A closing table with stacks of documents and a laptop screen showing an automated escrow workflow dashboard in a title company office
Project Management & Operations

Your Closing Had 120 Tasks. An AI Handled Most of Them Before You Sat Down.

By Frank DeLuca · June 20, 2026

← Back to all articles

In April 2026, a company called HomeLight announced EVA, an AI-powered escrow agent that can open orders, pull title documents, coordinate with lenders and HOAs, and initiate fund transfers. EVA works with more than 80 external tools. BlackRock put $40 million behind scaling it nationwide. HomeLight's CEO called it "just the beginning." Nobody at the closing table asked the buyer whether they were comfortable with that.

Completing an escrow requires roughly 120 discrete tasks. I have watched that number climb for twenty years as compliance requirements, documentation standards, and lender stipulations layered on top of each other like geological strata, each new regulation adding another form to print, another verification to perform, another party to notify, until a process that used to take a competent title officer about eight hours of actual work spread across three weeks now averages 22 hours per file according to the American Land Title Association. More work per closing, fewer people willing to do it, and a volume machine that processes seven million mortgages annually against more than two trillion dollars in originations.

Something had to give. What gave was human involvement.

Three Companies, One Bet

HomeLight is not alone. CertifID acquired CloseSimple on June 18, 2026, combining wire fraud prevention with closing communication and automation. CertifID has blocked $283 million in fraudulent wire transfers and recovered $132 million for victims, backing every protected wire with up to $5 million in direct coverage. Hundreds of title companies already use CloseSimple to keep agents, lenders, and buyers aligned through the close. The combined platform will push AI deeper into every step of the workflow.

Qualia launched Clear, one of the first agentic AI systems built specifically for title and escrow. It conducts preliminary title examinations, verifies closing data for accuracy, and monitors transactions for fraud. Not as a chatbot that answers questions, but as an agent that interprets context, makes decisions, and executes routine tasks with limited human oversight. Nate Baker, Qualia's CEO, called 2026 "the most consequential transformation the industry has ever experienced." He might be right, though I have heard that sentence applied to every technology since fax machines replaced courier services, and I am old enough to remember when "paperless closing" was supposed to arrive by 2015.

22 hrs
Average time to complete a single title file, according to the ALTA 2024 report. Up from roughly 8 hours two decades ago.

What Buyers Actually Think

While these companies were racing to automate the closing table, Cotality surveyed homebuyers and found something that should alarm anyone shipping these products. Trust in AI to help find a home dropped from 30% to 16% in a single year. Not a gradual slide. A 14-point collapse.

Seventy-five percent of buyers assume AI is already embedded in the process, and they are probably right. But assumption is not acceptance, and the numbers that follow reveal the gap between what buyers expect to encounter and what they are willing to trust with the largest financial commitment of their lives. Fifty-five percent prefer a human for their mortgage, up from 46% the year before. Sixty-six percent want a human for legal assistance, up from 54%. Nearly half said it is unacceptable for lenders or insurers to conduct automated AI valuations without their prior approval, and 44% would pay an additional fee to have a human expert verify AI-generated decisions about their home purchase.

Read that last number again: 44% would pay more money for someone to check the machine's work. In an industry where every basis point of cost gets fought over at the negotiating table, buyers are volunteering to pay extra for human oversight of a process that companies are spending hundreds of millions to make less human.

The Fraud Paradox

The FBI's IC3 reported $275 million in real estate fraud losses in 2025, drawn from 12,368 complaints. Their Recovery Asset Team initiated 3,900 freezes and managed to hold $679 million of $1.16 billion in attempted theft, which sounds impressive at 58% until you do the subtraction and realize $481 million in stolen closing proceeds made it out the door and into accounts controlled by people who never owned the house, never held the title, and never sat at the table where the documents were signed. In one case from Missouri, a buyer received wire instructions from what appeared to be their title company's email, sent $1.3 million to the address provided, and learned afterward that the entire email was fabricated.

AI is on both sides of this fight. CertifID uses machine learning to detect fraudulent wire instructions before money moves. Qualia Clear monitors transactions for anomalies. But the FBI's own 2025 report specifically called out AI as making fraud "more convincing and harder to detect," noting that synthetic content, personalized phishing, and deepfaked communications now operate at a scale and sophistication that did not exist two years ago.

An agentic AI escrow officer that can "complete critical steps like document ordering and fund transfers with minimal manual input," to use HomeLight's exact language, is also a system that reduces the number of human checkpoints between a fraudulent instruction and an irreversible wire transfer. Every automation that removes a human from the loop removes a pair of eyes that might have noticed the email address was off by one character, that the wire instructions changed at 4:47 PM on a Friday, that something felt wrong in a way that twenty years of closing experience can detect and no algorithm has been trained to quantify because the training data for "a feeling in your gut" does not exist in any dataset a neural network can ingest.

$275M
Real estate fraud losses reported to the FBI's IC3 in 2025. AI made schemes "more convincing and harder to detect."

The Automation Gap

Here is the calculation nobody in the industry wants to publish. HomeLight says EVA automates "the majority" of a typical file's 120 tasks, which I will conservatively peg at 67%, around 80 tasks handled without meaningful human involvement. Cotality says 16% of buyers trust AI to help with their home purchase, a number that was 30% just twelve months earlier and appears to be falling faster than anyone in product marketing anticipated. That is a 51-point gap between what the technology does and what buyers are comfortable with it doing, a chasm wide enough to swallow every press release about "revolutionary transformation" that the industry has produced this quarter.

Fifty-one points.

In project management, we have a name for what happens when a system's capability outpaces stakeholder buy-in by that margin. We call it a failure waiting to happen, not because the technology does not work but because the first time a closing goes wrong, the first time a wire ends up in the wrong account or a title defect slips through the AI's preliminary examination, the buyer's reaction will be proportional to the trust deficit rather than the error itself, and that deficit right now is the size of a canyon that the industry is pretending does not exist by publishing press releases about "revolutionary transformation" while their own customers tell surveyors they would pay extra to have a human double-check the machine.

What This Means If You Are Buying or Selling

Your title company is probably already using AI, or will be within the year. Only 17% of firms have formal AI policies in place, according to a Qualia industry survey, up from 6.6% in 2023 but still a minority. That means 83% of title companies adopting AI tools have no written framework for how those tools are governed, audited, or overridden when they produce a wrong answer.

Ask your escrow officer. Specifically: is any part of this closing being processed by an automated system? Which parts? What happens when the system flags something incorrectly, or misses something it should have caught? Who reviews the AI's output before it becomes a decision? These are not paranoid questions. They are the same due diligence questions you would ask a general contractor about their subcontractors, because the person responsible for your closing is now subcontracting portions of it to a machine, and you have a right to know which portions.

If you are a title officer or escrow professional, the industry's 17% AI policy rate is an embarrassment. You would never close a transaction without a title commitment. You should not deploy agentic AI without a written policy that specifies what the AI can and cannot do, what triggers human review, and how errors are documented and corrected. The CFPB and FTC are already scrutinizing AI chatbots for consumer harm. An AI that executes fund transfers with "minimal manual input" is going to attract attention that no title company wants, the moment it executes a transfer it should not have.

The Strongest Case Against This Article

Title work is drowning. Twenty-two hours per file, with volume that overwhelms available labor. The Bureau of Labor Statistics does not even track "escrow officer" as a standalone occupation, which tells you something about how the profession is valued. If AI can cut that 22 hours to 8 while maintaining or improving accuracy, the people who do this work every day will tell you it is the best thing that has happened to their profession in decades. The trust gap may not matter if the alternative is a system that cannot process closings fast enough to keep the housing market functional. Buyers who say they want a human may change their minds when the human-only option takes 90 days instead of 30, costs $800 more in closing fees, and still results in the occasional wire fraud because a tired human at 5 PM on a Friday made the same mistake the AI would have caught.

That argument has real weight. The problem is that nobody has measured whether the AI actually maintains or improves accuracy, because none of these tools have published independent error rates. We are running a live experiment on the largest financial transaction most people will ever make, and the hypothesis is "trust us."

Limitations of This Analysis

HomeLight's EVA launched in April 2026. No independent audit of its error rates, task completion accuracy, or security posture has been published. CertifID's $283 million fraud-blocked figure is cumulative over the company's lifetime, not an annual number. Cotality's survey was conducted between January 29 and February 9, 2026, before EVA was announced, meaning the trust numbers may have shifted in either direction since then. The FBI IC3 data is complaint-based and almost certainly undercounts actual real estate fraud. ALTA's 22-hour average is drawn from a 2024 report and reflects self-reported data from title professionals, which carries the usual caveats about self-reported data from any profession. The 67% automation estimate for EVA is my inference from "the majority" of 120 tasks, and HomeLight has not published a specific percentage. No published data exists on AI-specific errors in closing automation because the industry is too new to have generated the failure data that would make such analysis possible.

Frank DeLuca has managed residential construction projects for twenty years. He has closed on six homes personally and found something wrong with every single one of them, usually after signing. He is not optimistic, but he is paying attention.

← Back to all articles