Yes. AI can keep mortgage clients updated at each milestone (application submitted, valuation booked, valuation back, underwriter query, offer issued, completion) by turning your CRM's case status or lender emails into a short, plain message. The rules: it sticks to confirmed facts, never predicts lender dates or outcomes, and passes declines, down-valuations and new conditions to a broker to phone.
The biggest gain isn't the milestone messages themselves. It's the scheduled "nothing new yet" update. Clients in the middle of a purchase mind silence far more than slowness, and most "any news?" calls come in the gaps between milestones. A short weekly message saying where the case is and what's being waited for heads off those calls, and it's the kind of message AI writes well from a status field.
The milestones worth a message, and who delivers each
Not every milestone should be automated. The line is simple: good or neutral news can go automatically; anything that might worry the client or needs a decision goes to a broker.
| Milestone | Trigger | Sent by |
|---|---|---|
| Application submitted | Case status changes in CRM | AI, automatically |
| Valuation instructed or booked | Lender email or portal status | AI, automatically |
| Valuation received, at or above purchase price | Lender email | AI, automatically |
| Valuation received, below purchase price | Lender email | Broker calls; AI drafts a follow-up email for the broker |
| Underwriter needs more documents | Lender email | AI sends the document request, quoting exactly what's needed |
| Underwriter raises a concern or new condition | Lender email | Broker calls |
| Offer issued | Lender email or portal status | AI sends; broker follows with a call about any conditions |
| Offer declined or case referred | Lender email | Broker calls, same day |
| Weekly check-in, no change | Seven days since last update | AI, automatically |
| Completion | Conveyancer or CRM status | AI sends a thank-you; broker follows up about protection and next review |
Law firms face the same problem with their clients, and how small law firms answer client status questions with AI shows a similar split between automatic and human updates.
Where the AI gets its facts from
An update is only as reliable as its source. Mortgage cases have three, of varying quality:
- Your CRM's case status. Reliable if staff update it, useless if they don't. If statuses are behind, automation will broadcast out-of-date news, so fix the habit first.
- Lender emails and portal notifications. The most up-to-date source, but unstructured. AI can read each one, classify it and extract the facts.
- Conveyancer or solicitor updates. Relevant after the offer, and just as unstructured.
Here's how classifying a lender email goes in practice. The email arrives:
"Ref 88213447. Valuation report received. Valuer comments noted re: flat roof to rear extension; case referred to underwriting for review. Further update to follow."
Prompt: Classify this lender email for a mortgage case. Choose
ONE category: submitted / valuation booked / valuation received
- no issues / valuation received - issues / documents requested
/ underwriter concern / offer issued / declined / other.
Extract: case reference, date, facts stated. List anything the
client might need to know. Do not predict what happens next.
Illustrative output:
Category: valuation received - no issues
Reference: 88213447
Facts: valuation report received; case referred to
underwriting for review; further update to follow.
Client needs to know: valuation is back; underwriting review
is next.The classification is wrong, and in a way that matters. A valuer's comment about a flat roof, followed by a referral to underwriting, is an issue: the lender may ask for a roof survey, apply a condition or change the loan. Sending "your valuation is back, the underwriters are reviewing" would be technically true and would leave the client blindsided if a condition appears next week. The fix: add a rule that any mention of valuer comments, referral, retention or conditions puts the email in "valuation received - issues", which goes to a broker. When in doubt, the AI chooses the category that involves a person.
Message templates for each milestone
Templates with gaps for the facts are safer than letting AI write each update freely, and they read more consistently. A filled-in set for an illustrative purchase:
SUBMITTED
Hi [first name], your mortgage application went to the
lender today (Mon 5 Oct). Next, the lender will usually arrange a valuation
of the property. We'll let you know as soon as it's booked.
Questions? Reply here or call the case team on [number].
VALUATION BOOKED
Hi [first name], the lender has booked the valuation for
Thu 15 Oct.
The valuer will contact the seller's agent for access, so you
don't need to do anything. We'll update you when the report is
back.
DOCUMENTS REQUESTED
Hi [first name], the lender has asked for one more thing: your latest
payslip (September). Could you upload it here: [link]. As soon
as we have it, we'll send it on.
OFFER ISSUED
Hi [first name], good news: the lender has issued your mortgage offer.
We've sent a copy to your solicitor. [Broker] will call you
tomorrow to go through the offer and its conditions.
WEEKLY CHECK-IN
Hi [first name], a quick update: your case is with the lender's
underwriters, who are reviewing the valuation. Nothing is
needed from you right now. We'll be in touch as soon as we
hear, and in any case next Monday.Notice what isn't there: no "should", no "hopefully by Friday", no "don't worry". Each message says what happened, what happens next, whether the client needs to act, and how to reach someone.
The one update AI can draft more freely is the follow-up email after a broker's call about bad news, because the broker's call note supplies the facts. An illustrative note: "Down-val 285k vs 295k price. Options: renegotiate with seller, add 10k to deposit, or try another lender. Client to speak to agent, call back Thurs." The AI's draft opened: "As discussed, the valuation came in at 285,000, below the agreed price of 295,000. The good news is that this is common and usually resolved quickly." That second sentence has to go. The broker never said it, and "usually resolved quickly" is a prediction about a negotiation nobody has started. What stays: the three options in the broker's words, the client's next step, and the Thursday call.
Instructions that stop the AI guessing
Whether you use templates or let AI compose updates from the facts, the instructions need hard limits:
- Use only facts from the case record and the latest lender or
conveyancer message. Never add information.
- Never give or estimate a date the lender or conveyancer has
not stated in writing.
- Never describe how likely an approval, offer or completion is.
- Never explain what a condition, retention or valuer comment
means. Say the broker will call to go through it.
- If sources conflict, or anything is unclear, send nothing and
flag the case to the case manager.
- Every message ends with how to reach a person.The date rule causes the most trouble when it's missing. It plays out like this. A lender's automated email says underwriting is "typically completed within 5 working days". The AI, asked for a helpful update, tells the client "your offer should be ready by next Friday". The underwriter asks for extra documents, the offer takes three weeks, and the client, who has told the seller Friday, is furious with the broker, not the lender. The lender never promised a date; the AI turned a typical timescale into a commitment. Service levels quoted in lender emails are not dates for clients.
Document requests go wrong more quietly, when the AI tries to help by translating the lender's wording. An illustrative lender email says: "Please provide evidence of the source of the gifted deposit funds." The first draft to the client read: "Could you ask your parents for a short signed letter confirming the gift?" The client had never said who the gift came from, and the lender may want its own gift form plus the giver's bank statements rather than a letter. The client would have sent the wrong thing and the case would have stalled while everyone found out. The rule: quote the lender's request word for word, attach the lender's form if it sent one, and if the request needs explaining, the broker explains it.
The weekly "nothing yet" update, before and after
Before: No message for eleven days while the case sits with underwriting. The client phones twice and emails once. Each contact takes the case manager five or ten minutes to look up the case and reply, and the client's confidence drops each time.
After: Every Monday, any case without an update in seven days gets the weekly check-in template, filled from the CRM status: where the case is, whether the client needs to do anything, and when they'll next hear. The case manager glances at the day's list before it goes and removes any case where a call is more appropriate. Calls about those cases mostly stop, because the question has been answered before it's asked.
Choosing the channel: email, text, WhatsApp or portal
| Channel | Good for | Watch out for |
|---|---|---|
| Detailed updates, attachments, a clear record | Missed in busy inboxes; fine for the record, weaker for urgency | |
| Text message | Short milestone alerts people actually read | Length limits; avoid sensitive details |
| WhatsApp Business Platform | Conversational updates, easy client replies | Outside a 24-hour window after the client last messaged you, only pre-approved template messages can be sent |
| Client portal | Full case timeline, document uploads | Clients rarely log in unprompted; pair it with a notification |
WhatsApp's rule shapes how AI can be used there. Because messages outside the 24-hour customer service window must use approved templates (update notices fall into Meta's "utility" category), the AI's job on WhatsApp is to fill template fields from the case, not to write free text. Check Meta's pricing page before you rely on WhatsApp for volume: Meta announced changes to how service and utility messages are charged from 1 October 2026. WhatsApp Business app versus API for AI replies explains which version you need.
In practice the AI fills a template like this one, approved once and reused on every case (illustrative):
Template: valuation_booked (utility)
Hi {{1}}, the lender has booked the valuation for {{2}}.
You don't need to do anything. Reply here with any questions.
Sent: Hi [first name], the lender has booked the valuation
for Thu 15 Oct. You don't need to do anything. Reply here
with any questions.If the client last messaged on Tuesday and the valuation is booked on Thursday, the window has closed and only a template can go. If they messaged an hour ago you could reply in free text, but use the template anyway so the wording stays approved. Under the new pricing, free-form replies inside the window are charged once a number passes 1,000 service messages in a month, and utility templates sent inside it are charged with no free allowance, so cost it on your case volume before making WhatsApp the main channel.
Setting it up in four stages
- Clean up case statuses (one to two weeks). Agree a short, fixed list of statuses that match the milestones above, and get everyone updating them the same day something changes. If your CRM already has milestone notifications, switch those on first; many mortgage CRMs send basic status alerts, and the AI's job is then only to make them readable.
- Write and approve the templates (a few hours). One per milestone, plus the weekly check-in, each signed off by a broker. Include the wording for "we'll call you" messages so the automatic and human paths feel like one service.
- Connect the lender inbox (a day or two of building). Route lender and conveyancer emails through an automation that classifies each one, extracts the facts and either fills a template or flags the case. Zapier or Make can do this with an AI step; on Zapier, an AI by Zapier step uses one, three or five tasks per run depending on the model tier, so a busy lender inbox can add up. Price it on your real monthly email count (a worked sum follows this list).
- Run a shadow period (two weeks). Every drafted update goes to a case manager for approval before sending. Log every correction. When the good-news categories go a full week without a correction, let those send automatically and keep approval on everything else.
A quick sum for the 40-case brokerage described below, illustrative: say 150 lender and conveyancer emails a month. Each run is one AI step plus two actions (update the CRM, then send or flag). At the three-task tier that's 5 tasks an email, or 750 a month, which uses Zapier Professional's whole allowance before a single weekly check-in goes out. Add around 160 check-ins at two tasks each and you're at about 1,070. The one-task tier (or your own API key, which counts once) brings the emails down to 450 tasks and the total to about 770, still just over. So either price the next task level up on Zapier's pricing page, or compare the same workflow on Make's credits. Count your real lender emails for a month before choosing.
During the shadow period, keep the correction log as short as possible so people actually fill it in. An illustrative first week:
Mon VALUATION BOOKED Date taken from the email's sent date,
not the inspection date. Fix: prompt
names the field ("inspection date").
Tue WEEKLY CHECK-IN Queued for a case that completed on
Friday; CRM not updated. Fix: completion
logged the same day, no exceptions.
Wed SUBMITTED Joint application; message addressed to
the first applicant only. Fix: send to
every applicant on the case.
Thu DOCUMENTS REQ. Lender asked for two items; message
listed one. Fix: "list every item the
lender requests, in the lender's words".
Fri (none)Four corrections in five days isn't a reason to give up; each became a rule. If the second week brings one correction, in a document request, the good-news categories can start sending on their own while document requests stay under approval.
A small detail that saves a lot of confusion: give every automated update the same sender name and the same reply route, and make sure replies land in a monitored inbox rather than a no-reply address. A client who replies "is the roof a problem?" to an automated message deserves an answer that day.
Replies need sorting as much as updates do. An illustrative morning's replies, and where each went:
- "Thanks, that's great." No action; logged against the case.
- "Uploaded the payslip just now." The AI checks the upload folder, confirms receipt to the client and tells the case manager the document is ready to send on.
- "Does the comment about the roof mean they'll lend less?" Broker, same day. This is exactly the question the AI must not answer.
- "We might pull out, the survey found damp." Broker within the hour, marked urgent, because a purchase at risk changes the whole case.
Two of the four needed a person. If the sorting step ever puts a question about meaning or risk in the no-action pile, send every reply to a person again until you've found out why.
Forty live cases, before and after
As a worked illustration, imagine a brokerage with about 40 cases between application and completion at any one time, and a typical case taking six to eight weeks. Say each case generates three "any news?" contacts along the way, at around eight minutes each including looking up the status. For 40 cases, that's about 120 contacts and 16 hours of case-manager time per cycle, plus the milestone messages sent by hand.
With milestone templates and the weekly check-in, suppose chasing contacts fall by half and milestone messages take seconds to approve rather than minutes to write. That's around eight hours back per cycle from chasing alone, with more from the messages, and clients who feel better looked after. Setup takes perhaps 10 to 20 hours: templates, classification rules, CRM connection and a fortnight of reviewing every message before it goes. Adding human approval steps to AI automations covers how to run that review period without it becoming permanent.
Signs the updates are helping, or annoying people
- Chasing calls and emails per case. Should fall within a month. If not, the updates aren't answering the question clients actually have.
- Replies to automated messages. Some replies are good. A rise in "what does this mean?" replies means a template is unclear.
- Corrections. Any update that had to be corrected is logged, with the cause. Two in a month from the same source means fix the source or the rule.
- Frequency complaints. Weekly suits most clients during the underwriting wait. Daily is too much; fortnightly invites the calls back.
Updates sit inside the wider question of what a broker can automate at all; what mortgage brokers can automate and what stays advice draws that line across the whole case, and the enquiry-to-fact-find workflow covers the stages before submission. If you're choosing systems to run this, choosing a mortgage CRM with AI built in lists what to test.
Automated mortgage updates: common questions
Should the client know the updates are automated?
Yes, in a simple way. Sign automated updates from the firm's case team rather than pretending the broker typed each one, and make every message say how to reach a person. Clients generally don't mind automation for status news. They mind being unable to get an answer when something is worrying them, so the reply route matters more than the wording.
What if the lender's portal status and the email say different things?
Treat it as a conflict for a person to resolve, not something for the AI to pick between. Set a rule that when two sources disagree, no update is sent and the case is flagged to the case manager. Portals sometimes lag behind emails, and emails are sometimes sent before a status changes, so the safe default is silence plus a quick check.
Can the AI answer the client's replies to an update?
It can acknowledge them and handle simple ones, such as confirming a document has arrived or booking a call. Anything that asks what a condition means, whether a problem is serious, or what the client should do next needs the broker. A sensible setup lets the AI sort replies into those two groups and route the second group to a person the same working day.
Further reads
- Is It Safe to Let AI Listen to Mortgage Advice Calls? — The other end of the case: AI on the advice call itself.
- Best AI Tools for Mortgage Brokers at Each Stage of a Case — Tools for every stage, including case tracking.
- How Conveyancers Use AI to Cut Admin on Each Transaction — How the conveyancing side is automating its updates.
- Is It Safe to Let AI Reply to Customers on WhatsApp? — Safety checks before AI talks to clients on WhatsApp.
- How to Measure Time Saved After Rolling Out AI in a Small Firm — Measure the drop in chasing calls properly.
- How to Manage Your Inbox With AI: Triage, Drafts, and Follow-Ups — Sorting lender and client emails before anything else.
- Using AI to Spot Remortgage Opportunities in Your Client Bank — Turn an untidy client bank into a watchlist of deal-end dates, early-repayment-charge dates and loan-to-value moves, with AI filling the gaps.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Meta for Developers, WhatsApp Business Platform pricing and messaging documentation (customer service window, template categories, announced pricing changes); Zapier pricing and task-counting help pages. Checked September 2026. Timings and case volumes are illustrative.