AI for Swiss real-estate brokers and property managers: listings, leases, applicant triage – and AML obligations

What the industry covers – and where AI fits

Swiss real estate is broad: about 3,500 brokerage firms organised in SVIT Switzerland, plus around 2,000 property managers, several hundred valuation and architecture-adjacent offices, and the umbrella SVIT/USPI with about 12,000 industry members combined (SVIT annual report 2024). Business splits cleanly: (a) marketing and sale (brokerage), (b) leasing (first letting and ongoing), (c) management (operations on behalf of owners), (d) valuation and advisory.

In all four businesses text work is central: exposes, listings, applicant correspondence, leases, management reports, owner letters. That text work is the natural home for language models.

The Swiss real-estate software landscape: ImmoTop2, Garaio REM2, REM Verwaltung, AbaImmo, RIMO R5, immoNX, plus marketing platforms Homegate, ImmoScout24, Newhome, Comparis. Most industry tools shipped first AI features in 2024/2025 (listing generator in Homegate, applicant score in some portals). Whoever wants to orchestrate own workflows across platforms typically adds an n8n layer.

Particularly sensitive: since the 2021 AML revision, brokerages are AML-bound in two cases: cash transactions above CHF 100,000 and brokering to clients domiciled in high-risk jurisdictions. FINMA refined the duty in a 2024 circular – whoever uses AI in customer identification must show a documented review of the risk logic (see gwg-revision-2026).

Why it matters now

Three market trends hit the industry in parallel in 2025/2026.

First: rental market under pressure. Swiss residential vacancy is below 1.1 percent in 2025 (BFS vacancy count 2025), below 0.3 percent in Zurich, Geneva, Lausanne and Zug. Each listed flat attracts 60 to 200 applications – brokers must devote 30 to 90 seconds of attention per application or lose usable tenants. Pre-qualification AI sorts a pre-selection by formally verifiable criteria (income, debt collection register request, employment duration) which the broker then checks.

Second: purchase market under pressure, other direction. Mortgage rates rose 2023-2025, transaction volume slightly down, ownership properties stay longer on the market. Offices producing exposes faster, multilingually (DE/FR/IT/EN) and better personalised attract more inquiries.

Third: compliance pressure. The 2021 AML revision and the 2024 FINMA refinement require brokerages to document process review when AI tools are used in client identification or risk assessment. Producing an applicant score that co-decides on the lease puts you in EU AI Act high-risk territory (Annex III, personal scoring) – even though Switzerland does not adopt the AIA directly, the standard returns through DPA review (revDPA Art. 21, automated individual decisions).

For most offices that means: AI is welcome for text and triage – decisions on applicants stay with a human. The score is a tool, not a verdict.

Four high-leverage workflows

Four applications deliver the bulk of value.

Listing and expose generation. From the CRM object data (address, location, area, equipment, condition) a language model produces a listing text in the office tone. Multilingual (DE/FR/IT/EN) without retyping. Optional: SEO module (location terms, public-transport links, school district). Time per listing: 3 to 6 minutes instead of 20 to 35.

Applicant pre-qualification (NOT auto-decision). A pipeline reads incoming application PDFs (pay statement, debt-collection extract, ID copy) by OCR (ai-belegerkennung-ocr), extracts mandatory fields (gross income, number of open debt collection cases, employment duration, household size) and assigns a formal triage class (all fields OK / incomplete / formal disqualification). The broker receives a sorted list, NOT a rejection recommendation. The final decision stays human – this is DPA Art. 21 and AIA compliant.

Lease generation. Landlord and new tenant are set – the broker generates the lease from the SVIT model template, with object data, rent, ancillary cost, security deposit. The management lawyer reviews, the landlord signs. Time: 8 to 15 minutes instead of 45 to 90.

Owner reporting. In property management, a pipeline produces a quarterly owner report: tenant status, repairs, ancillary cost forecast, open items. AI structures raw data from ImmoTop/Garaio into readable form, the manager reviews and sends. That is 80 percent admin time spent manually today.

Further use-cases with less leverage: AI-assisted client-inquiry replies (see ai-mandantenanfragen – analogue to fiduciary), voice bot for first-contact appointments (voice-agent-telefon), automated market-value indication as an internal quick check. Legally binding valuations remain reserved for certified appraisers.

6 steps to a clean listing and pre-qualification setup

1. 01Audit: inventory CRM (ImmoTop2/Garaio/REM2/RIMO) and portal interfaces (Homegate, ImmoScout24, Newhome), clarify AML and revDPA situation, document the applicant process.
2. 02Build the listing generator: map CRM data to objects, tonality template in DE/FR/IT/EN, SEO module for location terms. Hand-check the first 10 listings.
3. 03OCR pipeline for applicant PDFs: pay statement, debt collection extract, ID copy read into structured fields. EU/CH residency, clear deletion deadlines (30 days post contract).
4. 04Define triage logic: mandatory fields OK / incomplete / formal disqualification – no accept/reject recommendation. Clearly mark the classification as machine-made.
5. 05Release workflow: all AI outputs (listings, pre-qualification lists, owner reports) go to a broker or manager first for review. No publication and no decision without a human click.
6. 06Audit trail: every AI action in a log with timestamp, input-data hash, model version, user. Minimum retention 5 years (AML retention) for AML-relevant operations.

When deployment is justified

Audit-pilot-managed as usual, with two industry-specific points.

The audit (ai-readiness-audit) additionally checks the AML situation (do you have cash transactions > CHF 100,000? Are you brokering to clients in classified jurisdictions?) and the DPA situation (do you keep applicant profiles? Is AI co-used in tenant decisions?). Those drive classification and data-processing duties that must be cleared BEFORE piloting.

The pilot focuses on listings and owner reporting – neither decision-critical, both clearly measurable. Applicant pre-qualification follows in phase 2 once the audit trail and release logic are in place.

AI pays off when (a) the office posts at least 8 to 12 listings per month OR a manager handles at least 200 units; (b) the CRM data is clean or ready to be cleaned; (c) leadership has understood that AI is a tool and does not replace decisions on tenants.

Where real-estate AI has clear limits

Four red lines – all stemming from DPA, AML and EU AI Act.

First: fully automated applicant decisions. An AI score that decides on acceptance or rejection of a rental candidate without human review is a high-risk system (EU AI Act Annex III) and under revDPA Art. 21 (automated individual decisions) subject to information and objection rights. Practical workaround: AI sorts and prioritises, the landlord or manager makes the decision and documents it.

Second: legally binding market valuations (bank lending, inheritance, divorce). These count as expert opinions and require accreditation (SVIT/SVE). AI may deliver internal sanity values – legally binding valuation stays with qualified experts.

Third: AI in AML identification without documented process. Whoever runs customer identification through an AI tool (ID OCR, sanctions list match, PEP check) needs the risk documentation required in the 2024 FINMA refinement. See ai-gwg-kyc-screening and finma-awareness.

Fourth: applicant data in an uncontrolled cloud model. Pay statements, debt-collection extracts and ID copies are sensitive personal data. No free ChatGPT web, no US endpoint without enterprise contract. EU region or on-prem, with data-processing agreement.

Trade-offs

FAQ

Related topics

Sources

1. SVIT Schweiz – Branchenbericht 2024, Mitgliederstatistik · 2024-12
2. BFS – Leerwohnungs-Zählung Schweiz 2025 · 2025-09
3. FINMA – Rundschreiben 2024 zur GwG-Sorgfaltspflicht bei Maklerbüros · 2024-06
4. EDÖB – Leitfaden automatisierte Einzelentscheidungen nach Art. 21 revDSG · 2025-08
5. Europäische Kommission – EU AI Act, Annex III (Hochrisiko Personenscoring) · 2024-07