What problem does it solve?
After closing, a mortgage becomes a long relationship made of small requests: what is my payoff amount, why did my escrow payment change, how do I enroll in autopay, where is my year end tax form, who services my loan now. A servicer receives these questions by phone and live chat, mostly in business hours, and every one needs a person who has the loan record open.
Volume is uneven. A regulatory event, a tax season or a servicing transfer can push calls up in days, and the questions repeat. The calls that need judgment (a borrower in hardship, an insurance premium the borrower cannot carry, a dispute) wait in the same queue as the routine ones.
How does it work?
- Identify the borrower. The assistant verifies the caller or chat user with the same checks an agent would use, and opens the loan record through the servicing platform.
- Understand the request. A language model classifies the request (payment, autopay, escrow, tax form, transfer, forbearance information) and, where the answer depends on the loan, reads the data from the servicing system.
- Answer or act. For information it explains the answer in plain language. For routine actions it completes the step inside the limits the servicer has set, for example autopay enrollment or sending a document. It can also place routine outbound calls the servicer schedules, such as occupancy or repair confirmations on a reverse mortgage, and log the outcome.
- Hand over. Hardship, disputes, complaints and any case the assistant cannot resolve go to a servicing specialist, with the conversation and the borrower's context attached. Borrowers can ask for a person at any point.
- Review. Servicing leads sample conversations, monitor missed intents and update the knowledge as regulation and investor rules change.
- Audience
- Customer facing
- Autonomy
- Supervised agent
- Adoption
- Early adopters
- Channels
- Phone and voice, Web chat, SMS
What is it worth?
Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.
| KPI | Median | Reported range | Data points | Claimed by |
|---|---|---|---|---|
| Containment rate | Too few to pool | 75% | 1 | 1 vendor |
| Interactions handled | Not pooled | at least 1 million | 1 | 1 organization |
| Contact deflection | Too few to pool | Not pooled: up to 12% | 0plus 1 up to | 1 organization |
Value drivers: Lower cost to serve, Customer experience, Speed and cycle time.
Indicative value
A mortgage servicer with 500,000 loans in its portfolio
USD 187,500 to USD 2.2 million
Human handled servicing contact cost avoided per year
How this is calculated
Formula: loans * contactsPerLoan * routineShare * containment * costPerContact. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Loans serviced loans, loans | 500,000 | 500,000 | The reference servicer. |
| Assisted contacts per loan per year contactsPerLoan, contacts per loan per year | 1 | 2 | Editorial assumption, replace with your own call and chat volume. |
| Share of contacts that are routine servicing routineShare, fraction of contacts | 0.5 | 0.7 | Editorial assumption, replace with the share from your contact reason report. |
| Share of routine contacts the agent resolves containment, fraction of routine contacts | 0.15 | 0.4 | Conservative against the evidence on this page. ServisBOT reports 75% containment for Cenlar FSB's chat bot during office hours, a vendor figure for chat only. This range also covers phone calls and a wider request mix. |
| Cost of a human handled contact costPerContact, USD per contact | 5 | 8 | Editorial assumption for a blended phone and chat servicing contact, replace with your fully loaded cost. |
What it leaves out: Gross avoided contact cost only. It leaves out the cost of running the AI, integration and compliance review, the value of 24/7 coverage and any effect on complaints or retention.
Who already uses it?
3 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.
Longbridge Financial
United States · Banking · 2026
Longbridge Financial, a US reverse mortgage lender and servicer, put AI voice agents live in January 2026 and announced them on 3 February 2026. They handle routine servicing calls, such as outbound checks that a borrower still lives in the home and that required repairs are done, and transfer callers to a servicing professional when needed. The agents give 24/7 coverage, borrowers can opt out and reach a person, and the company says staffing levels are unchanged. Its chief operating officer says it will take a few months of data to know how much time the agents save, so no outcome is published yet.
No outcome disclosed.
Rocket Companies
United States · Banking · 2026
Rocket Companies scaled an AI Voice platform across its mortgage servicing operations in the second quarter of 2026. In its earnings release the company says the system handled more than 1 million inbound calls within three months of launch, that more than half of those calls would otherwise have needed a servicing team member, that task resolution was nearly 25% faster than its old IVR and that client satisfaction reached 4.5 out of 5.
- Interactions handled: at least 1 million, inbound calls within three months of launch
"Within three months of launch, AI Voice handled more than 1 million inbound calls."
Claimed by: organization
Cenlar FSB
United States · Banking · 2021
Cenlar FSB, a US loan subservicer, deployed a ServisBOT bot in 2020 to answer borrowers' CARES Act forbearance questions across IVR, SMS and web, built in about 30 days, then extended it with a borrower bot in front of live chat for payments, autopay, loan transfers, tax forms and escrow. Cenlar's own executive says 10 to 12% of hardship callers chose the bot over an agent. ServisBOT reports a 75% containment rate during office hours for the chat bot. Cenlar's own site now describes newer generative AI bots for homeowners, including an authenticated escrow bot.
- Contact deflection: up to 12%, borrowers calling about a hardship who opted in to the bot instead of an agent
"Once the bot was turned on and offered in our IVR, 10-12% of borrowers who indicated they were calling about a hardship, opted in to an SMS with a link to our website instead of talking to an agent."
Claimed by: organization - Containment rate: 75%, during office hours, borrower bot in front of live chat
"During office hours, Cenlar is experiencing a 75% containment rate with only one out of four requests passed to the live agents."
Claimed by: vendor
How do you implement it?
A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.
Data you need
- Current servicing policies, investor guidelines and borrower facing regulatory notices
- Historical call and chat transcripts, labeled by request type and outcome
- Loan level data exposed through the servicing platform, with borrower verification
Systems to integrate
- Mortgage servicing platform (loan, escrow, payment and document data)
- Telephony or IVR and live chat platform, with handover to the agent desktop
- Identity verification and consent records
- Quality monitoring and call recording
Complexity: Medium
The build needs a read and write connection to the servicing platform, caller verification that matches the contact center's current checks, a knowledge base of servicing policies and regulatory notices, and a handover path into the agent desktop. Actions that change a loan record need approval rules from compliance before they go live.
- 1
Start with information only requests
Launch on questions that need no change to the loan record, such as payment status, escrow explanations and tax form requests. Measure containment and accuracy before any action is allowed.
- 2
Add routine actions one at a time
Enable autopay enrollment, document delivery and similar steps individually, each with its own limits and a compliance sign off. Keep an audit record of every action.
- 3
Build the handover before the assistant
Agree with the servicing team which topics always go to a person (hardship, disputes, complaints, bereavement) and make the handover carry the full context.
- 4
Offer an opt out at every step
Tell the borrower they are speaking with an AI agent and let them choose a human at any point, in line with the servicer's disclosure duties.
- 5
Review misses weekly
Read missed intents and escalations each week, add them to the knowledge base or to the always escalate list, and rerun the test set before each change goes live.
Guardrails
- The assistant discloses that it is an AI agent and offers a human at any point
- Hardship, loss mitigation, dispute and complaint topics always route to a person
- Answers about balances, payments and escrow come from the servicing system, not from model memory
- Borrower data is verified before any loan specific detail is shared, and personal data is masked in logs
KPIs to instrument
- Share of conversations resolved without a person, by request type
- Share of escalations that the borrower asked for versus the assistant triggered
- Accuracy of answers checked against the servicing record on a weekly sample
- Borrower satisfaction after the interaction
Human in the loop
Servicing specialists take every escalated case and own all hardship and dispute decisions. A quality team samples completed conversations each week and checks them against the loan record, and any change to the allowed actions needs compliance approval.
Common failure modes
- Wrong loan specific answer
- An escrow or payoff figure given from stale data or inferred by the model creates a complaint and possible regulatory exposure. Read figures only from the servicing system and sample answers against it.
- Borrower in distress kept in the bot
- A borrower in hardship who cannot find the route to a person is the "doom loop" the CFPB warns about in its chatbot report. Detect distress words and hardship intents early and route them straight to a specialist.
- Authentication gaps
- Weak verification on voice lets a caller reach another borrower's data. Use the same or stronger checks than the human desk and log every verification outcome.
What are the risks and rules?
EU AI Act
Limited risk (transparency)
A customer facing assistant that talks to borrowers is subject to the Article 50 transparency duty to tell users they are interacting with an AI system. It does not evaluate creditworthiness. If a deployment later influences loss mitigation or credit decisions for natural persons, reassess it against Annex III point 5(b).
Guidance
- Article 50, transparency obligations for providers and deployers of certain AI systems (European Union, Europe). People must be informed that they are interacting with an AI system unless this is obvious from the context.
- Chatbots in consumer finance (Consumer Financial Protection Bureau, North America). Warns that chatbots which cannot resolve a problem and block access to a human ("doom loops") can harm customers and lead to violations of consumer financial law.
Controls to put in place
- Disclosure that the borrower is talking to an AI agent, with a human option at every step
- Compliance approval for each action the assistant may take on a loan record
- Audit trail of identity checks, answers and actions
- Recorded and reviewable conversations with personal data masked in analytics
Frequently asked questions
- What share of servicing calls can an AI assistant resolve?
It depends on the request mix, and few servicers publish a resolution rate. ServisBOT reports a 75% containment rate during office hours for Cenlar FSB's borrower bot, a vendor figure for chat. Rocket Companies reports volume rather than a resolution rate: its AI Voice handled more than 1 million inbound servicing calls within three months of launch.
- Which servicing topics should never be automated?
Hardship and loss mitigation, disputes, complaints and anything where a borrower is in distress should go to a person. Longbridge Financial's chief operating officer, for example, says it wants its human agents to talk with borrowers on fixed incomes who are worried about rising homeowners insurance costs.
- Must the borrower be told it is an AI?
Under Article 50 of the EU AI Act users must be told when they interact with an AI system unless it is obvious. In the United States the CFPB has warned that chatbots which leave customers without a route to a person can harm them and risk violations of consumer financial law. Longbridge Financial lets borrowers opt out of the AI interaction and speak with a person.
How to cite this page
Blits.ai AI Use Case Library, "AI assistant for mortgage servicing borrower requests", last verified 11 November 2026, https://www.blits.ai/ai-use-cases/mortgage-servicing-borrower-assistant. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 11 November 2026: First published