Alkami Chief Technology Officer Deep Varma is pushing the digital banking provider’s 300 regional bank and credit union clients past the era of replicating branch interactions online and into one defined AI-driven personalization. 

Varma is a 35-year Silicon Valley veteran with stops at Yahoo, Zillow and Varo Bank, where he helped launch the first fintech to receive a charter from the Office of the Comptroller of the Currency. Varma has spent nearly three years driving Alkami’s technology roadmap.  

“Only reason I’m here is because we are going through a massive transformation,” he told FinAi News. “When you compare us … to the European Union and some of the Asian countries, we are still lagging the innovation and adoption.”

That transformation, in Varma’s view, is shifting banking from a transactional relationship of log in, check balance, log off, to an emotional, proactive one in which institutions provide relevant guidance before customers ask.  

Varma spoke with FinAi News about the company’s AI-driven tool launches last month, agentic AI and where regional banks can still beat the megabanks. 

What follows is an edited version of the conversation.

FinAi News: You just launched Alkami Code Studio. Who is it for, and what gap does it fill? 

Deep Varma: Alkami is a digital banking platform used by 300 financial institutions, mostly regional community banks and credit unions. They use us to serve their customers. Many of them embrace and extend our platform using our SDKs and APIs. But here’s the gap: When I’m building the platform, I’m not building it for one bank — I’m building for 300. If Bank of America comes to me and says, “I want to change the header,” but 299 others aren’t asking for it, I’m not going to do it. So what choices do you have? You wait for my roadmap, you keep complaining or you use the SDK and make the change yourself. 

We have 90 financial institutions out of 300 already using our SDKs, which shows there’s a real desire. Code Studio is a layer on top of the SDK. Based on natural language prompts, it generates the code for you. So instead of hiring 10 engineers, you hire five, and Code Studio is your engineer. The remaining 210 institutions that haven’t been using SDK can now tap into customization, too. Under the hood, we’ve trained our own LLM models — Star Coder Tomb on AWS Bedrock — on our entire code base, so it knows what good code, bad code and ugly code look like in our system. The conversational layer is Claude, talking to our MCP [model context protocol] servers, which talk to our LLMs to generate the code. 

FinAi News: Where is AI making the most measurable impact in banking right now, and where will it take longer? 

DV: Front office, mid office, back office — those are the three layers. The mid and back offices are where AI is making an impact today. I’ll give you the simplest example: reconciliation. Every bank runs day-end reconciliation — debits, credits, money movement — and that’s a six-hour process, every day.  

Month-end takes 10 hours. Year-end can take 24. AI agents can crunch those numbers and get you to the right point much faster. Same with mail handling. At Varo, we used to receive thousands of letters — someone would open the envelope, read the form, upload it somewhere. A human still has to open the envelope, but everything after that — OCR, document scanning, classification — AI handles. RPA [robotic process automation] was fantastic at automating repetitive tasks, but RPA only does what you tell it to do. Agentic AI can analyze, learn, discover and change. 

The front office takes longer, and it’s not because of regulation alone. Changing consumer behavior is the hardest thing. When we launched Zestimates at Zillow, no one had any idea what a Zestimate was. It took us seven years for people to get used to checking their home value every month. Plus, there’s the explainability problem — you can’t tell a customer their mortgage was declined because an agentic AI said so. We’re not ready — and consumers aren’t ready — to delegate their financial needs to a bot yet. 

FinAi News: Every bank is talking about agentic AI. Where are we on the curve from pilot to production? 

DV: What you’re seeing is the emergence of vertical agents first — banks building one flow to do step A, B, C, D within a contained scope they can manage. The horizontal layer, where one agent hands off to another, is going to come later, because that requires more governance and human intervention. I was talking to a CIO recently who told me their agent runs the ACH file, does all the sanity checks, and says, “My ACH file is ready to submit to the Federal Reserve.” But a human still uploads it. I asked why. He said, “We’re not ready yet — we can’t make mistakes.” That’s where we are. The day will come when one agent validates the file and another submits it, but we’re not there. 

There’s also a new dynamic I’d call token currency. Anthropic, OpenAI and others are putting limits on usage and sessions — $20 plans, $100 plans. I have 250 engineers. One developer uses 10 prompts to get from A to B, but the platforms don’t easily let that engineer share those prompts with the rest of the team, so everyone has to burn tokens running through the same process. That’s token currency, and we’re managing it carefully. We’ve given our developers access to Claude, OpenAI and Copilot, but we’re not at unlimited tokens yet. We’ll get there. 

FinAi News: With megabanks like JPMorgan throwing massive sums at AI, can regional banks realistically catch up? 

DV: The good news is, with this technology, over time it gets cheaper and cheaper. Moore’s Law applies — accessibility increases for everyone. Today Chase is throwing money at it because they have deep pockets, but the technology is going to become so pervasive that regional banks can use it too. The gap is going to reduce. It always happens this way: new technology arrives, the gap is biggest at the beginning, then it shrinks until everyone can use it. 

But the more important question is why we still have 5,000 banks and credit unions in the U.S. when Chase and Bank of America have such deep pockets. There’s a reason. In a lot of states, relationship-based banking still exists — common people in those areas want to walk into the bank and have that relationship. That’s not going anywhere. Readiness for regional banks isn’t a technology problem anymore — the technology is available. It’s a will problem. Sometimes you have the technology but the people are resistant to change. Sometimes the people are ready but the technology doesn’t exist. In our case, the technology exists. The mega banks have deep pockets, the technology and the will. If a regional bank has the technology but no will, they’ll never compete. That’s how I see it. 

FinAi News: Where will AI’s biggest disruption to banking show up over the next 12 to 18 months? 

DV: Back office is No. 1 — that’s where AI is going to make the most impact, in a positive way. Reconciliation, mismatches, the places where banks lose money on operational drag — AI is going to make all of that better and smarter. Then it sprinkles into the mid office. We’ll start seeing more of the agents people are building, and I’m confident some front-office AI, chatbots and assistants will roll out, mostly in the customer service space. There’s no reason to call an 800 number to ask, “When is my card coming?” AI will answer most of those questions, and you’ll start to see it play a proactive role. 

The bigger shift is from reactive to proactive banking. I went into a branch once and intentionally tested it — I had $20,000 in savings and $2,000 in checking, and I asked to withdraw $100. The teller could see I had money. I was hoping they’d say, “Hey Deep, you can move this $20,000 to a CD.” They said nothing. I had to ask. That’s reactive. Where the swing is going to happen is using the data — the data is your data, but are we helping you understand what you can do with it? When you log in, the system should say, “Hey Deep, we can open this savings account and you’ll get 5% APY,” That’s proactive. That’s where the next phase of banking is heading, and AI is what makes it possible at scale. 

Register here for the FinAi Lending Summit, set for Oct. 7-8 in Las Vegas.