Iro Tasitsiomi, head of AI and investments data science at T. Rowe Price, discusses how artificial intelligence, particularly generative AI, is transforming investment and trading workflows. Once focused mainly on research and alpha generation, her team now drives AI adoption, governance, and training across the firm.
She shares her outlook on how AI is reshaping trader skill sets, emphasizing critical thinking, data understanding, and the ability to challenge AI outputs. While data quality, system integration, and regulation remain key challenges, Tasitsiomi envisions a more automated, data-driven trading desk where traders become “power users” of AI, combining human insight with machine intelligence.
Interview
Josephine Gallagher – Tell me a bit about your background and your role at T Rowe Price.
Iro Tasitsiomi – So currently, I’m the head of AI and investments data science at T Rowe. Before that, I was a director at BlackRock’s AI labs. I started my career in finance as a quant at Goldman Sachs, where I actually learned finance for the first time. Because before that, I was an academic. I was actually teaching and researching at Princeton, after receiving a Master’s in physics and a PhD in astrophysics from the University of Chicago. Now, when it comes to what I’m doing really at T. Rowe. Up until a year ago. A lot of the emphasis was around research and development, alpha generation, helping investors, traders, etc, get an edge, right, from data and technology. As of last year, and after assuming responsibilities for AI adoption across the enterprise, I find myself working on many other additional things, including, of course, governance and training of our employees.
Josephine Gallagher – Where do you see AI, particularly generative AI, making the most meaningful impact across investment and trading workflows?
Iro Tasitsiomi – Frankly, with artificial intelligence, anything from automation to augmentation, to differentiating intelligence is possible, and everything, of course, in between in that spectrum. Basically anything we do on investing and on trading can benefit from some kind of AI. It is worth mentioning before touching on Gen AI in particular, that trading, in particularly liquid markets, have been benefiting from machine learning and AI for many years, if not decades. Every stage of the trade, from idea origination to placement to confirmation, settlement and post trade analysis, there is so much machine learning we have been doing, of course, smart order routing, etc. At the same time, on the liquid markets front, you can actually use AI and data in particular, for price discovery. Now with generative AI, we take it a step further, because generative AI introduces our ability to consume so called unstructured data, meaning textual data. So it’s no longer the case that a trader has to wake up every morning. They need to spend hours trying to consume all the market colour information they receive in emails. Instead, generative AI can help us do that in a matter of minutes.
Josephine Gallagher – Some believe that AI can dilute the value of decision-making on the desk or even the development of trader skill sets. What is your view on this, and how can traders adapt to this quickly changing tech environment?
Iro Tasitsiomi – I would say that it’s not really that much about dilution. Instead, I would say that it is around reshaping and redefining what real differentiating skills are on the desk. The ability to consume multiple emails fast is not necessarily a trader differentiating skill. Instead being able to understand inputs and data, being able to use effectively AI and models, as well as being able to stress test AI, since there’s no guarantee that it will always be correct, will be the new skill set required by traders. So when I’m thinking about the future, I see a hybrid kind of model where actual traders become really power users of AI. And in this hybrid model, we get to see the strongest elements from both humans and machines.
Josephine Gallagher – And so what are the biggest data and operational challenges to using AI in trading today?
Iro Tasitsiomi – When it comes to data, the problems can be diverse, but you can probably describe it all by talking about data incompleteness. So this incompleteness can be biases. For example, you shouldn’t use liquid market information to infer anything about illiquid markets. So in that sense, data can be incomplete. Also, the same is the case with data availability. Often we may not have enough in terms of the time span, or data can be just captured wrongly. So lots of challenges on the data front. When it comes to operational challenges. I think one of the most important is integration with legacy systems. So you find yourself having built something cutting edge, something AI, but then it has to be integrated in something that cannot keep up in terms of performance. Another challenge is around governance, and in particular to the fact that since the Gen AI explosion, regulation has been constantly changing. So even keeping up with what is the right regulation to follow can be a challenge. Last but not least, explainability. We are in a highly regulated industry, and that means using black box models can be tricky. We all agree, I think that going to our boss and saying, AI made me do it when something goes wrong will not be a satisfactory answer.
Iro Tasitsiomi – Explain what agentic AI is and how is it different to traditional workflow automation.
Iro Tasitsiomi – Traditionally, workflow automation pertains to processes that consist of a specific number of rules-based steps, so if X then Y. Usually the challenge around those is that there may be many if and then conditions, but that’s it, and it is deterministic. It doesn’t change. When it comes to agents, we have something much more dynamic. Agents adjust and adapt to their environment, they actually also try to optimize a certain objective. It can be maximizing wealth, or it can be optimal trade execution. So in the case of trading, you can imagine that you have an agent that monitors liquidity, adapts tactics, even escalates when needed.
Josephine Gallagher – Looking ahead, what do you think the future trading desk will look like, and given the pace of AI adoption, what will be the big differentiator in the use of AI?
Iro Tasitsiomi – Well I see a future trading desk that is much more data-driven, automated, and efficient. When it comes to differentiation. I often ask myself the same question, especially given that more or less everybody will have access to the same tools, as well as the same data, then indeed, what will make one trader better than the other? When I think about this question, first, I remember that AI is good at replacing junior-level execution, not really leader-level thinking. So the differentiation will be critical-thinking traders. You need traders that are leaders in their field and in their way of thinking. As we said earlier, you also need traders that will be power users who will be able to embrace and leverage the technology maximally. A trader who knows what question to ask. Because AI is better at answering questions rather than asking the right questions or generating ideas on its own. There was a time that I used to say that 50% of the value of an exercise comes from asking the right question. In the future with AI, it will be 99%.
Catch Trader TV shows each week:
To watch Trader TV This Week shows go to – Video
Or make sure to follow us on LinkedIn – Trader TV’s LinkedIn
Available as a podcast – Link
