AI for mortgage brokers: document chasing and follow-up

Where the admin time goes, and what AI can take over

Most of a mortgage brokerage's office week is not advice. It is the trail of messages that sits between a borrower saying yes and a lender receiving a complete file. Someone asks for a pay stub again. Someone checks whether the three months of bank statements are the right three months. Someone emails the borrower's accountant for a notice of assessment in Canada, or a letter from an employer for a payslip in the UK, then calls again a week later. Someone updates the referring estate agent or accountant. Then the lead that went quiet in the spring gets nothing until a rate change prompts a call.

An AI worker is software that runs a defined set of these tasks from a set of rules and a connected inbox, document store or CRM. In this business it can request documents from a checklist that matches the lender's requirements, track what has been received and what is still outstanding, send reminders on a schedule the broker sets, send plain status updates, and draft follow-up messages for dormant leads. It does not decide whether a borrower qualifies. It does not read a bank statement and conclude that income is stable. It does not tell a borrower which product suits them.

That division matters for the office team. Licensing assistants, processors and administrators do not disappear from the file. Their work shifts from typing and chasing to checking the exceptions that the AI worker flags, such as a statement with a missing page, a name that does not match the application, or a borrower who has not replied after several reminders.

The whole workflow, end to end

The order below follows a typical file from first enquiry to renewal. The detail varies between brokerages, but the stages are common across Canada and the UK.

1. Enquiry and first contact

Today, a receptionist or licensing assistant logs the enquiry from a website form, a phone call or a referral, checks the caller's basic details and books a first call with a broker. Enquiries often arrive out of hours and sit in a shared inbox until the next morning. With an AI worker, the enquiry is logged in the CRM the moment it arrives, the borrower receives an acknowledgement that names the next step, and the broker's calendar is offered two or three slots. The broker still takes the call and decides whether the file is one the firm will take on.

2. Fact-finding and consent

Today, the broker or an assistant collects income, employment, property and existing debt details, often on a paper or PDF form, then types them into the lender's system. The AI worker can send the fact-find questionnaire, check that every mandatory field is answered, and return incomplete answers for correction. It does not interpret the answers. Where the firm holds consent for credit checks or for sharing data with lenders, the AI worker can record that consent date and store the signed form, but the broker confirms that the consent wording is the one the firm uses.

3. Document request and collection

This is the largest block of repeated work. Today, an assistant writes a document list for each borrower, sends it by email, saves attachments into a file folder, renames them and notices what is missing. The AI worker sends the list built from the lender's checklist and the borrower's employment type, files each upload to the right file record, and reports the gaps. A self-employed borrower and a salaried employee get different lists. The AI worker is set up to follow the list, not to decide which list a borrower needs. The broker makes that call.

4. Chasing missing bank statements

Chasing is where most follow-up time goes. A borrower who has sent five of six statements is usually not the problem. The sixth is. Today, the assistant emails, calls, emails again, and records the outcome in a spreadsheet. The AI worker sends a reminder at the interval the broker sets, names the exact statement period still missing, and escalates to a person after a set number of attempts. It does not read the statements for affordability or flag a transaction as suspicious. A person does that review.

5. Submission to lender and underwriting queries

Today, the broker or processor checks the complete file, prepares the lender submission and answers underwriter questions, which often arrive as short emails with a deadline. The AI worker can assemble the file index, check that each document on the lender's checklist is present, and route underwriter questions to the right person with the original document attached. The broker writes the response. The AI worker does not write the explanation of income or the case for the borrower.

6. Status updates

Borrowers ask the same question several times a week: where is my file? Today, the answer depends on who picks up the phone. The AI worker can send a short status message at each stage change, such as documents received, submitted to lender, conditional approval received, or awaiting valuation. Referring professionals can receive a separate, shorter update if the broker has agreed that. Any message that states a rate, a term, a condition that changes the deal, or a view on the outcome is drafted for the broker to approve first.

7. Conditional approval, offer and completion

Today, the admin team tracks conditions, valuation reports, insurance certificates, solicitor or conveyancer details and completion dates across several inboxes. The AI worker can keep a single list of open conditions per file, send reminders to the party who owes the document, and notify the broker when a deadline is close. A licensed person confirms that a condition has been met in substance, not only that a document arrived.

8. Dormant leads and renewals

Many brokerages hold a long list of past enquiries and past clients whose fixed rate is coming to an end. Today, this list is worked by hand, if at all. The AI worker can send a planned message to leads who have not replied in a set period, and flag renewal dates for the broker's review. The broker decides which clients get a call and what the conversation covers. Consent rules apply to marketing messages, so the AI worker should send only to contacts the firm is permitted to contact for that purpose.

Staff today and with AI workers

The table below lists the office roles a mortgage brokerage usually employs, using the titles that appear in day-to-day use. It shows what each role does now, what an AI worker takes over, and what stays with a person. The table describes task types, not headcounts. Every firm's split will differ.

In practice, the first two rows change most. The processor and receptionist rows change in volume, because the number of chase emails falls. The bookkeeper row changes least, because most accounting work still needs a qualified person's judgement.

Six to ten automations ranked by effort and impact

The ranking below weighs how hard each automation is to set up against how much routine work it removes. Effort includes the time to map the checklist, connect the systems and agree the rules with the compliance owner. The ranking is a judgement by the team, not a measured result. Impact will depend on the file volume and the number of chase cycles in the firm's own records, which a free audit measures.

Items one to four are the best starting point. They touch the most repeated work and need the least change to how the broker runs a file. Items nine and ten need clean records and a stable accounting system first.

Compliance and regulatory points

Mortgage brokerage is regulated, and the regulator's rules decide what any message can say and who must approve it. The rules differ by country and by province or nation, so the firm should read the rule in its own jurisdiction before switching any automation on.

In Canada, mortgage brokers and agents are licensed at provincial level. In Ontario, the Financial Services Regulatory Authority of Ontario licenses mortgage brokerages and agents and sets conduct and disclosure duties, so the firm should check the FSRA's own guidance on what must be disclosed and when. Canada also applies federal anti-money laundering rules to mortgage businesses. The Financial Transactions and Reports Analysis Centre of Canada publishes the obligations on identity verification and record keeping, and an AI worker that stores identity documents must keep them within those rules. Canadian privacy rules under the federal Personal Information Protection and Electronic Documents Act apply to how borrower data is collected, stored and shared.

In the UK, the Financial Conduct Authority regulates mortgage advice and administration. Its Mortgage Conduct of Business sourcebook sets out what must be disclosed to borrowers and how. The UK General Data Protection Regulation governs personal data in the UK, and the Information Commissioner's Office publishes the guidance on automated messaging and records.

Three points apply in both countries. First, a status update that states a rate, a term or a likely outcome is advice or a regulated communication, and a licensed person must approve it. Second, marketing messages to dormant leads must respect consent and opt-out rules. Third, the brokerage keeps the record of who approved each outgoing message. An AI worker can log what it sent, but the firm remains responsible for the content.

This explains the rule in general terms; it is not legal or tax advice for your case.

How the transition happens without disruption

Start with the existing team and the existing file process. Do not replace a licensing assistant on day one. Ask the assistant to write down every chase message they send in a typical month, and the replies they get. That list becomes the first set of rules for the AI worker.

Run the first automation on one stage only, usually document requests and statement reminders, for a set period. Keep the person who previously chased documents as the owner of the exceptions queue. Exceptions include a borrower who replies with a complaint, a document that does not match the application, a request for a different product, and any message that mentions a rate or a term.

Roles shrink to review only after the exceptions queue is stable and the broker has checked a sample of outgoing messages. A role that once spent most of its day chasing now spends that time on the files the AI worker cannot settle. That is a change in the job, and it should be agreed with the person. Where a reduction in hours or a role is being considered, the notice, consultation and severance rules in the country apply, and the change should be agreed with the employee or checked with a lawyer.

Common mistakes

• Letting the AI worker pick the document list. The list must follow the lender's checklist, and the broker must confirm which list applies to each borrower.
• Sending status messages that quietly state a result. A message that says the file is with the lender is fine. A message that implies approval is not.
• Building the workflow before the checklists are clean. If the firm's document lists disagree with each other, the AI worker will repeat the confusion faster.
• Treating reminders as harmless. A borrower who receives six reminders in a week may complain, and the complaint is the brokerage's responsibility.
• Skipping the record of approvals. Without a log of who approved each outgoing message, the firm has no evidence for a regulator or a complaint file.
• Automating a firm with very few files. A brokerage with a handful of files a month may spend more time on setup and checking than it saves. Manual chasing is the better choice at that volume.

Limits of automation in this business

An AI worker does not make lending decisions. It does not assess affordability, choose a lender, interpret a credit file, or give mortgage advice. It cannot replace the broker's judgement on whether a borrower should proceed, and it should not be set up to try. It also cannot fix a poor document checklist or an unclear file process. Those need a person to decide the rules first.