Ask most lenders where their loan technology slows them down and you’ll rarely hear about the screen a loan officer looks at. You’ll hear about the invisible middle, the place where a file waits for a person to read each document, key in each number, and check it against the right investor’s rules before anything moves. That invisible middle is where the cost to originate has quietly grown past eleven thousand dollars a loan, according to the Mortgage Bankers Association, even as the software sitting on top of it kept getting more polished.
Plenty of newer mortgage technology entrants are racing to point AI at that problem, and the AI is often the easy part. The hard part is knowing where loans fall apart, which conditions a given investor truly cares about, and why a file that looks clean, still isn’t. That judgment comes from a platform built on thirty years of origination experience, and it’s the part you can’t spin up with funding. Our product evangelist Craig Rebmann puts it as “domain expertise plus AI = magic”, and he’s right that one without the other doesn’t get you much. We’ve spent three decades building the expertise. Now we’ve added the AI layer within it – and that’s allowed us to build out AI agents going into the point of sale, that lives inside Empower.
Why build an AI Agent into our POS?
The point of sale is where the borrower first meets the loan, and it sets the tone for everything that follows. A borrower uploads documents, answers questions, and waits, while behind the scenes a person works through whether what came in is complete, accurate, and ready to move. When that handoff drags, the borrower feels it as silence and the lender feels it as cost. Putting intelligent agents at the front door means the routine parts of that intake start handling themselves the moment information arrives, so a file shows up to your team already organized rather than as a pile to sort.
That only works if the intelligence understands mortgages, not just documents in the abstract, which loops back to the point Craig made. An agent that knows what a complete file looks like for a specific investor is useful. One that simply reads paper is not.
Process orchestration is the part nobody sees
When people picture loan software, they picture the interface. The part that determines how fast and how cleanly a loan moves sits underneath that interface, and we call it an orchestration engine. It’s the engine deciding what happens to a loan at each step, in what order, against which investor’s guidelines, and what gets handled on its own versus handed to a person.
When that engine is built right, a loan moves itself through the predictable steps without anyone pushing it along, which means most of the value in modern loan technology lives in a layer the end user never directly sees. When it’s built poorly, you’ve added a nicer face to the same manual grind.
Intelligence belongs inside the platform, not bolted onto it
A good deal of the industry has approached AI as something you add on top of an existing system, a feature in one place and an assistant in another, which demos well but changes little because the work underneath still runs the old way.
We take the opposite approach and build AI and automation into Empower itself. Our intelligence layer, Aiva, runs inside Empower as part of the flow, reading what borrowers submit and checking it against the actual investor guidelines while the loan moves. Because the intelligence lives in Empower, the difference shows up in the loan and not only in the interface.
The first piece our customers can use today is Ask Aiva, an assistant built directly into Empower that answers questions about a loan from the file itself, traceable back to the source rather than offered as a guess. Let’s clarify that for a second – when you ask it a question, it gives you the answer AND the source. That means you can click the source to read more about the root of the answer. It’s assistive intelligence, and its job is to help a person work faster.
The step we’re most focused on now goes further than answering questions. We’re soon shipping out AI agents that take action on the repeatable work at the point of sale, organizing and verifying what a borrower submits and surfacing only the items that need real judgment, so the loan officer who used to chase missing pieces can step into a command center, free to ask and direct and receive the answers back instead of digging for them.
What is AWS and where does it fit?
Building intelligence into the engine securely, and running it across thousands of loans at once, calls for a foundation we don’t build alone, so we build ours on Amazon Web Services. Their role is to make the underlying cloud and the modern intelligence services it runs on reliable and effortless, which frees our energy for the part only we can add: the mortgage expertise on top. Craig and I represented Dark Matter at an AWS event in Orange County, CA this week, walking a technical audience through exactly this, and working alongside a partner operating at that scale is part of what makes the future we’re building credible rather than aspirational.
What this means for lenders
Where a team does the work today by reading and keying and checking and clearing…that same team moves toward reviewing the work while the routine steps run on their own. Essentially, moving from DOING to REVIEWING.
If you’re weighing where mortgage technology is headed, watch the LOS and the intelligence embedded within, rather than the interface, because the companies changing the work underneath are the ones who’ll move your cost to originate and your borrower experience in the right direction. The lenders who win the next few years will be the ones whose technology pairs deep mortgage expertise with intelligence built for it, and that pairing is exactly what we’ve spent thirty years earning the right to deliver.
We’ll have a great deal more to show in the weeks ahead (trust me, its really freaking cool), and if you’d like to see where this is going before it’s public, that’s a conversation worth having now.