Ramon Richards landed at Baltimore-based T. Rowe Price at an interesting moment in the evolution of technology.
2023, when Richards joined the global asset management firm as its Chief Technology Officer, was the year generative AI hit the mainstream.
“When I arrived, we had a team that [had been] focused on building AI solutions at T. Rowe since 2017, and that team was already exploring and experimenting with what generative AI could do for our business,” says Richards. “The enthusiasm and excitement around generative AI was really taking shape.”
Since then, T. Rowe has created tools that help its employees reduce time-consuming tasks, get better insights from its data, and improve client service.
Founded in 1938, T. Rowe has more than 8,000 employees, and manages $1.79 trillion in client assets. In an interview in late December, we spoke with Richards about AI training, use cases, and measuring return-on-investment.
• • •
You started working at T. Rowe Price a little more than two years ago, which is less than a year after OpenAI introduced ChatGPT. Can you walk me through your approach when you began the job?
When I joined, T. Rowe already had a team that was exploring and experimenting with generative AI to look at what the technology could be to our business. …My approach was, let me first understand T. Rowe’s priorities. I wanted to understand how AI could help advance some of the priorities, and understand the technology foundation and what was our readiness to do more.
While there’s a lot of excitement and enthusiasm around AI, having the right governance structure in place is important, [and] having your data ready and prepared and managed well is important.
…In our investment space, we were focused on can we use these tools to provide deeper insights on our own data. In our distribution space, we were exploring how AI could help our sales personnel with some of the work they have to do in having client insights. At that time, or soon after I arrived, our technology teams were looking at how AI [can] help us accelerate our software engineering work that we have to do.
…In those early days, we were really trying to figure out where does this fit? Let’s do some experiments, let’s do some pilots. A lot of those experiments aren’t things that we ever scaled across the company, but they created really good opportunities for learning and sharpening our understanding of what are the right use cases where we can apply AI.
…We have prioritized our use cases that we are executing against. We have implemented infrastructure that supports and enables AI agents. We started to scale some of the solutions that we’ve put in place…
I’ll just close out with how we have evolved since then. We have established business AI leaders. So from our investment and distribution teams, we have prioritized our use cases that we are executing against. We have implemented infrastructure that supports and enables AI agents. We started to scale some of the solutions that we’ve put in place, and some [that] we’ve actually scaled across the enterprise. We have a governance structure in place that we have been able to scale as we evolve, and we have launched internal AI training and are committed to increasing our AI literacy as we move forward.
Of all that you have listed, what areas where the hardest areas for you to get through?
I think it’s a few. There’s a lot of curiosity around generative AI, so there definitely are a group of people who are really leaning in and doing a lot of self-learning…. [The challenge is] really about [providing] the right type of training and engagement opportunities for all of our employees, so that we are all kind of learning and keeping pace with the change. That was an area [that] took and still takes deliberate focus around how are we investing within.
The second one was identifying the right use cases where we should focus, and as we got clarity on that focus, it was then building an efficient execution team that not only was addressing how we implement the solution for this use case, but making sure that our data was ready and can support the work we are doing.
Give me an example of the clarity of a use case and then making it a priority. How do you get to that point in the AI journey?
I’ll give you an example of one around productivity that impacts all of our associates. We had done initial experimentation with Microsoft Copilot… Initially it took some time for our associates to appreciate what is the value in leveraging this kind of tool. Part of the work that we had to do was [rolling] out the right training — let’s highlight examples, let’s have the demonstrations, let’s share the videos.
We have an innovation hub in our headquarters where we would have 15- to 20-minute learning sessions. “Here is the art of the possible with this capability. The capability is evolving and here is how you can tie it to everything work.” So whether it is summarizing meeting minutes, whether it is helping you manage your emails and be more efficient, whether it’s helping you craft documents that you need to draft — it required that change management work…for more people to start to benefit from this productivity tool.
Can you give me two or three example of high impact generative AI use cases that demonstrate the productivity and efficiency that you speak of?
There’s a number of use cases that have had positive impacts… With generative AI agents, it’s one of those things [where] as you do more you learn more.
One area where we have seen a lot of benefits in our investment space is enhancing our research capabilities. This is deeper insights into our proprietary data. We have methods today in terms of how we get insights from our data — analytics reports, tools that we use. By leveraging generative AI, we have been able to get deeper insights, [and] we have been able to marry our proprietary data with some of the external data that we have, and unlock new insights that are very helpful from a research perspective. …We are able to operate faster, so there is a productivity gain. We have been able to identify emerging themes that help us with research and provide good input to our investors and [do] things like generating commentaries about the work that we are doing.
We have been able to leverage [Amazon Web Service’s] generative AI tool to help accelerate some of our software engineering work, and we have been able to measure some of the gains that we have there.
…In our distribution space, we have been able to apply these AI tools to get deeper insights in our sales data — data we have about clients, data that helps inform how we can better serve our clients, or are there insights that may help with some of the capabilities that we may want to roll out for clients? And then the reduction of manual work in that space from a code generation perspective. We have been using Amazon Q. …We are in a multi-cloud space. One of the cloud providers is AWS. We have been able to leverage their generative AI tool to help accelerate some of our software engineering work, and we have been able to measure some of the gains that we have there.
In terms of another example… this year we were able to leverage our AI tools to help draft some of our performance reviews, and then edit and ensure that the review has come together in the right way, but [that] accelerated how you pull together a lot of different inputs from different data sources to come back with an outcome.
So overall, when I think about the impacts of these different use cases, we’ve been able to reduce manual work, create some efficiency, gain new insights and we are seeing more and more people incorporate tools into their way of working. This ties back to some of the metrics that we capture around how is this taking shape, and whether these are the right tools for now. We also know we have to be disciplined about turning tools off if they are not working, or don’t have the intended impact.
Do you have data that you can share that shows the ROI in these examples?
We do. Maybe I will highlight just a few key performance indicators that we look at. And I will tell you that this is a space that is still maturing, in terms of how do you capture the value that is being created?
Are the tools being used? Are people sustaining that usage? Is the usage increasing?
…We typically focus on two kinds of core dimensions: one is adoption, and the other is productivity. With a lot of our generative AI use cases, we will look at monthly activities, [or] how many people are using the tools? For some of our use cases, we have satisfaction surveys that we run quarterly just to capture some of that qualitative feedback. We have some engagement metrics that track…[people may be] using the tool, but what is [their] usage level? So that’s one way we kind of look at it from the productivity gains perspective: Are the tools being used? Are people sustaining that usage? Is the usage increasing? And then you marry that up with some of the survey feedback.
For some of our use cases, we capture hours saved. There are cases where that translates into call savings that are easy for us to identify. In use cases to automate a process, we will look at exception handling — how many exceptions require manual intervention. The thought there is that we would want that to decrease that over time. From a research and insights perspective, we are tracking the number of new insights that we are attributing to our AI capabilities.
But do you have any hard data to share?
We have data, but not that we are ready to publish. We are still maturing our capabilities, and learning a lot…
How do you prioritize projects?
There are several factors. There are projects that we are executing on that are delivering against business priorities. We work with our business partners to make sure that we are sequencing and prioritizing the most important work that we have to deliver on. We also have a set of projects that are focused on how do we…put ourselves in the position where technology data, our cyber security teams, our operations teams are operating as efficiently as possible. …Sometimes we prioritize projects based on impact, [and] sometimes it’s based on [needing] this capability to be in place, because we know what’s coming from a business standpoint and we see this as a prerequisite. …Then, there’s other things that are really about how [we can] reduce our legacy infrastructure and put ourselves in a position where we are as nimble and efficient as possible, so we can support the needs of the business.
Would you say that you are proactive or reactive on implementing guardrails for your AI implementations?
Proactive for sure — but always focused on learning and then determining [whether] there [are] adjustments we need to make based on what we’ve learned. The first priority is we have to make sure we are protecting the company…
You have an AI lab. Is that the Technology Development Center in New York City that was established in 2017?
That is right. It is now the T. Rowe Labs. That’s what we call it.
How are you using that lab to [advance] AI innovation?
It’s a team of engineers and data scientists, and these individuals are focused on delivering innovative solutions for our investments and distribution business areas. And this is the team that has been operating AI since 2017, and a lot of focus early on was around machine learning. But they partner closely with our business partners on solutions. When we started, and we were doing more pilots, this team was doing a lot of the delivery work for the pilots. Now, as we continue to expand the work we are doing in AI, they are still very much in the middle of some of our solutions…
It is a really important team for us, and their focus now is on our generative AI solutions.
How do new vendors with AI tools get your attention — or are you simply sticking with well-known brands?
We can’t, because technology in this space is evolving so quickly that we are paying attention to the landscape… how it evolves. We are not intending to chase everything, because that does not make sense.
…Microsoft is an important partner. Microsoft Copilot, Amazon Q, Open AI — we’ve leveraged [all of] their products. For our agentic agent infrastructure we are leveraging Amazon’s products called Bedrock and Agent Core. Then, we have third-party platforms that are core to our technology foundation. Salesforce and ServiceNow are a couple of examples. They have rolled out AI capabilities.
…We are continuing to evolve and figure out what is the right combination of tools that are available that helps us deliver on our business priorities.
…We are always assessing the landscape. We talk to startups and fintechs all the time. I won’t name any today — I’m not here to pick winners — but…if there is an AI company that is interesting and aligned with problems that we are trying to solve then we will take a closer look. …We are continuing to evolve and figure out what is the right combination of tools that are available that helps us deliver on our business priorities.
How has agentic AI changed the depth and breadth of your client engagement?
…It will allow us to look at some of our processes that exist today and significantly change them, introducing a whole new form of automation. So we have identified areas where we think agents will be really helpful. For example, in our distribution and investment space, we’ve done some early experimentation to start to build our confidence, and my reference to Agent Core earlier was that that’s an important prerequisite — the underlying infrastructure that allows us to do more with agents.
What have you learned during the last two-plus years, with regard to generative AI?
…It is an important time to be vigilant about learning and [to challenge] how these technologies fit [into] the way that we operate. I can see a future…where some of the things we are talking about now — agents and AI and productivity tools — become table stakes. And so it is important that we are positioning and enabling the company to take full advantage… that we are increasing our AI literacy so we are well positioned for the future.
You must be logged in to post a comment.