AI in Swiss property management: less rental admin

Switzerland’s dwelling vacancy rate fell to 0.93% on 1 June 2026, from 1.00% a year earlier. The Federal Statistical Office counted 34,690 vacant dwellings offered for rent, 6.7% fewer than in 2025. The 0.93% measure covers the overall housing stock, including properties offered for sale; it is not a rental-only vacancy rate. ¹

For an agency handling strong demand, finding more applicants may be less urgent than managing the applicants it already has. A full inbox does not tell you which applications are complete, which viewing slots remain available or which cases need a property manager’s attention.

The priorities will differ across a portfolio. A popular urban apartment may create an intake peak. A harder-to-let property may need faster follow-up and more flexible viewings. Choose the workflow using your own listing and workload data.

The strongest starting points are repetitive enquiries, viewing coordination, missing-document follow-up and copying approved information between systems. Choose the task that repeatedly interrupts your team. Automating it should remove work from the day, not create another inbox to supervise.

These steps combine process automation and AI agents. Workflow automation follows defined rules: when a viewing is booked, reserve the slot and send a confirmation. AI helps interpret variable material: a free-text question, an uploaded document or an explanation from an applicant.

Use a calculation for arithmetic. Use AI where language or document variation makes a fixed rule insufficient. Give both clear limits on what they can change or send.

You open an application. One required document is unreadable. You find the applicant’s email, ask for another copy and move on. When it arrives, you need to find the case again, save the replacement and check what is still missing.

Begin with staff approving the follow-ups. Once the team has checked their accuracy on representative cases, routine messages can be considered for automatic sending under agreed rules. Set a reminder limit and stop messages when an applicant withdraws or the case closes.

The applicant should receive a specific request, not another generic checklist. The property manager should see the current case, not reconstruct it from several email threads.

That is a useful pilot outcome: less chasing, fewer duplicate files and a clearer review queue.

The workflow keeps each follow-up attached to the application in six steps:

1. Associate the submission with the correct apartment and application.
2. Check the items required at this stage against the files received.
3. Flag an unreadable file or missing page without guessing its contents.
4. Draft a precise follow-up: which item needs replacing and how to submit it securely.
5. Attach the replacement to the existing case and update its status.
6. Give the reviewer a summary of what is complete and what still needs attention.

AI can extract agreed fields, compare information across documents and prepare a summary for a property manager. It should show uncertainty and link back to the evidence. A summary is useful only when the reviewer can check it quickly. For each important value, show the original source and any unresolved uncertainty.

For example, if your agency uses a rent-to-income indicator, define the calculation explicitly. Which rent components count? Which household income is included? How is variable pay treated?

With monthly rent and charges of CHF 2,100 and documented gross household income of CHF 7,000, the ratio is 30%. That is arithmetic. Whether the household meets your agency’s approved criteria is a separate review.

An applicant starting a new job may have an employment contract but no payslips from that employer. A self-employed applicant may need a different evidence route. The system should surface these circumstances and allow corrections, rather than turn missing standard documents into an automatic rejection.

Keep tenant selection and rejection with an accountable person who can inspect the evidence and change the outcome. Do not use names, photographs, accents or an AI-generated impression of “suitability” to rank people.

Likewise, reading a debt-enforcement extract is different from authenticating it. A model can extract a date or flag a mismatch; that does not establish that the document is genuine. Any authenticity check needs a supported verification method and a route for cases that cannot be verified.

Rental applications contain personal and financial information. Making it easier to process should also make it easier to control.

The Swiss Federal Data Protection and Information Commissioner says applicant data must be relevant and proportionate to selecting a tenant, collected for a recognisable purpose and kept only as long as needed. Its guidance says unsuccessful application forms should be destroyed once a lease is signed, unless a specific justification exists, such as an agreed waiting list. ²

Before building the workflow, agree which information is needed at initial application, which evidence is justified later, who may access it and how closure triggers deletion. Check where copies remain, including inboxes, document storage and connected services.

Ask any AI provider where documents are processed, which subcontractors receive them, whether they are used for model training and how deletion works. Review the actual data flow and contracts. A Swiss hosting label or a human approval button alone does not establish compliance.

Give applicants a clear explanation of the process and a reachable person when something is wrong.

You may not need new software. First check what your existing rental platform and property-management system can already do. Forms, viewing scheduling, message templates and application tracking may already be available in your subscription.

Then identify the gap. Perhaps the application platform works well, but your team still copies approved details into the property system. Perhaps enquiries are centralised, but staff still read every attachment to find missing information.

That gap is the integration brief. Confirm that your particular software version and contract provide the access needed. An interface advertised by a supplier does not establish that every required action is available in your account.

For a first pilot, a review queue or controlled export may be enough. If you later add direct updates, require duplicate prevention, visible failures and a record of what changed. Your team should be able to tell whether information reached the destination.

Build the case from your own handling time. Separate staff minutes spent on a case from calendar days spent waiting for an applicant.

Here is an illustrative calculation, not a measured SPS result or a forecast:

The calculation is 200 × 8 ÷ 60 × CHF 60. From that value, subtract recurring software, processing and support costs. Account separately for setup and integration costs.

Released capacity becomes cash savings only when it changes actual spending. It may instead let the same team handle more work, reduce overtime or spend more time with tenants and owners. Decide which outcome matters before approving the investment.

Choose a small set of properties and one repeated task. Measure the current process, then compare it with the pilot using similar cases.

Track staff minutes per application, time until a dossier is ready for review, incorrect follow-ups, document-reading corrections and cases needing manual intervention. Include difficult examples: poor scans, multiple household members, different languages, withdrawals and replacement documents.

Agree the conditions for expansion in advance. The workflow should save time after corrections, preserve the evidence reviewers need and give staff a workable fallback when a connection fails.

Once that works, extend the process to the next bottleneck. There is no need to automate the whole rental lifecycle in one project.

What is the difference between an AI agent and workflow automation?

A workflow follows predefined steps, such as confirming a booked viewing. An AI agent can interpret a request and choose among permitted actions, such as checking approved property information before drafting a reply. Use clear rules for deadlines and calculations, and human review for consequential decisions.

Can a pilot work with our existing property system?

Potentially. First confirm the access your software version and contract provide. Start with an approved export or a review queue if direct updates are unavailable. The pilot should make failed transfers visible and prevent duplicates before you expand it.

Can we start without sharing real applicant documents?

Yes. You can map the process and test the initial workflow with synthetic examples or properly anonymised documents. Before using real applications, agree the permitted data, access, providers and retention rules. Synthetic tests alone do not establish performance on real dossiers.

Book a 30-minute operations review with Swiss Product Studio. Bring one recurring rental task, the tools involved and a synthetic or properly anonymised example. Do not send applicant identity documents or salary files to book the call.

The aim of an initial consultation is a rental workflow opportunity map: where time is spent, what existing software can cover, what needs connecting and which bounded pilot is worth testing. That gives you a concrete basis for deciding whether to proceed.

Book a 30-minute operations review.

Not sure which task to choose? Use our fill-in guide, Your First Automation Plan, to identify a recurring process and map out your first automation.

1. Federal Statistical Office (FSO). Dwelling vacancy rate falls below 1% in 2026, 10 September 2026. Figures refer to 1 June 2026. Official federal release covering the overall vacancy rate and vacant dwellings offered for rent. ↑1
2. Federal Data Protection and Information Commissioner (FDPIC). Data collection when renting accommodation, accessed 15 October 2026. Guidance on purpose, proportionality and retention of prospective-tenant information. ↑1