tiffany Wilding, and presents to the firm's quarterly investment forums
Ekeberg is based in Colorado on purpose. Hall's line was that it is "basically the only place you can get to everywhere in the country in a three-hour flight"; Ekeberg's answer was that he can reach anywhere direct
Hall admitted to his own gamification. "I check my I check my ring every morning to know how I'm feeling. I used to just wake up and know how I was feeling, but now I have to check an app to find out"
Hall declined to end on the bearish note he had opened. After the exchange about who remembers past crises, he said fixed income people tend to be pessimistic and that the conversation had been too interesting to finish there
The bottom line from both guests is that structure beats self-knowledge: the biases are documented, the attempts to train people out of them do not work, and what does work — a survey before the meeting, a committee of five, a premortem, a plan the client agreed to when calm — is procedural, which is also why Ekeberg thinks the one part of an adviser's job software cannot copy is the part that happens when a client is frightened.
Products, Companies & Tools Mentioned
PIMCO (The firm behind the podcast, which Hall said has spent the last few years embedding behavioral finance in its own investment decision-making and which employs both guests)
HEC Paris (Where Buyalskaya is an assistant professor, the academic half of what Hall called the academic and applied sides of the subject)
Caltech (Where Buyalskaya was taught by Ralph Adolphs, whose work on emotions as functional states she uses against the idea that emotion is superfluous to investing)
Fitbit (Her example of gamification used for good — health and well-being apps running the same dopamine mechanism as brokerage apps)
Books & Resources Mentioned
Thinking, Fast and Slow – Daniel Kahneman (Published in 2011, a decade after his Nobel Prize; Buyalskaya said she read nothing else until she had finished it)
Prospect theory – Daniel Kahneman and Amos Tversky (The 1970s paper Buyalskaya says opened the field, by showing risk preferences change with the reference point)
Vanguard's Advisor's Alpha (The study Ekeberg cites for the roughly 300 basis points a year an adviser adds, across the six or seven categories he says AI now covers six of)
Artificial intelligence and illusions of understanding in scientific research – Lisa Messeri and Molly Crockett (The paper Buyalskaya recommended, recalling the title as AI and illusions of knowledge; the source of the illusion of explanatory depth she warns about)
The Neuroscience of Emotion – Ralph Adolphs and David Anderson (The book Buyalskaya credits, without naming it, for the framing of emotions as functional states that carry information)
Ulrike Malmendier's research on experience effects (The work Buyalskaya cites for why lived crises change financial behavior and read-about ones do not)
If this was worth your time, send it to someone closer to the industry than you are.
Get the latest market chatter as it happens: