‘The answer to everything is always data’

This interview has been edited for length and clarity.

Sarah Wheeler: How do you decide what technology you need to focus on for clients? What does that process look like?

Joe Trapani: Thinking from a 10,000-foot view: everybody wants to do everything faster, cheaper, better. So everything that we build needs to make us faster, cheaper, better.

My philosophy is very much that the days of the tech components sitting in a dark closet, where no one speaks to them directly, is over. I have more conversations with CEOs than CTOs. We sit with our clients and really dig into the problem space, understanding what their primary problems, challenges and bottlenecks are, and then craft the solution.

A couple of years ago we started what I call the blueprint strategy. Because Moder is across a bunch of different clients within all of our verticals that we service, we hear a lot of the same, common problems and we’re not, today, a platform company. So what we developed were blueprints that are easily dropped in, easily configured to your environment and solve a common problem. We now have a catalog of solutions that are really easy to deploy to clients.

SW: Your team has expertise in insurance, banking and financial services. How does that broad expertise benefit your mortgage clients?

JT: My biggest ‘aha’ coming into mortgage was that this is the one industry where knowing the business is more important than knowing the technology. So, I think us having that background, especially in the consulting and outsourcing space that we that we’re in, is invaluable. And I’ve never been a vendor before — I’ve always been on the client side of it and always done the build. So, I’ve often seen vendors come in and pretend to understand the space but they really don’t, or they only understand on a surface level.

We are so hyper-focused on this vertical that when you come in and start talking to us, whether it’s on our operations or our tech side, you’re talking to experts. You’re talking to people who have been there, done that, seen it. We’re not starting from scratch. We’re not learning what an LOS is the day that we come on. We also have a huge stable of not only operational expertise, but tech expertise. So we can immediately start giving a lift manually, and then gradually phasing technology for our clients as well. It’s something that we see immense value in doing.

SW: What are you doing with gen AI that’s impactful?

JT: I’ve been around technology long enough to know that there’s always this emerging ‘next big thing,’ and today that’s gen AI. Early on with gen AI, my biggest concern was security and privacy and compliance. We’re in a highly compliant industry, so one of the things that we immediately focused on was private LLMs. One of our blueprints is a private LLM that can be deployed into our client’s environment. We do not want to own client data. Anytime anybody owns any of your data, it represents a risk. So we immediately developed a private LLM that can be dropped in a client’s environment.

Now, on day one, your agents and employees can have access to the typical chat GPT functionality that we’re all used to and love, but it’s private. You know that the queries, responses and data that’s being uploaded or downloaded isn’t going over the internet. It’s all happening within people’s environments, and it’s secure.

Then, we’re tailoring those LLMs and making them extensible and really targeted. Our blueprint chatbot, which is Moder AI, we can tailor that to specific use cases by creating custom knowledge bases for it. So you’ve got your general and then let’s say we’ll use the scenario desk, or the conforming guidelines that we can then train that model on very quickly. And so now users are getting targeted responses citing sources.

We’ve also added things like general document repos that we can train that on, so we have clients who are using our chatbot across document repositories so that their brokers or even their employees can ask questions and get responses tailored back based on the documents that are driving their business forward.

This is a very risk-intolerant business, so we’re not going to over trust. We’re starting out with these small, targeted copilots or agents that can help. And now we’re moving into the space of process automation with a human in the loop. Targeting the right processes and introducing AI safely is one of our primary focuses.

SW: How do you decide when there needs to be a human in the loop?

JT: The answer to everything is always data. When we use AI, we get a confidence level back. Over time, the data tells us when we can safely automate fully. There are processes that I think probably always will require a human in the loop — the final underwriting decision, for example. But there’s a huge difference between the traditional process and AI having a decision ready and presenting that information to an agent for their review. It’s still a significant uptick in your productivity when you’re now reviewing what that AI may do.

And if you look the emergence of agentic AI, that’s where that trend is going. This [AI] agent sits there and it’s taking multi-step workflows and automating that, and then asking the human, “hey, this is what I think I’m going to do. What do you think?” And I think that is achievable in the very near future.

SW: What differentiates Moder?

JT: One of the things I genuinely think that we are the best at is our blind obsession. We are dedicated, end to end, to our clients. We want to constantly be pushing out things when we work with our clients on the tech side — it’s every two weeks. We want to be the partners that we wanted. We do internal development the exact same way — the tech side sits down with business and understand the business problem before we go out and build things.