What problem does it solve?
A home loan is a large, long commitment, and the questions start well before an application: how much can I borrow, what deposit do I need, fixed or variable, what happens when my fixed rate ends, which documents will you want. Customers research when it suits them, often outside the hours when specialists are available, and a slow or vague answer can send them to a broker or a competitor.
Every hour a specialist spends on early questions, or on applicants who were unlikely to qualify, is an hour not spent on applications. And because rates and credit policy change, a generic assistant, or a web page that is out of date, can give answers that are wrong in a way that matters: a borrowing estimate that is too high, or a rate that no longer exists.
How does it work?
- Answer from current, approved content. Questions about rates, fees, loan to value limits, product features and documents are answered by retrieval over versioned product and policy documents, with the source and date shown, and a refusal when the content does not cover it.
- Estimate, clearly labelled. Borrowing power and repayment estimates come from the bank's published calculator logic, called as a tool, never from the model's own arithmetic, and are shown as indicative, not an offer.
- Pre qualify against published rules. Basic checks (deposit, income bands, residency, property type) come from a rules service and tell the customer what is likely to be needed, without a credit decision.
- Prepare the handover. The assistant captures the customer's situation and questions, lists the documents they will need and books a call or meeting with a specialist.
- Serve existing borrowers too. Fixed rate expiry, switching products, extra repayments and offset questions come from the same content, with account specific actions behind authentication.
- Audience
- Customer facing
- Autonomy
- Supervised agent
- Adoption
- Early adopters
- Channels
- Web chat, Mobile app, WhatsApp, Phone and voice
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 |
|---|---|---|---|---|
| Interactions handled | Not pooled | about 900 | 1 | 1 vendor |
Value drivers: Revenue growth, Customer experience, Employee productivity, Speed and cycle time.
Indicative value
A lender handling 100,000 home loan enquiries a year
USD 300,000 to USD 1.5 million
Specialist time freed from early enquiries per year
How this is calculated
Formula: enquiries * handledShare * minutesPerEnquiry * costPerMinute. The low scenario uses every low input, the high scenario every high input.
| Input | Low | High | Basis |
|---|---|---|---|
| Home loan enquiries per year enquiries, enquiries per year | 100,000 | 100,000 | The reference lender. |
| Share of enquiries answered without a specialist handledShare, fraction of enquiries | 0.3 | 0.5 | Editorial assumption; early questions about rates, fees and documents are the ones the assistant can answer. |
| Specialist minutes per early enquiry minutesPerEnquiry, minutes per enquiry | 10 | 20 | Editorial assumption, replace with your own time data. |
| Fully loaded cost of a specialist minute costPerMinute, USD per minute | 1 | 1.5 | Editorial assumption, replace with your own fully loaded cost. |
What it leaves out: Counts specialist time only, which is capacity freed rather than cash saved unless staffing changes. It leaves out extra settled loans from faster, out of hours answers, retention of borrowers at fixed rate expiry and the cost of the AI and content upkeep.
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.
Figure
United States · Banking · 2025
Figure, a US fintech that offers home equity lines of credit, uses Gemini models to run chatbots that simplify and speed up the lending experience for consumers and for its own staff. No outcome figures were published.
No outcome disclosed.
Loft
Brazil · Real estate · 2025
Loft, a real estate technology and financial services company active in Brazil and Mexico, moved its data to Google Cloud and built the Assistente Loft on Gemini. Brokers at the real estate agencies connected to Loft (about 9,000, according to Google Cloud) use it on WhatsApp, by text or voice, to compare home financing conditions from different banks in seconds, so a buyer knows their borrowing power before choosing a property. Google Cloud reports about 900 financing simulations a week on the company's WhatsApp channel.
- Interactions handled: about 900, per week, home financing simulations on WhatsApp
"Cerca de 900 simulações de financiamento por semana no WhatsApp e corretores de 9 mil imobiliárias conectados"
Claimed by: vendor
Safe Rate
United States · Banking · 2025
Google Cloud describes Safe Rate as a digital mortgage lender that uses Gemini models to create an AI mortgage agent with chat features called "Beat this Rate" and "Refinance Me", which let borrowers compare rates and get a personalised quote in under 30 seconds. Safe Rate's own website now presents it as a US mortgage shopping service that ranks lenders, and says its AI assistant answers plain English questions about rates, lenders and local costs of ownership. Neither source reports outcome figures.
No outcome disclosed.
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 rate sheets, product terms and credit policy summaries with owners and effective dates
- The bank's borrowing power and repayment calculator logic as a callable service
- Published eligibility criteria for pre qualification
- Specialist calendars and booking rules
Systems to integrate
- Rates and product content management
- Calculator and rules services
- Scheduling system for specialists and brokers
- CRM for leads and handover summaries
- Loan origination for existing application status, behind authentication
Complexity: Medium
Answering from documents is straightforward. The hard parts are keeping rate and policy content current to the day, reusing the bank's own calculators instead of letting the model compute, and a clean booking handover to specialists.
- 1
Build the content spine first
Collect every rate sheet, product term and policy summary, assign an owner and an effective date to each, and remove anything unofficial. The assistant is only as good as this set.
- 2
Wrap the calculators as tools
Expose the bank's existing borrowing power and repayment calculators to the assistant, so every number matches what a specialist would show.
- 3
Write the boundaries
Decide what the assistant may say about eligibility and what is reserved for a specialist, and phrase estimates as indicative every time.
- 4
Make booking the default next step
End qualified conversations with a specialist booking and a summary, so the conversation turns into an application rather than a dead end.
- 5
Test with rate changes
Include rate change days in the test plan: update the content, rerun the test suite and check that no old rate survives anywhere.
Guardrails
- Rates, fees and limits only from dated, approved content, with a refusal when not covered
- All numbers from the bank's calculators, labelled as indicative estimates
- No credit decision or approval language
- Personal financial details masked in prompts and logs
- Handover to a specialist for complex situations and vulnerability signals
KPIs to instrument
- Share of enquiries answered without a specialist, with repeat contacts counted
- Specialist bookings and applications started per conversation
- Accuracy of answers on a weekly checked sample, including rates quoted
- Out of hours share of conversations
- Complaints mentioning the assistant
Human in the loop
Mortgage specialists own advice, applications and every credit decision. Product owners approve the content set and the calculators, and a sample of conversations is reviewed weekly for accuracy and for any wording that sounds like advice or approval.
Common failure modes
- Yesterday's rate
- A superseded rate stays in the index. Version content with effective dates and retire old versions on the day of change.
- The model does the maths
- The assistant calculates a repayment itself and gets it wrong. Route every number through the calculator tool.
- Advice by accident
- The assistant recommends a product for a person's circumstances in a market where that is regulated advice. Keep to general information and hand over for recommendations.
- Booking dead ends
- The customer is told a specialist will call and nobody does. Book into real calendars and track kept appointments.
What are the risks and rules?
EU AI Act
Depends on design
Answering questions and giving indicative estimates from published rules is limited risk with an Article 50(1) disclosure that the customer is talking to an AI system. If the assistant evaluates an individual's creditworthiness to decide or filter access to a loan, it falls under Annex III point 5(b) and is high risk.
Rules that apply
Guidance
- Directive 2014/17/EU on credit agreements for consumers relating to residential immovable property (European Union, Europe). The Mortgage Credit Directive sets rules on advertising, standard information and creditworthiness assessment that the assistant's answers must respect.
- Responsible lending (Australian Securities and Investments Commission, Asia Pacific). Example of national responsible lending obligations that shape what pre qualification may say.
- Annex III, high risk AI systems referred to in Article 6(2) (European Union, Europe). Annex III of Regulation (EU) 2024/1689 (the AI Act). Point 5(b) covers creditworthiness evaluation of natural persons.
Controls to put in place
- AI disclosure and a statement that estimates are indicative
- Content inventory with owners, effective dates and retirement on change
- Calculator and rules services under change control and tested
- Log of every estimate shown, with the inputs used
- Weekly accuracy sampling and complaint monitoring
Frequently asked questions
- Can a home loan chatbot tell me how much I can borrow?
- It can give an indicative estimate from the bank's own calculator and published criteria, clearly labelled as such. A real borrowing amount needs a full application and a credit assessment by the bank.
- Who uses AI assistants for mortgages today?
- Examples include Safe Rate in the US, whose AI assistant answers questions about rates and lenders, and Loft in Brazil, whose Gemini assistant lets real estate brokers run about 900 home financing simulations a week on WhatsApp, according to Google Cloud. Figure uses AI chatbots in home equity lending. None of these sources report conversion results.
- How do you keep answers current when rates change?
- Treat rates and policy as versioned content with owners and effective dates, retire old versions on the day of change and rerun an automated test set that checks the quoted rates.
How to cite this page
Blits.ai AI Use Case Library, "AI home loan assistant with pre qualification", last verified 27 September 2026, https://www.blits.ai/ai-use-cases/home-loan-assistant-and-prequalification. Licensed under CC BY 4.0. Method: how we verify use cases.
Changelog
- 27 September 2026: First published