HomeLight has launched an AI escrow agent to automate dozens of tasks
HomeLight recently launched EVA, an artificial intelligence-powered escrow agent designed to automate most of the tasks required to complete a residential real estate transaction.
And the company announced $40 million in new debt financing from funds managed by BlackRock to scale the platform nationally.
Drew Uher
A typical escrow requires more than 120 separate tasks that are time-consuming, difficult to coordinate and error-prone. HomeLight says EVA automates most of themfrom opening orders and retrieving HOA documents to communicating with lenders and transferring funds.
“When we look at the intersection of AI, title and escrow, it’s truly mind-boggling,” said Drew Uher, founder and CEO of HomeLight. Inman. “We believe that it is essentially a foregone conclusion that this industry will look radically different in five to ten years, as a lot of back-office work needs to be done to close an escrow.”
From frustrated to ‘lights out, excited’
The product has been in development for approximately 19 months, with the first automated workflows going live in early 2025. Uher said accuracy quickly rose from 25 to 30 percent at launch to the 80 to 90 percent range — and then stuck there for about six months.
“By December 2025, I was honestly frustrated with the project,” he said. “I wasn’t sure if this was a completely solvable problem.”
Apparently so. In the first quarter of 2026, Uher said EVA essentially achieved 100 percent accuracy – first with one workflow, then another, then another.
“All four major workflows achieved 100 percent accuracy between February and early May of this year,” Uher said. “We’re extremely excited about this. We’ve now automated about 25 percent of the footprint of the escrow process, and it gives us confidence that we have visibility into automating the majority of it.”
There is a caveat to this formulation: Uher does not define 100 percent as zero uncertainty, but as zero false confidence.
“When I say 100 percent accuracy, it’s not a problem if EVA comes across a really strange case and says, ‘I don’t know how to handle this, I’m going to escalate.’ That can still count as 100 percent,” Uher said. “What I can’t have is EVA saying, ‘I think I know how to do this,’ and doing it wrong and making a mistake. Owned and guarantorYou are dealing with people’s money – often savings – and the stakes are extremely high.’
Within the EVA workflow
EVA stands for ‘Escrow Virtual Assistant’. Uher described it as a virtual employee, a collection of different AI agents. “The secret sauce is that we built EVA with more than 80 different tools that she can use to access the outside world and get real work done,” he said.
Uher was quick to say that EVA is not a “chatbot experience.”
“We’re talking about an AI agent – or a collection of AI agents – that can actually go out and do things,” Uher said.
For example, EVA monitors the escrow officer’s inbox, waits for an open order request to arrive, opens the email and PDF, and checks it for more than 20 key pieces of information needed to open the order. Uher said “she” then goes into the registration system and enters all that information.
“If something is missing, she contacts the agent directly: ‘Hey, I’m missing the buyer’s phone number, do you have it?’ – waits for the response and adds it to the file,” Uher said.
Once the order is open, she emails the agent the order number, initiates the title search process with the title seller, initiates the HOA process if one is associated with the file, sends intake forms to the buyer and seller, repeats with them if they have any questions, and lets the buyer know where to send their earnest money deposit.
On the HOA flow, Uher said certain states have websites with databases of HOA documents. In Texas, there are two or three major sites that contain the most HOAs in the state.
EVA can search those sites, retrieve the relevant documents and even use them HomeLights credit card to purchase them on behalf of the company. She also has access to hundreds of provincial tax and assessor offices and various regulators.
Things still on the EVA product roadmap include advanced title analysis, curative work, and signing planning. But EVA has also automated lender applications, notarization quality control, post-closing preparation, and several other parts of the flow.
The competitive window
HomeLight’s entry into ownership and escrow in 2019 predates the AI tools that now make this automation tractable. Uher acknowledged that the first few years were slow. All relevant information in a typical escrow is stored on paper, and extracting it reliably required more robust AI models that didn’t arrive until 2023 or 2024.
According to him, the competitive window is really open.
“If you look at the title and escrow industry, it’s probably in the top 1 percent of all industries where AI can drive real, meaningful change — and no one is really doing anything about it,” Uher said.
Uher went on to say that the “big established players” are generally not technologists.
“There are no new startups being funded at the moment due to the state of the capital markets,” he said. “And the existing title and escrow players have largely been sold or merged. So there’s really no one innovating in this space right now.”
He noted that some software companies build systems for title and escrow services, but they have thousands of customers, each with their own workflows, so they build for them all.
“We are an AI-first title and escrow agency with one workflow – our workflow – and we build technology to close escrows reliably, 100 percent of the time, on time and with 100 percent accuracy,” said Uher.
HomeLight is currently the only customer of its own technology. EVA does not yet grant licenses to other title agencies.
The bigger story
Victor Lund, managing partner at WAV Group and CEO of RE Technology, sees the launch as legitimately significant, but says the newspaper article may not give enough insight into what is actually at stake.
Victor Lund
“The data extraction of escrow is extremely valuable,” Lund said Inman.
He pointed to the market caps of data companies like Cotality and Black Knight as a framework for what HomeLight could eventually access as it scales.
“If you capture all this information at scale, you can license it to those companies or directly to capital market players,” Lund said. “Brokers and agents don’t even think about the value of that data. They just process transactions.”
Lund also emphasized the security dimension. “When AI processes wire transfer instructions, there is no human with a screen to compromise,” he said, noting that bank fraud is one of the most common and costly forms of real estate transaction fraud.
Whether EVA is actually safer than the status quo is unknown, he warned: “None of us have seen it yet.”
The bigger unknown, says Lund, is the go-to-market strategy.
“How do you sell escrow?” Lund said. “You have to retrain agents and brokers. And first of all, agents are going to want to use their own escrow company, so there’s built-in competition. The go-to-market is the hardest thing that we don’t know about yet.”
HomeLight’s likely path, he said, is through its 2020 acquisition of Disclosures.io, which became a standard listing and listing management platform for transactions in Northern California.
“The game is, go to all those Disclosures.io customers and say, ‘Do you want us to handle the escrow for you?’” Lund said. “They can offer a lower price and better service, and as long as there are no kickbacks or revenue sharing, that’s RESPA compliant. Just a better deal.”
‘That’s all a win’
Uher is direct about the consequences for the labor market, but draws the line.
“Right now, 80 to 90 percent of the work we do in escrow is repetitive back-office work that, frankly, no one wants to do,” he said. “AI can handle all that.”
His formulation for the people who remain in the process: own one of the two numbers. The first is a service role focused on agents: someone who answers to agents and customers when things go wrong, which Uher called “a throat to choke.”
The second is escalation and edge cases, the truly customized corners of a transaction that still require human judgment.
Lund said it more bluntly. A good escrow officer may handle twenty closings at a time. If AI handles the audit work and a human handles the exceptions, that same officer can manage a hundred. “That’s all a win,” he said.
Email Nick Pipitone
