For the past five years, financial services company the United Services Automobile Association has taken a hands-on approach to experimenting with and deploying AI technology.
The company, which many know from its public profile as an insurer for military members and their families, uses AI technology across its organization and has deployed more than 200 AI tools and products.
However, before the advent of generative AI, the United Services Automobile Association (USAA) was experimenting with AI technology for a long time and still also uses traditional or machine learning AI technology, not just generative AI.
In this Q&A, Ramnik Bajaj, chief data analytics and AI officer at USAA discusses the company's strategy with GenAI and agentic AI.
How does USAA use AI technology?
Ramnik Bajaj: Our technology strategy is to have multiple platforms and pick the one that's fit for the use case that we are trying to solve. For some use cases, we'll use providers with large models based on the capabilities that we need.
We have deployed technologies to help our workforce with their knowledge work. We use what we call the EagleGPT, which our workforce uses as their assistant. We have AI technology deployed to help our software developers with their code-writing tasks. We call it code assist, and that's used by all our software engineers. We have deployed some technologies in our back-office and front-office functions and are continuing to work on more automation of both.
The front office is our call center because it picks up the calls. The back office is all our processing, claims, and servicing policies and processes. Across the board, we are using AI for the front office. For example, we have deployed an AI that can help our front-office staff respond to our members' (we call our customers members) questions faster by putting an AI system on top of 20,000 documents containing the knowledge they need to answer the questions. So now, instead of searching for a document and reading it and then answering the question, they can use the AI to ask the question, get the response from the AI and serve the member that much faster.
Ramnik Bajaj, chief data analytics and AI officer at USAA.
For our back-office work, particularly in the data management space, we have used IBM Orchestrate. We are deploying Watsonx Orchestrate to automate some of our data management functions. For example, if a particular team needs a data set, they request that the data be provided. We would have a Watsonx Orchestrate agent take that request and figure out what to do with it, either to approve it directly, if it's a low-risk data set, or to fire off the proper approvals to the right people.
There's some amount of reasoning and orchestration of the workflow.
We expect to do more use cases, particularly ones that are internally focused and have to be run across our on-prem platforms with Orchestrate. We can deploy this tool internally, in our data centers and in our environment, and use its ability to execute workflows across our internal platforms.
Other than IBM Watsonx Orchestrate, what other AI vendors does USAA incorporate into its workflow internally or externally?
Bajaj: At this point, we want to use all of the major ones to learn the ins and outs to put the right one to use for the case, because they all have subtle differences. Some work better for specific use cases.
We use Google's Gemini models extensively, and we also work with AWS and Bedrock [generative AI platform]. We keep an eye on open weight models, such as the Meta Llama model, and we train our own models.
So, our insurance business involves a lot of computer vision, processing images and automating workflows that involve images. We also train our own models for those.
What benefit has USAA found with Orchestrate for data management?
Bajaj: Data management workflows are internal. They're used by our own employees for our own back-office functions, not impacting our customers directly, and so that's where we chose to start. That's not to say that's the only thing we'll use it for, but we chose to start there because it's internally facing. It's a good place for us to gain confidence in the agentic platform and framework and learn the technology.
All of our AI solutions ultimately depend on how good our data is and how well managed it is. Over the last three years, we have created our unified data environment. Many companies are still on that journey. We were fortunate to get an early start on those three years back. We have essentially taken all of our enterprise data and made it accessible from one place in one analytic environment, and we've connected the dots. If you think about a customer's interaction, they could interact with our property and casualty business today, interact with our banks tomorrow, or do something in the call center. They could do something else on their mobile app. All of these interactions, through whatever channels they happen, and all of the transactions that get executed because of those interactions, we all have it all in one place. So that gives us the advantage of having an end-to-end view of what we do with our customers.
How does USAA decide which AI technologies to use??
We experiment and decide. In 2023, we ran many pilots with GenAI. We learned from that how to implement solutions and what works best. In 2024, we implemented about a dozen generative GenAI solutions into our environment.
As a conservative company, we want to manage our risk around these things very well.
Ramnik BajajChief data analytics and AI officer, USAA
I would say we are moving quickly, but cautiously. As a conservative company, we want to manage our risk around these things very well, so we want to be assured that we can do what we need to with the GenAI solution before rolling it out.
You are a financial service organization. How do you implement governance in AI technology rollouts?
Bajaj: We have an AI council. I lead that. We have four executives from our executive council who sit on the AI Council, and they guide us through what we call the intake. When a use case comes in, we assess if it's feasible to do that with AI, given today's maturity of AI. Can it handle the use case? Would it produce value for us or our members? Will it improve their experience? Will it give us better innovation? Will it generate either revenue or cost savings? We look for the value, and then we look for any red flags.
On the back end of the process, we have a robust model risk management function that validates whether the solution we created performs as we would expect it to. It's a thorough check of whether the model is performing as we expect it to. The third layer is ongoing monitoring. We look at the right metrics coming out of the instrumentation we put in place to make sure the solution continues to perform as we expect it to.
How important is it for organizations to adopt agentic AI?
Bajaj: Adopting agentic AI will be important to achieving end-to-end automation, which will really enhance our productivity and experience. We absolutely intend to do that this year, focusing on some of the lower-risk use cases to begin with and then moving into more complex use cases.
What challenge have you had as an organization adopting generative AI and agentic AI?
Bajaj: The technology is indeed changing fast. It's evolving every day, and we keep pace with that. We have a dedicated lab environment where we constantly test these technologies and learn our way into them.
The challenge for us and all businesses today is: How do you truly reimagine a business process that is being used to run your entire business while you are still using the whole process? The real challenge is: How do we transform those business processes to fully use AI solutions without disrupting our business?
Editor's note: This Q&A has been edited for clarity and conciseness.
Esther Shittu is an Informa TechTarget news writer and podcast host covering AI software and systems.
