Andrew Chin is laser-focused on how artificial intelligence can create a competitive edge at AllianceBernstein — not by simply using tools like ChatGPT or Llama, but by applying them in ways that differentiate the firm’s approach to investing and client service. “Everybody has access to ChatGPT, so where’s our edge?” Chin says. “Our edge comes from how we use those tools…”
Chin, who joined AllianceBernstein 28 years ago, became the firm’s first Chief Artificial Intelligence Officer in July 2024. He now leads efforts to embed AI across the global investment management firm, which had $792 billion in assets under management as of December, and reported revenues of $3.5 billion in 2024 . His previous roles — including Chief Risk Officer, Chief Data Scientist, and Head of Investment Solutions and Sciences — have helped guide the firm’s steady adoption of AI and machine learning models over the past decade.
We spoke with Chin about the results AllianceBernstein has seen so far from AI; the biggest opportunities he sees ahead; and how he’s thinking about enabling employees to work smarter.
• • •
How does the company define your role, and where do you sit in the organization?
I report to our Chief Operating Officer, Karl Sprules… It was important for this role to report to an individual with responsibility across the whole firm, because we believe that AI can impact the whole organization, not just one particular area.
This role came about eight years ago [when] I started building our data science efforts…because it was going to provide us with better and faster insights—whether we were making investment decisions, engaging with clients, or managing risk. We’ve been building these capabilities for some time…what has changed over the past year-and-a-half is we realized the impact these tools can have on our decision-making…so we wanted to make sure we are fully invested…which is why the role was created—the first in our industry.
[The mission] is really to transform how we make decisions…with the use of AI. What we really wanted was somebody who knows how to run our business, someone who’s matched portfolios, engaged with clients, done risk management, operations, and things like that. All my experiences were really ideal for that…they inform me as to how we can leverage AI to transform the whole organization.
What has changed over the past year and a half is we realized the impact these tools can have on our decision-making.
Could you share examples or case studies that illustrate how AllianceBernstein is using artificial intelligence?
On the traditional side, you started by saying that AllianceBernstein is an asset management firm. We manage money for the people. Our goal is to hopefully meet our client’s objectives and help them make more money. …My very first use cases were all around investments. How can we improve the decision making on the investment side? We use natural language processing (NLP) to analyze text and extract investment signals from them… [e.g. corporate filings, earnings call transcripts]… to help us make better investment decisions. For example, we look at Chinese transcripts to help us perform better in the Chinese market.
These tools are also very good at comparing things, so instead of extracting things, [we] extract sentiment…
If the CEO said our quarter was great, our revenue grew a lot, that’s good sentiment. If he or she said the last quarter we lost money, we had fraud issues, that’s obviously bad. The way people used to measure sentiment is using a dictionary—so if I look for words which are good versus words that are bad. You can imagine looking at the dictionary—that’s like version one. That was just five or six years ago when that was really state-of-the-art. But…the dictionary is imperfect, because if I look for the word “good,” for example, if I said “not good,” obviously that’s not good. You want to be context aware and look at whole sentences rather than just a couple of words. …We don’t use a dictionary-based approach—we use a context-aware approach to assess sentiment. And then we know what the context is when they are describing a product or a company and that helps us assess the sentiment better.
There’s a lot of data to maintain and analyze. What technology are you using to support your AI tools?
We use cloud platforms like Azure or AWS… That’s where we store the data. Once you get the data in, though, you have to cleanse them… [with] natural language processing, NLP-based, tools because they understand text well. On the AI side, we use a variety of vendor or closed-source models, like Open AI’s ChatGPT [and Google’s] Gemini. We also use open-source models… The most common one, as you likely know, is called Llama from Meta AI.
Hugging Face is an online platform that allows you to test different models…we have access to that to test the latest AI models. When I first started seven or eight years ago, there were literally a thousand models…today there are 1.5 million. So we’re trying to figure out which one to use, how to use it, and where to use it. It’s actually pretty complicated.
One of the things I care a lot about is whether these tools are fine-tuned or customized for finance. Finance is really specific—the words we use, the acronyms… If they’re not, we do some work to make sure they’re customized for the needs that we have.
When scouting for AI vendors, how do you navigate the buy, build, or partner decision — and what would you recommend for other CAIOs?
I would say whoever does the best job wins… Because I started this seven or eight years ago, many of the tools were not built yet, so I built many of the tools in-house. The NLP signals that I described earlier, they were all built in-house and we still use those tools. But as time went on, vendors had much better tools, so we are a lot more open to vendor-based tools. And then when you think about large language models like Open AI’s Chat GPT, there is no way I have a hundred million dollars to spare to do a large language model, so in that instance we would just rely on GPT as an example.
I always let the problem drive the solution… if you don’t have anything in-house, you don’t have any tools [or] expertise and it costs a lot of money to do it, then buy is a really good option… You want to be problem-driven. That’s very important.
Because the AI space changes so quickly, I think some sort of partnership makes sense… We learn from other vendors in terms of how they are developing tools… they learn by applying their tools to very specific problems. So I think there’s a benefit to both, but again I would just let the problem drive the solution.
Some companies are requiring employees to use one GenAI tool. Is that the case with your company?
We’ve built our own version of AI, called ABAI. It’s really our firm’s gateway to AI. We want to make sure that the data we upload onto the platforms stays within our organization. Data privacy, particularly for our clients, is really important for us. I also want to be very careful about the tools our employees have access to. There are 1.5 million models out there—how do you know which one is good and which one is bad? I want to make sure the models we use on this platform are good. That’s why we have this platform that all our employees can use.
But there’s a lot of innovation, and I want to make sure employees have access to some of the latest tools. Our investment teams have access to five to ten other types of AI models. They’re testing them to see if they help make better decisions. Same thing on the marketing side… I want to manage that risk well. At the firm level, we have an AI risk governance committee that ensures we understand all the use cases and that there’s no data loss, and that employees are using the tools appropriately.
So we use a hybrid approach. We give them a way to access these tools in a safe, secure way. I want to make sure that we can change, explore, innovate—and that’s how I think about it.
It’s not AI replacing humans; it’s humans with AI skills, I think, [who will be] succeeding in the future.
How did you address concerns that generative AI might replace employees or make their roles obsolete?
My role is a firm-wide role, so I have AI responsibility for the whole firm which is, by the way, different from other organizations where they may have an AI lead in investments, marketing, or internal operations. The benefit of doing this firm-wide is you have one message: Make sure you share best practices across the organization.
Employees are subject matter experts in what they do, [and] AI augments them… It’s not AI replacing humans, it’s humans with AI skills, I think, [who will be] succeeding in the future.
Is there a business unit in which you are not using generative AI?
In general, we are trying to see how these tools can help everybody be more efficient and more effective, but there are some places where it’s going to be harder. I think about making investment decisions; that’s really hard [and] complex. That’s why there’s a human at the end of that decision-making process. The human needs to be accountable and responsible for all the investment decisions that we make. They may have tools that guide them, but ultimately, the human has to make a decision.
How do you calculate ROI when an AI integration project is deployed into the workforce? What specific ROI data have you been documenting?
…At the end of the day, we are there to deliver performance for our clients. That’s the most important metric… These tools have been able to deliver outperformance for our clients. And then you move to some of the ROI on the operational side. Have they saved money? Did they save time? Many of the processes that we’ve built on the operational and features side save our employees 10 percent or 20 percent, even 50 percent, of their time, which is fantastic. So now they can spend [more] time with our clients.
Given recent market volatility, would you say this environment is a strong test for AI analytics—and how well does AI perform when it comes to fast-moving financial predictions?
I would say everyday is a test—not just the last couple weeks. But I think to your point, on a prediction side, these tools are not good. To predict whether assets can go up or down tomorrow, these tools are not good… because in finance there is a very low signal-to-noise ratio. It’s actually very difficult to determine whether a stock is going to outperform tomorrow or not. This is very different from other GenAI examples. For example, “Can you tell me if this is a cat or a dog?” — those are what I call high signal-to-noise problems, and these tools are very good for that.
…While we don’t have these tools to make predictions for us because they are not good, they may definitely help us synthesize information much faster. When the tariffs were first announced, our analysts were able to use these tools to synthesize the tariffs very quickly to understand the impact on the stocks that they hold. Typically, what may have taken them several days to figure out, understanding the nuances across the countries and across the sectors, they are able to use these tools to help them do the analysis much faster.
What challenges are you facing as you navigate the AI skills market?
I do have a centralized team, and the idea is to make sure that this team helps [spread] best practices across the organization. We also have data scientists and data engineers embedded across the different business units. In terms of talent, what I find is that there are lots of people who are trying to develop the AI skills, which is good, but not many of them also have the skills in finance. In our industry, the combination is really important. Just because you know AI [isn’t enough]. The problems are very different on the financial side. I actually like folks to have experience in that space so that they know how to apply these tools to our particular problems.
In which areas are you using chatbots?
One of the lowest-hanging fruits is the ability to interact with internal documents. We created a chatbot that allows me to search for my own internal documents. When was the last time I did research on x? And what did I say? Can I combine it with what I’ve been doing recently? Can you also imagine a chatbot with our product characteristics? We have a chatbot that allows us to just find out what the fee is, who the portfolio managers are… we can find information a lot faster.
Everybody has access to ChatGPT, so where’s our edge? Our edge comes from how we use those tools—in two flavors. One is prompting, how you ask the questions. Another is fine-tuning, how you train the models with the data that you have.
Looking ahead, how do you see AllianceBernstein using traditional AI, generative AI, and AI agents in the future?
I talked about using these tools to make better predictions going forward. That’s a big one, because at the end of the day, we are here to grow assets for our clients…that will be a big focus for me. You mentioned AI agents — that’s going to be a big one for us, for our industry, and for society. People [will] try to use these tools to make some decisions autonomously. It’s not going to be simple, but I think many people will try to do that.
I think another one is folks trying to see how they can use these tools to create an edge, in terms of what they do. Everybody has access to ChatGPT, so where’s our edge? Our edge comes from how we use those tools — in two flavors. One is prompting, how you ask the questions. Another is fine-tuning, how you train the models with the data that you have.
I really believe the elements we have in-house are really good. So if I can incorporate their insights into an AllianceBernstein-specific model, that would hopefully help us make better investment decisions moving forward.
You must be logged in to post a comment.