Build it, buy it, or find a partner? That's the decision facing financial institutions when it comes to AI capabilities. Abeer Thomson, Q2 senior director of Partner Engineering and Operations, joins CTO Adam Blue to work through what that decision looks like in practice and what separates the AI partners worth trusting from the ones who just demo well.
Federal Banking Agencies Issue Revised Guidance on Model Risk Management
“Burning Chrome” by William Gibson
Adam Blue
Hey everyone, welcome to today’s Cut to Context. I’m Adam Blue, and I’m here with Abeer Thomson, our senior director of Partner Engineering and Operations here at Q2. We’re going to talk about a topic that I think is top of mind for a lot of folks today in the banking space as they think about AI. Should we be building, buying, or partnering for AI? It’s unclear what the right choice is, and today we’re going to dig in a little bit. The OCC recently made some recommendations about how you should be managing AI at your bank, bringing exciting new complexity to trying to bring the most advanced technology to bear. Abeer really sits in a place in our organization where she’s at the forefront of working with partners and working with customers around how we would bring AI into the space. I’m excited to have her here. Let’s just jump in, Abeer. What do you think? Buy, build, partner? How should we be thinking about it?
Abeer Thomson
It’s a question we’re getting a lot from our customers today around what really is the right choice. I will say all three are really possible and all three are really the right answer. There are different layers where you would apply each of those strategies. What I see in Innovation Studio is really a tiered approach.
Building things that are genuinely institution-specific—things tied to their risk appetite, their account holder relationship, their brand experience, how they open accounts—that really makes sense to build. On the buy side, you’re looking for that horizontal infrastructure: core processing, compliance tooling, things where the standardization of the tooling is the key feature that helps them in a highly regulated industry. And then on the partner side, that’s really where you can be at the forefront of leading-edge AI capabilities, where maybe the innovation pace is too fast to specialize deeply internally. That’s where partnership becomes really critical—especially where you need to go deep and specialized in certain verticals and certain areas of your offering to account holders, where building that contextual knowledge internally just wouldn’t make sense. So the decision is really all three, in different strategic areas of your business.
Adam Blue
I think that’s interesting. Maybe to frame it a little bit: the way I think about it sometimes is you should be building as close as possible to the customer outcome. And you should be buying more or partnering for more when we’re talking about infrastructure and the foundational components. If it’s customer-facing and it’s fun, think about building it. If it’s deep in your back office and sitting underneath something else, it may not make a lot of sense to invest in creating new technology when you can get something off the shelf. I think that’s another framing you can lay on it, and I think it’s a great approach.
So let’s talk about partners a little bit. Innovation Studio has been an extraordinary approach for us to bring partners into the space that are subscale and help them through compliance, regulatory, and operational challenges, and then make it easy for our customers to bring fintech experiences to their customers without monthlong waiting cycles and product roadmaps and engineering. What does it look like, and what are you seeing from our partners, about moving to an AI-native approach to delivering these fintech capabilities that we would normally pick up in Innovation Studio?
Abeer Thomson
It’s a hard question because I really think about what “AI-native” actually means. I think there’s a subset of partners that can truly say they were AI-native—they grew up in that space from the very beginning, very early in that frontier area. And then there are a lot of legacy partners that are creating new products with AI bolted on top of a very legacy system. That’s not necessarily meaning they aren’t AI-native, because the product they’re building could become AI-native. It’s a very interesting paradigm that a lot of companies are going through.
I think the companies that are winning in the AI space are the ones that have been in it from the very early beginnings. I think of some of our AI chat partners—they were very early in providing conversational banking, leading with AI and then transferring to a human. They’re going to be the more traditionally AI-native ones. And then you’ve got these AI products bolted onto legacy systems. What I’ve seen become a problem is where data access, how you’re auditing various areas, or latency issues become a problem when partners aren’t thinking about building those AI capabilities natively within their legacy systems. That’s how I’ve been thinking about it as I have conversations with our partners.
Adam Blue
There’s a big difference between, on the engineering side, using AI to build traditional products faster—through assisted coding, improved tooling, and moving to an AI software development life cycle—and actually bringing AI into the experience itself. And as you pointed out, getting AI into the experience and demoing it is one thing. Having that experience be meaningful, reasonable, and effective is another.
I use a Fitbit because I’m old and it’s good to measure all those things—I like to calculate every day when I will likely expire, and the Fitbit really helps me with that. Google just flipped me from the Fitbit app to the Google Health app. And it’s just Gemini in there. It’s great and it’s better in some ways, but I don’t like waiting nine seconds for something to come back from the app. As someone who works in this space, I can always tell when I’m just about to get the AI engaged because I’ll get that spinner, and then it’s eight, nine, 10 seconds. And then I get something great—like, hey, your walk this morning was fantastic, you will probably live to see another day. But I don’t want to wait nine seconds for it. Some of the concerns around operationalization and performance and the feedback loop just aren’t really dealt with yet. That’s a big thing to think about as you operationalize AI—just plunking it in the middle of a user workflow and having it take 10 or 15 seconds may not be the right answer. A lot of people are starting to run up against that wall, and we’re trying to figure out how to maintain the responsiveness of experiences while bringing the richness of AI at the same time.
Abeer Thomson
You have a great point there. I think this is why some of those companies that have been around a long time and weren’t born in the AI era are taking longer to build AI products within their systems. They have to be really thoughtful about what that experience looks like, because at the end of the day, that experience is what they were selling and what they are selling. It can’t degrade what that experience is. Figuring out how to deliver that experience in a really native way that’s impactful to the customer is really important.
Adam Blue
What are some of the red flags when you’re out in the market evaluating partners and figuring out which ones you might want in your infrastructure? What makes you think a partner may not be there yet when they talk about or demonstrate AI?
Abeer Thomson
It’s interesting because when you think about how you would evaluate a vendor previously, it was pretty simple—the product isn’t working, the product isn’t getting delivered, and you could figure it out in the first few months. With AI features, sometimes you can’t tell for a number of months. A product could be performing really well, and then you have that one instance where it drifts or hallucinates or something goes wrong. So I think asking the right questions to the partner as you go through due diligence on their AI capabilities is going to be really critical.
When a partner is doing a demo, I really think it’s important that they show you the good use cases and the bad use cases—to really show they’ve been thoughtful about how the bad use cases are treated, how they operationalize it, how they take that data back into their learnings to make sure it doesn’t end up in front of the customer in a problematic state. Digging in and asking questions a little differently than we’re used to is going to be really important.
Adam Blue
What are some of the great answers you get back from partners who are really doing this well? What separates a thoughtful approach from one partner to another?
Abeer Thomson
I think it really comes down to how they think about the governance of the data around their AI and how they tackle those questions head-on. The partners that really understand the regulatory environment that our financial institutions operate in—and start with that, start with “We understand the space you live in and this is how we are ensuring you’re protected in that experience with your customers”—are really making you believe they’re in it with you. Those are some of the partners that do really well.
Adam Blue
Let’s talk a little bit about the evolution of what I think of as an ecosystem. What can you see getting unlocked as things continue to get more robust and more interesting, both in the Q2 platform and with partners? What are some of the interesting use cases you think we might be able to develop as the ecosystem continues to evolve and mature?
Abeer Thomson
I’m really excited about the potential for us to unlock our AWS infrastructure to our external developer community—not only for the builders at our customers, but also at our partners—where they can start to utilize infrastructure that is very highly constrained, very highly regulated, and really governed for the space we operate in. That would let them bring AI features truly more natively to account holders, because it’s using a very governed infrastructure.
I get excited about this idea of being able to unlock the latest Bedrock large language models to our developers so they can offer really high-value, AI-native features to account holders. Being able to expose that to our developer community would make it that much more native, because it’s on top of our integration layer with all of our data layer there. They can pull in data they wouldn’t have access to within just their own system. We’re bringing so many different systems together that we have all of this robust data that one individual partner would never be able to have, because they’re not sitting where we’re sitting in the space.
Adam Blue
I think all the time about the way we can provide this kind of fabric within which people—from banks, from partners, our own engineers—don’t have to bring their own batteries. If you want to make a cake, nobody wakes up in the morning thinking, I’m going to go reap some wheat and run it through the mill, and then go find a cow and get some milk. You start with the ingredients. You don’t think of flour and sugar and milk and butter as high-value-added components, but they really are. The difference in baking with high-quality butter and great milk—it’s dramatic. There are people who wake up every morning and just think about cows. They’re doing cow stuff all day, and they produce fantastic products from that.
One of the roles I think we have at Q2 is to imagine a world where people on our team just wake up every morning and think about how they can deliver an environment within which you can innovate and do interesting things without worrying about model management and compliance and regulatory issues and governance and security—because all of that will just kill you. If you have to do all of it every time you do something, it’s such a drag on innovation. Being able to leverage our investment in the AWS environment and bring that into the ecosystem is just tremendous. It accelerates all the innovation we get access to in that environment.
Abeer Thomson
I also think about all the monitoring we have in place with our security team around all of that. You’re really not only protected within the AWS environment that we would be able to give you, but also by the monitoring we have within that environment that you don’t have to set up yourself. You don’t have to spin up a whole security team to monitor a side AWS environment—that all comes baked in. I think it’s also the operationalizing of those environments that comes with the access to it, and that sometimes gets overlooked. Everyone wants to build all these great features, but the operationalizing of it is just as important.
Adam Blue
That’s a great example. I did a class in the building where I live where the pastry chef for the restaurant in the complex—who is on the world competitive pastry team, which it turns out is a thing—is the chocolate specialist. He did a chocolate class and brought all the supplies, and we all made chocolate sculptures and painted them. It turned out fantastic. And then I thought about it later: I didn’t have to choose the chocolate, I didn’t have to temper it, I didn’t have to cut the little pipettes they made for sticking the pieces together, I didn’t make any of the shapes, I didn’t have to source the paint. I got to be about nine years old for the afternoon. I made this great thing and thought, I’m an accomplished chocolate artist. And then I realized I could not have done any of the underlying pieces of that myself. I was entirely relying on those folks. It just underscores how valuable it is to have someone who really has and understands that knowledge.
All right, as we start wrapping up here, just a really easy question for you, Abeer. What should banks do next? Just be perfect so they can take your advice, not make any mistakes. Make it easy.
Abeer Thomson
I really think everyone, regardless of your role, should try to solve a problem—and try to solve it in the digital channel. There are so many workflows that get overlooked in the back office, whether it’s for a customer service representative or for an account holder. Everyone at a bank or credit union has an idea to make the experience better. And regardless of your role, with all of these AI tools made available, anyone can really improve that experience.
I would really encourage anyone to tinker. Come into the Q2 ecosystem, try out Q2 Code, try out the SDK, and see what you can build to solve a problem that gets you excited. The best solutions I’ve seen our customers build are the ones that really excite and energize the people building them. So that’s what I would encourage.
Adam Blue
Fantastic. Thank you for that. Here at the end of Cut to Context, we always have a recommendation of a piece of art or culture for everyone to enjoy as a companion to today’s episode. Today I’ll drop a book called “Burning Chrome” by William Gibson. It’s a collection of short stories, all very different tonally, which is really interesting. They represent the beginning of a science fiction movement called cyberpunk, which now is pretty much the default mode for sci-fi films—that dark, semi-dystopian near future. But as a literary movement, it has this very much “20 minutes into the future” feel, and a lot of this writing is from the early ‘80s. I think it’s still very resonant today.
In the same way that some engineer at Motorola watched Star Trek and then invented the cell phone because he saw people using communicators, a lot of people in my generation and the generation before me read a lot of this speculative fiction and imagined what might be possible. It’s really interesting as a body of literature to think about what the future looked like 20 or 30 years ago, how close that vision actually is today, and where the divergence is. As we think about this progression and the way the technology works, just having that framing could be really valuable. “Burning Chrome” by William Gibson—it’s all short stories, so if your attention span has been diminished by the algorithm like mine has, you can still get through them. Super enjoyable. That’s the recommendation for Cut to Context. Thanks so much for being here today, Abeer. This was really great. And remember, listen to Cut to Context wherever you find all of your highest-quality podcasts.
Abeer Thomson
Thank you.