What works and what does not at mortgage firms
Late last year, we carried out an AI audit at a Dutch mortgage firm with 35 employees and an average of 60 financing requests per day. It surfaced eight concrete opportunities, five of which are now in production or in implementation. We did not take on several others, because they sit too close to the Wft line legally or professionally.
The overview below is based on what we see working there and at comparable firms. No hype, no marketing, but figures and pitfalls.
Use case 1: classifying and pre-filling financing requests
What it does. A request comes in by email, form or the adviser portal. An AI layer reads it, recognises the type of financing (home, rental property, refinancing, bridging), the complexity (NHG limit, BKR, tax aspects) and the urgency. It immediately fills the right fields in the CRM and links the file to the adviser who fits in terms of specialism and schedule.
What it delivers. In practice, around 70 hours per month on a team of 35 people. More importantly, the lead time from first customer contact to full intake roughly halves.
Pitfall. Never rely on the classification alone. AI output must always be escalatable to human validation within the first 24 hours, otherwise an error slips through. So always build in a review step in the first month.
Use case 2: advice-meeting summaries straight into the file
What it does. During an advice meeting (phone, video or in person), a transcription tool runs along. Right after the meeting, the AI generates a structured summary, mapped onto the Wft-required advice elements (customer situation, objective, knowledge and experience, risk appetite, advice and rationale). The adviser reviews, corrects where needed, and signs off.
What it delivers. Per adviser, 4 to 6 hours less admin per week. With 8 advisers, that is already 32 to 48 hours per week.
Pitfall. The summary is an aid, not a replacement for the written advice the Wft requires. The adviser remains ultimately responsible for the content. That must be explicitly recorded in your workflow too. Do not make a grey area of it.
Use case 3: document recognition during document submission
What it does. The customer uploads payslips, annual statements, BKR reports and purchase deeds through a portal. AI immediately recognises which document it is, checks completeness (3 months of payslips, the correct employer's statement), extracts the relevant fields and places them in the file. For missing or incomplete documents, the system automatically sends a reminder to the customer.
What it delivers. For an average file of 12 documents, this saves 30 to 45 minutes per file of manual entry and checking work. On a busy Monday's pile, that means considerably less backlog on Tuesday morning.
Pitfall. Document recognition struggles with poorly scanned PDFs, screenshots of mobile banking apps, and old Word documents without an OCR layer. Always plan a fallback for "AI does not know" with human handling, otherwise 10 to 15 percent of your files get stuck in an invisible queue.
Use case 4: automated customer communication at milestones
What it does. When a file passes a milestone (feasibility check approved, lender approves, valuation in, deed signed), AI automatically generates a suitable email to the customer in the firm's house style. The adviser sees the draft in the queue, adjusts where needed, and sends it.
What it delivers. Customers experience tighter communication, advisers no longer lose time on standard messages. On a team of 8 advisers, around 6 hours per week per person.
Pitfall. AI must never send independently when it comes to financial communication. Human in the loop, always, even if it only takes 30 seconds. Otherwise you get situations where an approval email goes out while an additional lender condition is still outstanding.
Use case 5: after-the-fact compliance monitoring on advice meetings
What it does. After an advice meeting, an AI runs through the transcript and flags potential Wft points of attention. Did the adviser ask about knowledge and experience? Was risk appetite explicitly discussed? Was a suitable rationale given for the final advice? It flags, it does not judge.
What it delivers. For the compliance officer or senior adviser, this saves considerable time. Instead of listening back to meetings at random, he gets a list of meetings where the AI saw a point of attention. He then listens in a targeted way.
Pitfall. AI is not a Wft regulator. It is an aid to monitor more systematically, not a replacement for the human compliance eye. Make that explicit in your internal governance.
What this concretely delivers for a 30-FTE office
If you implement use cases 1 through 4, conservatively calculated, that saves between 200 and 280 hours per month at the team level. That is 1.5 to 2 FTE no longer tied up in repetitive work but available for advisory time or growing the office. At average adviser rates, you recoup the investment within 6 to 9 months, whether that runs through a five-to-eight-week implementation project or through an AIOS retainer in which we stay structurally connected as a partner.
Use case 5 is not primarily about saving time but about reducing risk. A well-documented Wft file is worth its weight in gold if the AFM comes knocking.
The order we recommend
Do not do all five at once. They are not all equally quick to achieve, and the dependencies run between them. In practice, this order works for most firms. First use case 1 (classifying) and use case 3 (document recognition), because they give immediate overview and break open the intake bottleneck. Then use case 4 (customer communication), because it sits operationally close to what you have just built. Use case 2 (meeting summaries) and use case 5 (compliance monitoring) come after that, because they require more governance work up front.
Each step builds on the previous one. Whoever does it in this order sees the first effect within three months and the full 200 to 280 hours per month within a year.
The first step
In an AI audit, we map for your specific office which of these use cases deliver the most, in which order you best implement them and which integrations you need. Fourteen days from the interview day, fixed price, a report you can put on the management agenda tomorrow.
AI that works in your field starts with knowing where the most time stalls for you and which use case is the logical first step for your office.
