Your phone rang at 7:42 last night. Nobody answered.

That borrower did not leave a voicemail. They called the next lender on their list. You paid to generate that call, and your competitor closed it.

The cost of missed calls for loan officers is the largest untracked leak in the funnel. I have spent 23 years in mortgage technology. Every team I work with tracks lead cost, conversion rate, and pull-through. Almost none of them track missed calls. That is a mistake. The lead already exists and the intent is already there. The only failure is that nobody picked up.

Price it first. Then decide what to do about it.

The cost of missed calls for loan officers, run three ways

You do not need a study to price this. You need three numbers you already have.

(Missed calls per week × 52) × close rate × revenue per funded loan = annual loss

Count nights, weekends, lunch hours, and every call that arrived while your LOs were on another line. Pull last month’s report from your phone system. Look at the calls that rang out or hit voicemail. Then apply a close rate you would defend in a pipeline meeting.

Here is that formula at three volumes. Revenue per funded loan is held at $3,000.

Missed calls per week Missed calls per year At a 3% close rate At a 5% close rate
4 (solo LO) 208 $18,720 $31,200
10 (small branch) 520 $46,800 $78,000
25 (8-LO branch) 1,300 $117,000 $195,000

Ten missed calls a week at a 3% close is $46,800 a year.

Shape’s arithmetic here is deliberately conservative, and it is arithmetic rather than a study. Nobody surveyed anyone. Vendor pages in this category routinely model 15% or more of missed calls converting to funded loans. We use 3% because that is a number an operator will actually recognize. Replace every input with your own and the structure holds.

The worst part is that none of it shows up as a loss. Your CRM never saw the lead. Your reporting never counted the call. The deal funded somewhere else.

Why voicemail fails mortgage borrowers

I want to be direct about the evidence. Search this topic and the same three statistics appear on nearly every page. They claim precise rates of voicemail abandonment and callback failure. Chase them and they lead to compilations of compilations, with no public raw data. One competitor is honest enough to flag them as such. I will not repeat numbers I cannot trace, so take what follows as opinion formed over 23 years.

Most borrowers do not leave a message. A purchase borrower whose agent just asked for a pre-approval letter wants an answer now. A callback window is not an answer.

Rate shoppers work down a list. The lender who answers first frames the entire comparison. Inbound calls decay faster than web leads, because the borrower is already mid-decision when they dial. Shortening that window is a system problem rather than a discipline problem, which is why we wrote up how speed to lead with AI works across text and voice together.

Even when a message gets left, follow-up waits on a human noticing it. Messages sit. Callbacks land hours later.

After-hours makes it worse. Borrowers shop mortgages at night and on weekends. If coverage ends at 6pm, there is a hole in your pipeline every day.

What loan officers actually do about missed calls

An answering service is not the only option. There are five.

Option Relative cost What you get Where it breaks
Forward to mobile None A live voice, sometimes No routing, no record, no overnight coverage
Rotate an after-hours phone LO time and goodwill Real coverage on a schedule Comp fights over whose lead it was
Answering service High per call A human voice and a message No qualification, no calendar, no CRM record
Accept the loss Your missed-call number Nothing Only defensible under ten missed calls a week
AI voice agent Low per call Qualification, booking, CRM record Wrong callers, covered below

Two of these deserve more than a row.

Accept the loss is a real answer. At four missed calls a week you are looking at roughly $18,700. That does not justify a vendor, a compliance review, and a rollout. Fix your call routing and move on.

Rotating an after-hours phone looks free and is not. The first time the on-call LO closes someone else’s referral, you have a compensation fight. Coverage quality swings with whoever holds the phone.

What an AI voice agent adds

An answering service takes a message. An agent works the lead. Inside Shape that is four things.

It answers every call, including nights, weekends, and the Tuesday when every LO is busy. It asks your qualifying questions: purchase or refi, timeline, loan amount, whether an agent is involved. It writes the transcript, the score, and the contact record into the CRM. Then it books time on the right loan officer’s calendar.

The fourth part is what separates it. Most missed-call fixes capture the call. Shape hands the loan officer a documented conversation with an appointment already on it. Your LO opens the record and starts selling.

Cost per call runs below a live answering service. There is no shift change. It handles simultaneous calls, which matters on the Monday after the Fed moves. No answering service staffs for that morning. The call-handling mechanics are covered in more depth on our page for AI calling agents.

What compliance questions come first?

Four, and a branch manager will raise them before they ask about price. Short answers first.

Question Answer What it means operationally
Does the TCPA apply to answering an inbound call with AI? No The FCC’s 2024 ruling governs calls you initiate. Answering is not initiating.
Does it apply to what happens after the call? Yes Callbacks, confirmation texts, and nurture sequences are outbound. Consent rules apply.
Do you have to disclose the AI? Yes, in practice State law is uneven and moving. Disclose on every call and stop tracking it.
Can the agent take an application or quote terms? No That is licensed activity. Scope the agent to intake, scheduling, and routing.

None of this is legal advice. Your compliance counsel gets the final word.

On the TCPA. In February 2024 the FCC ruled that AI-generated voices count as artificial under the TCPA. That governs calls you place. A borrower dialing your number is not one. Your exposure sits in the workflow behind the agent, not on the answer itself. Most failures in this category happen there.

On disclosure. Utah requires disclosure at the outset for regulated occupations collecting sensitive financial information. Colorado is a moving target. Its 2024 AI Act never took effect as written. The legislature replaced it with SB 26-189, signed in May 2026 and effective January 1, 2027. The new statute covers automated decision-making that materially influences a lending decision. It also removed the exemptions that federally regulated entities previously had. Shape’s position is to disclose on every call in every state. It costs four seconds and it ends the question.

On licensing. Under the SAFE Act, a mortgage loan originator takes a residential loan application and offers or negotiates terms. Regulators have not addressed what happens when software does that work. You do not want to be the test case. Shape scopes its agent to contact capture, qualification questions, scheduling, and routing. It does not quote rates, confirm that a borrower qualifies, or discuss program terms.

That constraint is worth stating plainly, because competing products market the opposite. Several advertise AI that runs rate discussions and matches borrowers to FHA, VA, and jumbo programs. One says in its own FAQ that most callers never realize they are talking to a machine. Read those as feature lists and they sound better than ours. Read them as a compliance review and they are a problem. We built AI calling for mortgage around that boundary rather than against it.

When is an AI voice agent the wrong answer?

In four situations. Shape sells this product and I will still tell you when not to buy it.

  • When the first call is the sale. Jumbo, self-employed, non-QM, damaged credit. These callers were told no somewhere else. They are testing whether you will listen. An intake script loses them in ninety seconds.
  • When an existing borrower has a problem. Nobody calling about a closing delay wants a bot. Route those to a human, or configure escalation on the first signal.
  • When a referral partner is calling. Realtors and builders get your cell. Nothing else.
  • When your CRM data is already a mess. An agent writing hundreds of records into a broken system makes a bigger broken system.

Anyone selling this as a universal answer is selling. It fixes one failure. Nobody picked up, and now nobody ever will.

How do you measure the fix?

Four metrics. Shape holds deployments to these and you should too.

  • Answered call rate. This should move to effectively 100%.
  • Contact-to-appointment rate on after-hours calls. These were zeros before. Every booking is recovered revenue.
  • Speed to first meaningful touch. Not speed to log a lead. Speed to a real conversation about the borrower’s scenario.
  • Funded loans sourced from previously missed windows. Tag them at origination. In two quarters you will know what the voicemail era cost.

Run your own numbers

Pull your phone system report for last month. Count the calls that rang out or hit voicemail. Run them through the formula above with your close rate and your revenue per loan.

If the number clears five figures, it is worth a conversation. If it does not, fix your call routing and spend the money elsewhere.

Frequently asked questions

What is the cost of missed calls for loan officers?+

Multiply missed calls per week by 52, then by your close rate, then by revenue per funded loan. Ten missed calls a week at a 3% close and $3,000 per loan is $46,800 a year. A solo LO missing four a week is closer to $18,700.

Is it legal to answer borrower calls with an AI voice agent?+

Yes. Answering an inbound call is not a robocall. The FCC’s 2024 ruling on AI-generated voices governs calls you place, not calls you receive. Consent rules do apply to any outbound call or text the system triggers afterward.

Can an AI voice agent take a loan application?+

No. Taking an application and offering or negotiating terms is licensed activity under the SAFE Act. Scope the agent to contact capture, qualification questions, and scheduling. Route anything about rates or terms to a licensed loan officer.

How is an AI voice agent different from a mortgage answering service?+

An answering service takes a message a human still has to process. An AI agent asks your qualifying questions and books the loan officer’s calendar. It also writes the full conversation into your CRM.

When is an AI voice agent not worth it?+

When your book is concentrated in jumbo, non-QM, and self-employed borrowers, where the first conversation is the sale.