Skip to content
homePad
Back to the blog

Innovation14 min read

AI in property management: practical uses, limits and safeguards

AI in property management: what it really does on an inspection report or a portfolio, where to start, how to check its output, GDPR rules.

The homePad team

AI in property management is used today for four practical tasks: recognising what a photo shows, transcribing what you dictate, drafting a first version and suggesting an order of magnitude for a cost. It does not visit the home and it signs nothing: it prepares a draft that the professional corrects and approves. Properly supervised, it cuts data entry and makes reports more consistent; poorly supervised, it produces findings that nobody has reread.

Artificial intelligence (AI) in property management: software that analyses photos, voice or text to suggest a description, a summary or a cost estimate, using models trained on large volumes of examples.

On-device AI: AI whose calculations run on the phone or tablet itself, without depending on a remote server for the analysis.

What does AI in property management actually do?

AI comes in mainly during the visit and just after it, where data entry takes the most time. Here is each use, what you check and the most common mistake in the field.

UseWhat the AI suggestsWhat you checkTypical mistake
Detection from photosThe elements in the room and their visible defectsOmissions, duplicates, condition givenA reflection on varnished parquet read as a scratch
Voice dictationThe text of the observation, while you lookTechnical terms, location of the defect"Ceiling" transcribed as "sealing" in a noisy stairwell
Voice commandsChanging room, adding an element without touching the screenThat the action carried out is the right oneAn element added to the wrong room
Works reportThe list of works based on the defects recordedQuantities, surface areas, responsibilityA paint touch-up counted over the whole wall instead of one section
Report summaryAn overview an owner can readThat nothing important has disappearedWater damage reduced to "marks on the ceiling"
Cost estimateAn order of magnitude per itemConsistency with your prices and your quotesAn estimate presented as an amount owed

What are the limits of AI during a visit?

AI knows neither the lease, nor the history of the home, nor what was agreed with the occupant: it describes what it sees or hears, nothing more. Three limits keep coming up in the field.

  • What is out of frame does not exist for it. A chip at the bottom of a door or a blackened seal behind the washing machine is only detected if it is photographed.
  • It gets materials and light wrong. A shadow becomes a stain, wood-effect tiles become parquet, backlighting wipes out a crack.
  • It does not settle questions of law. Wear and tear is the normal wear of a material or a piece of equipment caused by time and use in line with the lease; it is not the tenant's responsibility. Deciding whether paint that has yellowed after nine years of occupation is wear and tear or damage remains a matter for the professional's judgement, with a wear-and-tear grid or a table of useful lives if needed.

Wear and tear is the state of wear or deterioration resulting from time or from normal use of the home's materials and equipment (decree no. 2016-382 of 30 March 2016); the lease may provide for a wear-and-tear grid. Any deduction from the deposit must be justified, and the deposit is returned within one month if the move-out matches the move-in, two months otherwise (art. 22 of law no. 89-462 of 6 July 1989). Disputes: the departmental conciliation commission (commission départementale de conciliation), then the court.

Where should an agency or a property management firm start with AI?

Start with a single use, the one whose result can be checked at a glance, then add the others one at a time. Switching everything on at once stops you from knowing what is actually useful.

  1. Voice dictation. The benefit is immediate and a mistake shows when you reread.
  2. Detection from photos. It requires a photo-taking instruction shared by the team: an overall view of each room from the door, then a close-up of each defect, in daylight.
  3. The summary and the works report. They are useful as soon as a report is passed on to an owner or a contractor.
  4. The cost estimate. It comes last, because it depends on the quality of everything that comes before.

Test each step with two or three volunteers from your staff, on a variety of properties: a furnished studio, a two-bedroom flat occupied for ten years, a flat with a cellar and a parking space. For the visit method itself, see our guide to the digital inspection report.

How do you organise human approval?

Nothing goes out to a tenant, an owner or a contractor without being reread by the person who carried out the visit. The report commits the company that signs it, whatever tool prepared it. Write down what the review must cover:

  • each room contains all its elements, with nothing missing and nothing duplicated
  • the condition given to each element follows the team's shared scale
  • each photo properly illustrates the observation it is attached to
  • the quantities and surface areas in the works report are plausible
  • no sentence assigns a responsibility to the occupant without an established fact
  • the cost estimate never appears as a sum owed

A finding generated and then signed without being reread is hard to defend before a conciliation body or a court.

Which rules apply: GDPR, Swiss law and the EU AI Act?

Photos of an occupied home, a member of staff's voice and a tenant's name are personal data: their processing by an AI follows the same rules as any other processing. In France, Belgium and Luxembourg, the GDPR and the EU AI Act apply; Switzerland applies neither, but its data protection act applies directly to AI. In practice, everywhere:

  • collect only what is necessary: no faces, mail or personal belongings in the frame;
  • sign a data processing agreement with the tool's vendor;
  • enter the processing in your record of processing activities;
  • assess the risks before using any tool that decides on its own, for example a tool that automatically scores prospective tenants.

GDPR: data minimisation (art. 5), processors (art. 28), record of processing activities (art. 30), data protection impact assessment where the risk is high (art. 35); art. 22 governs decisions based solely on automated processing. Regulation (EU) 2024/1689 on AI, in force since 1 August 2024, was amended by regulation (EU) 2026/1744 (the "Digital Omnibus on AI", OJ of 24 July 2026): the obligations for the high-risk systems in Annex III apply from 2 December 2027. Since this reform, article 4 requires companies that use AI to take measures to promote AI literacy among their staff, without a set level. A tool that helps draft inspection reports does not, on the face of it, have the profile of a high-risk system; assessing the creditworthiness of natural persons is listed in Annex III (point 5(b)), so a tool that scores applicants deserves a specific analysis. Data protection authority: the CNIL.

Before choosing a tool, put four questions to its vendor: where the calculations take place, on the device or on a server, and in which country; whether your data is used to train its models and how to object to that; how long it is kept; which contract governs the service. Our GDPR checklist for real estate covers the rest.

Which mistakes should you avoid with AI?

The most costly mistakes come from how the work is organised, rarely from the tool itself.

Confusing an estimate with a quote. A deduction from the deposit is justified by a quote, an invoice or a costed finding (rules for each country in the block on wear and tear and deductions). An automatic estimate prepares a decision; it does not justify it.

Neglecting how photos are taken. A dark or blurred photo gives poor detection and a longer review.

Letting the vocabulary drift. If everyone approves their own wordings, reports stop being comparable from one visit to the next and the portfolio's data loses its value.

Measuring nothing. Record before and after the rollout: data entry time per report, number of corrections at the review stage, time taken to send the signed report, disputes received. Your own figures are worth more than a vendor's promises.

Aiming too far, too soon. Predictive maintenance or the digital twin require years of structured data. Today's reports are that data, including for preparing repairs across the portfolio.

What homePad does for you

homePad, used every day by more than 10,000 professionals, builds in AI on one constant principle: AI suggests, the professional approves.

  • homePad App: detection of elements and defects from photos, voice dictation and voice commands, with on-device AI.
  • Works report generated by the AI, report summaries and cost estimates, to be reread before anything is sent.
  • homePad Reports: AI move-in / move-out comparison report, from the Pro plan.
  • homePad One: dashboards, KPIs, digital twin and export of your portfolio's data.
  • For property managers and real estate investment companies: AI reports included from the Pro plan.

Frequently asked questions

Can AI do an inspection report instead of an agent?

No. AI prepares the description from the photos and the dictation, but the inspection report is still drawn up in the presence of the parties or their representatives, who reread it and sign it. It is the professional present on site who is accountable for it.

Is AI reliable for detecting defects on photos?

It detects well what is sharp, well lit and in the frame. It gets reflections, shadows and imitation materials wrong, and ignores anything that has not been photographed. A shared photo-taking instruction and a systematic review make up for these limits.

Is an AI cost estimate enough to make a deduction from the deposit?

No. In all four countries, a deduction must be justified, in practice by a quote or an invoice, or accepted by the tenant (details for each country in the block on wear and tear and deductions). The estimate helps you decide quickly which quotes to ask for.

Is AI in property management GDPR-compliant?

Yes, if the processing is governed like any other: a defined purpose, minimised data, a data processing agreement with the vendor, an entry in the record of processing activities. In Switzerland, the same logic applies under the Federal Act on Data Protection. Ask where the data is processed and whether it is used to train models.

How should a real estate agency start using AI?

With voice dictation, whose mistakes show immediately when you reread. Then add detection from photos with a photo-taking instruction, then the summaries, and the cost estimate last. Measure data entry time and corrections before and after each step.

  • AI in property management
  • AI inspection report
  • photo defect detection
  • EU AI Act

Ready to digitise your property management?

Book a personalised 30-minute demo with our team.

Talk to an advisor

Personal data: see our privacy policy (in French).

How many pages are viewed and how visitors move through the site. Measured with Google Analytics, for our own use.