AI in property management: three workflows for when capacity runs short

Starting point: the bottleneck is processing, not the market

The German association of property managers published its industry barometer in September 2025. 70% of property management firms report staff overload, a third of them severely. The reason: vacancies stay unfilled. The consequence appears in the same survey: 57% part with unprofitable mandates, 63% with time-consuming properties, and almost 14% no longer take on any new clients at all, rising above 20% among the smallest firms.¹ A market in which owner associations are left without a manager does not have a demand problem. It has a processing problem.

The rest of Europe knows the same work under different names. In France, the syndics of a single association alone manage some 4.5 million lots, meaning individual flats, cellars and garages (as of 2023). 57% of its members work as administrateur de biens or as syndic de copropriete.² In Spain, administradores de fincas face a stock of 25.7 million dwellings. 42.7% of them were built before 1980, and the annual renovation rate is between 0.11 and 0.15%.³ In Italy the management office is itself a micro business: 94.8% of all Italian enterprises have at most nine employees, and management offices typically sit below that.¹⁰ At the same time, the condominium has for years received the invoices of its tradespeople and suppliers electronically through the Sistema di Interscambio (SdI), the state exchange system for electronic invoices. Assigning them to the individual owner associations is the management’s job. In Sweden, 100 public housing companies worked together in the digitalisation programme of the Allmännytta. They cover around 65% of the stock and more than a million tenants.⁴ And in Norway, borettslag and sameier are a widespread form of management made almost entirely of documents: statutes, house rules, minutes of general meetings and board meetings, maintenance plans, correspondence with the shareholders.

Two things are the same everywhere. First, the work hangs on text: resolutions, minutes, house rules, maintenance and insurance contracts, tradespeople’s quotes, years of correspondence. Second, the small offices doing that work reach AI less easily than the large ones. In the EU, 19.9% of enterprises with ten or more employees use at least one AI technology. Among those with 10 to 49 employees it is 17.0%, and from 250 employees upwards 55.0%.⁵ The gap between small and large is wider in every country than the gap between countries.

Use case 1: answering enquiries from the resolution and contract record

An owner wants to replace her windows and asks whether she may and who pays. A tenant reports damp in the cellar storage and asks whether the repair has now been approved. A board member wants to know what the maintenance contract for the lift covers. Three enquiries, three times the same preparatory work: looking up what applies.

The AI colleague has those documents as a knowledge base: the declaration of division, the deed that divides a building into individual units, or the equivalent house regulations, the house rules, the collection of resolutions from recent years, maintenance and insurance contracts, past correspondence about the property. It produces a draft reply and names the source with it: the meeting, the date and the agenda item, including the wording. You read, trim and approve. The valuable part is not the finished text but the source reference underneath it.

The line has to be drawn sharply. The assistant finds the passage and summarises it, it does not interpret it. Whether a declaration of division assigns the window front to individual ownership, meaning the part that belongs to a single owner alone, is the application of law, and if there is a dispute the specialist lawyer belongs at the table. No draft goes out without approval. Permissions apply down to individual tools: whoever may not see the files of another property does not see them through the AI either.

Use case 2: preparing the meeting from your own history of minutes

Eigentümerversammlung, assemblee generale, assemblea condominiale, junta de propietarios, generalforsamling: the meeting has a different name everywhere and means the same thing everywhere. Beforehand the agenda, the draft resolutions, the documents. Afterwards the minutes and the follow-up. In between a year in which nobody quite remembers what was deferred under item seven in the minutes before last.

The assistant reads the property’s series of minutes and compiles what is still open: deferred items, resolutions subject to conditions, works commissioned but not completed, recurring motions from the owners. Out of that comes a draft agenda, together with draft resolutions in the wording of comparable earlier ones. After the meeting, your notes become the first version of the minutes, in the structure your office uses anyway.

In Germany a change of format comes on top. Owners can resolve, with at least three quarters of the votes cast, to hold a purely virtual meeting for at most three years. If such a resolution is passed before 1 January 2028, a meeting in person must be held at least once a year up to and including 2028, unless this is waived unanimously.⁶ So virtual meetings create more written preparation and follow-up, not less.

And this is where the sharpest limit in this article sits. A sloppily worded resolution can be challenged, and nothing about that changes if the first draft came out of a machine. The assistant supplies a template from your own records; it is worded, checked and owned by the management, and nothing is sent on its own.

Use case 3: explaining statements and renovations in the language of the owners

Every statement of costs produces follow-up questions, and those questions look much the same year after year. In Germany, since 2023 the allocation of carbon dioxide costs between landlord and tenant along a tiered model has been added: the more carbon dioxide a building emits per square metre, the larger the landlord’s share. On top of that comes a duty: where remotely readable equipment is installed, users have to be given monthly consumption information, compared with the previous month, the same month of the previous year and a standardised average user.⁷ That is exactly the kind of calculation rule that has to be explained per property and per year: to the owners, to the tenants and to the board.

The assistant writes that explanation once from your own documents and then again and again: as an annex to the statement, as a reply to the follow-up question, as a section in the circular. The same pattern carries the second big topic. In Spain the ministry is aiming for 1.2 million renovated dwellings by 2030, while at the current pace around 285,000 can be expected.³ A renovation resolution rarely fails for lack of will and often because nobody has put the funding conditions, the state of the building and the cost consequences side by side. That is precisely the compilation the assistant can prepare, provided the management has stored the funding conditions itself.

Because owner associations on the Spanish coast and in French holiday regions are internationally mixed, the same explanation is produced in several languages on request. Two limits go with that. The calculation itself is done by the accounting software or the metering service; the AI explains the result, it does not recalculate it. And the legally authoritative version remains the original language, with the translation being a service that carries this note on the document. The assistant gives no assurance about eligibility for funding.

Connectors: the benefit starts before the interface

Now the uncomfortable truth: management software in this sector is largely closed across Europe. For the widely used German systems Immoware24, DOMUS and objego, no public developer documentation could be found in August 2026. casavi is the exception and publishes the documentation of its interface without a login, which is why the product appears in the first row of the table. The Spanish fincas programmes Gesfincas and NetFincas name neither an API nor a web service on their sites. Both are described routes along which another program could fetch data. For the French product group there is no verified finding, which is why it does not appear in a single row of this table.

That matters less than it sounds. The three workflows above do not depend on the specialist software. Resolutions, minutes, contracts and correspondence exist as files anyway, and the knowledge base needs no interface for them. The connection itself runs through MCP, an open standard for linking external systems. Where an AI connection of that kind is genuinely documented, it is listed here, checked in August 2026:

If the software has no interface, the route runs via exports, and here that is no makeshift. Every management system hands out property and unit lists as a table, statements and dunning runs as PDF. Those outputs go into the knowledge base together with the document store. The difference from a real connection should be named honestly: you get no live status, but a status with a date on it. For the resolution record, contracts and history that is no loss, because those documents do not change by the hour. For open items and payment status it is one. There the route runs via the connected accounting system.

How a connection via MCP works in principle, and which systems are prepared, is described on the integrations page.

Data protection: owner and tenant data stays in Europe

A management office holds names, addresses, payment behaviour, arrears and occasionally details about accessibility needs. With data like that, the question of where it is stored is not theoretical. Kisuno runs on your own server or managed in data centres with ISO 27001 certification; the certification belongs to the data centre operator. The models from OpenAI (ChatGPT) and Anthropic (Claude) run in the AWS data centre in Frankfurt. AWS is a US company, so the US CLOUD Act applies even with servers in Frankfurt, and we say so openly instead of leaving it out. For higher data-protection requirements, open frontier models such as GLM, Kimi and DeepSeek are available, operated by European companies on servers in Europe; the CLOUD Act does not apply there. Whatever model you choose: inputs and uploaded data are neither stored by the model provider nor used for training. The data processing agreement and the subprocessor list are published openly on the website.

The advantage should not be made bigger than it is. If personal data is transferred to a recipient in a third country, that is, outside the European Economic Area, Chapter V of the General Data Protection Regulation applies. It calls for its own legal basis and an assessment of the legal situation in the destination country. If storage and processing stay within the European Economic Area, that assessment step falls away and what remains is ordinary processing on instructions under Article 28. European processing does not make the processing lawful automatically. It makes it easier to justify.

On top of that comes a point that is rarely spelled out. In Norway, 71.0% of the enterprises that use AI use commercial off-the-shelf software, in Sweden 62.1%.⁵ Such tools are usable, but they do not know the business: not the resolutions, not the contracts, not the history of minutes. The difference lies not in the model, but in what it is allowed to reach and where it does its computing.

Finally, on the AI Act: an internal assistant that writes drafts and searches your own documents is as a rule minimal risk. This sector needs to pay attention elsewhere: an AI that assesses the creditworthiness of natural persons counts as a high-risk system under Annex III of the AI Act. Anyone pre-sorting prospective tenants by machine also has to observe equal treatment law. Kisuno supports the administrative work and does not select prospective tenants. This article is not legal advice.

Frequently asked questions from property management

Sources

All figures in this article come from the following sources, last checked in August 2026. National surveys are marked with the country they belong to.

1. VDIV Deutschland, industry barometer 2025, press release of 12 September 2025 (Germany): vdiv.de
2. FNAIM, Chiffres cles, lots under management and the professions of its members (France): fnaim.fr
3. Consejo General de Colegios de Administradores de Fincas de España, report on the housing stock and energy renovation, presented on 8 April 2025 (Spain): cgcafe.org
4. Sveriges Allmännytta, Digitalisering i allmännyttan, programme 2020 to 2022 (Sweden): sverigesallmannytta.se
5. Eurostat, dataset isoc_eb_ai, artificial intelligence by size class of enterprise, survey year 2025: ec.europa.eu
6. German Condominium Act (WEG), Section 23(1a) and Section 48(6), virtual owners meetings (Germany): gesetze-im-internet.de
7. German Carbon Dioxide Cost Allocation Act, in force since 1 January 2023 (Germany): gesetze-im-internet.de. Heating Costs Ordinance Section 6a on billing and consumption information
8. Danea, Guida alle API per Domustudio Cloud, APIKey and the Cloud Pro requirement (Italy): danea.it
9. Vitec Fastighet, overview of interfaces and integrations, access through the partner programme (Sweden): vitec-fastighet.com
10. ISTAT, Annuario Statistico Italiano 2025, chapter 14 on enterprises, size classes in the reference year 2023 (Italy): istat.it
11. casavi, documentation of the manager API, viewable without a login (Germany): api.mycasavi.com