Industry · Real estate & AVM

AI in real estate. From valuation to due diligence.

We build AI software for property investors, mortgage banks, valuers and housing associations. Automated valuation models, computer vision for maintenance condition, document AI for due diligence and climate risk modelling, built on BAG, Kadaster and your own portfolio data.

ApplicationAVM & valuation
ApplicationComputer vision
ApplicationDocument AI
ApplicationClimate risk

The Dutch real estate AI market in numbers.

~10,3M
Addressable BAG objects in the Netherlands
~8,1M
Homes in the housing stock (CBS, 2024)
~190K
Home transactions per year (NVM/CBS)
~1,0M
Non-residential properties (offices, retail, logistics) in BAG

Source: CBS, Kadaster, BAG, NVM annual figures 2024.

Standard AVMs don't solve everything.

The major AVM providers in the Netherlands (Brainbay, Calcasa, Altum AI) deliver solid generic valuations. For the average terraced house in a typical neighbourhood, their model comes close to the transaction price. The problem lies at the edges: unusual properties, office, retail and logistics real estate, leisure property, listed buildings, and anything with atypical cash flow or renovation profiles. Applying a uniform model there produces unexplained outliers that cannot be justified in an acceptance or acquisition process.

Mortgage lenders want their own model that fits their acceptance policy and is explainable to the AFM (the Dutch financial regulator) and under the AI Act. That means training on their own acceptance data, SHAP values per feature, and a human-override flow with an audit log. Property investors and REITs want an in-house valuation tool that reflects their specific methodology, energy roadmap and CSRD reporting, not a black-box output from an external vendor. Insurers need rebuild and contents values based on photo and BAG (Dutch Basic Registration of Addresses and Buildings) data, not a comparable-properties model that leans on market prices.

We don't replace Brainbay or Calcasa. They are data suppliers that can also feed our model. We build the layer around them: white-label AVMs, drone pipelines, due-diligence LLMs and tenant portals that fit seamlessly with the way you work. How that works technically is covered in more detail on our page about custom real estate software; this page focuses specifically on the AI layer on top.

Our usual clients include mid-market estate agents (NVM/MVA), housing associations with asset management and energy challenges, smaller REITs (NSI, Eurocommercial-type), investment funds, developers who want to speed up feasibility studies, and valuation firms (RICS, NRVT) that want to scale their work without compromising on quality. For enterprise organisations with a broader AI programme, we sometimes also look at enterprise AI implementation as a wider framework.

AI software that fits every real estate segment.

Each type of property has different data, different risks and different models. We show how we typically work for each segment, from residential AVMs for banks to due-diligence platforms for REITs.

Residential valuation (AVM)

Hedonic models supplemented with gradient boosting (XGBoost, LightGBM) and computer vision on facade and aerial photos. We train on your own transaction history, supplemented with BAG, Kadaster (the Dutch land registry) and energy label data from RVO (the Netherlands Enterprise Agency). The result is an AVM with a measurable confidence score, explainable features and an audit trail you can defend to the AFM or an internal validation committee.

For consumer mortgage processes, we explicitly take the AI Act into account: this application may count as high-risk and requires documentation of training data, bias monitoring and the option of human review. We include an AI Act compliance package as standard.

For commercial real estate (offices, retail, logistics), alongside comparable transactions we also use cash flow modelling: tenant mix, indexation clauses, vacancy scenarios and break options from the lease. For REIT and fund clients, this results in an in-house valuation tool that, alongside external AVM data, also reproduces their own methodology. Asset managers use the same model for portfolio-wide revaluation, scenario analyses and CSRD reporting.

  • BAG and Kadaster integrationReal-time characteristics, ownership and transaction history per property.
  • Computer vision on facades and roofsMaintenance condition, solar panels and roof type from aerial photos and street view.
  • Energy label predictionRVO database + year of construction + photos for properties without a valid label.
  • AI Act explanation per valuationSHAP values per feature, audit log per model version.

AI touches every layer of your real estate portfolio.

Valuation is the best-known AI application in real estate, but by no means the only one. Our projects span the entire value chain, from acquisition screening to ongoing portfolio management.

Computer vision on drone, street view and satellite imagery provides insight into maintenance condition, roof condition, solar panel inventory and energy label indicators. We prefer to build these CV pipelines as part of a broader computer vision programme so the same models can be reused in multiple ways.

Document AI with LLMs extracts clauses, indexations, notice periods and break options from lease agreements, title deeds and valuation reports. In large due diligence projects, this is the difference between weeks of manual work and a structured file within a few sprints. For clients with a broader DD programme, we build a custom due diligence platform where this lands as a module.

Climate and risk modelling combines PostGIS data from the Klimaateffectatlas (Dutch Climate Impact Atlas), BAG construction year and façade data to determine flood, heat-stress and foundation risk at property level. For REITs and insurers, this is relevant for both CSRD reporting and underwriting policy.

Cash flow and demand forecasting based on transaction and tenant history helps with sale timing, vacancy projection and tenant churn. Smart-building AI combines IoT sensors with portfolio data for climate control, energy optimisation and occupancy analysis. Customer portals with retrieval-augmented generation give tenants and buyers instant answers to contract and service questions, with human escalation when a question falls outside scope.

Anomaly detection and tender/procurement AI are two applications we see more often with asset managers and commercial landlords: anomaly detection flags rent or vacancy outliers within a portfolio; tender matching links incoming office demand to the right properties in your portfolio, including lease terms and tenant-mix fit. Models like these run in production daily and need relatively little data once the feature set is well defined.

Built on the right real estate and AI regulation.

AVMs for consumer mortgages touch on both valuation standards and AI legislation. We work within this framework from the first sprint, not as an after-the-fact document.

EU AI Act

High-risk classification for consumer AVMs

An AVM that co-decides on a mortgage for a consumer falls under Annex III of the AI Act. We deliver risk classification, data governance, bias monitoring and a human-override flow as standard. The governance is not a separate PDF produced afterwards; it is built into model version management, the deployment flow and logging.

GDPR

Home occupancy is personal data

An address plus a resident is PII as soon as it is traceable. Models are trained on pseudonymised data, and production APIs log only what is needed for audit. A DPIA is delivered at the start of every project. For clients with their own DPO, we align with their template; for those without one, we provide a workable starting point that can answer any supervisory authority question straight away.

NRVT & RICS Red Book

Valuation guidelines

For valuation-support tools, we work within NRVT guidelines (the Dutch register of certified valuers) and, for international clients, the RICS Red Book. AVMs are an explicit subject there; we document the methodology according to their requirements.

Wft & DORA

Financial advice and digital resilience

For banks and insurers, the AVM system falls under the Wft (Dutch Financial Supervision Act) and the broader DORA requirements for digital operational resilience. We build in testability, monitoring and incident response from day one.

CSRD & EPBD

Energy and climate reporting

Real estate investors and REITs must report on their portfolio under the CSRD, and the EPBD sets requirements for building energy performance. Our platforms deliver the underlying data at property and portfolio level, including energy label prediction for properties without a valid label. That saves you hiring an external energy consultant for each individual property, while European regulatory requirements only get stricter.

Seamlessly connected to the real estate ecosystem.

We integrate with the data and software your organisation already works with, from public sources to commercial AVMs and valuation tools. Our AI complements; it doesn't replace.

BAG
Basic Registration of Addresses and Buildings
Kadaster
BRK, BGT, BRT
RVO
Energy label register
Brainbay
Dutch market-leading AVM
Calcasa
Consumer AVM
Altum AI
Real estate data APIs
PostGIS / ArcGIS
Geo-analysis
Klimaateffectatlas
Flood and heat risk

One abstraction layer above the data source.

We have standardised the integrations mentioned above across multiple real estate projects. For a new project, we set up the connector using the same abstraction layer, which means fewer bugs, faster turnaround, and your IT team can manage it themselves if we step out. No lock-in on a vendor-specific data contract.

For commercial AVM data (Brainbay, Calcasa, Altum AI) we do not handle subscriptions; those run directly between you and the supplier. We build the pipeline that combines their output with your own transaction history, photo data and portfolio context. That way, your IP position does not depend on a single supplier.

For geo-analyses we work with PostGIS, QGIS and, for larger portfolios, ArcGIS or deck.gl for visualisation as standard. For the model training pipeline we use Python (scikit-learn, XGBoost, LightGBM, PyTorch), MLflow for versioning and SHAP for explainability. International clients with portfolios outside the Netherlands also get a CoStar or HousingAnywhere integration, using the same abstraction approach so the Dutch stack does not have to change.

Not sure about a large project yet?

Test your idea first: a working prototype in 1 day

With OneDayBuild, we make your idea tangible in a single day for €1,150, so you know whether further development is worth the investment. Decide to go ahead with the full build? Then we deduct the cost in full.

View OneDayBuild →

From data check to production in clear steps.

A real estate AI project has its own rhythm: first inspect the data honestly, then train a narrowly scoped model, then scale up to production with monitoring. Five phases that are identical for every project.

01 · Data audit

What do you really have in-house?

Two days with your data team. Which transaction history, photo library and portfolio data is usable, what quality, and where the gaps are. Result: a feature inventory.

02 · Baseline

Comparable properties as a starting point

A simple hedonic model on your data. This is the baseline the ML model must beat later. It also gives a first indication of model explainability.

03 · ML iteration

Gradient boosting and CV features

XGBoost or LightGBM supplemented with computer vision on facade and aerial photos. Retraining each sprint, comparison with the baseline, and a bias test per neighbourhood type.

04 · Production

API + explainability

Model version in production with SHAP values per valuation, an audit log, and a human-override flow. Integration with your underwriting or acquisition system.

05 · Monitoring

Drift, fairness, retraining

Continuous monitoring of model drift, fairness across neighbourhoods and regions, and performance. Fixed monthly price for retraining and first-line support.

Work in the real estate AI domain.

Residential AVM · Mortgage lender

White-label AVM for consumer mortgages

Gradient boosting on transaction history supplemented with BAG and RVO data, with SHAP explanations per valuation and an AI Act compliance package. The model runs alongside, not instead of, Calcasa data and provides a second, explainable opinion for borderline cases in the underwriting flow.

a few tenths
of a percent better fit than baseline
3 regions
phased rollout
Computer vision · Housing association

Drone-based roof maintenance planning

CV pipeline on drone imagery for roof condition, moss/algae detection and a solar panel inventory across a housing association portfolio. The output lands in the existing asset management system so planners can immediately translate it into work packages.

one pilot neighbourhood
run before scaling
3 damage categories
labelled automatically
Due diligence · Asset manager

LLM platform for lease agreement analysis

Document AI for extracting indexation, termination and break clauses from mixed commercial lease agreements, with an audit log and human review flow. A DD team that used to spend weeks reading through documents manually now repeats the same work within a few sprints, with greater consistency.

ten clause types
extracted in structured form
audit trail per document
for compliance

Recognised in the real estate and proptech domain.

DNB Study · Mortgage market 2024

"AVM models are increasingly used for revaluation and portfolio monitoring, with explainability and bias control as the main points of attention for regulators."

PropertyEU Tech Briefing 2024

"Computer vision on aerial photographs and drone data is gaining ground in the Netherlands, particularly in housing association and asset management portfolios where manual inspection is too time-consuming."

Client survey, REIT client

"Appfront's AI team built our internal valuation platform on the same methodology our appraisers used manually — but across the whole portfolio instead of property by property."

Answers for real estate organisations considering AI.

The questions we hear most often from risk managers, asset managers, appraisers and proptech teams.

What exactly is an AVM?
An Automated Valuation Model is a statistical model that estimates the market value of a property based on comparable transactions, property characteristics (floor area, year of construction, type) and environmental factors. Classic AVMs use hedonic regression; modern variants use gradient boosting (XGBoost, LightGBM) and computer vision features from facade or aerial photographs. An AVM is not a replacement for an appraiser, but a tool for scalable valuation, portfolio-wide revaluation and continuous monitoring. In the Netherlands, most people know the concept through Calcasa valuations or the WOZ-style estimates on housing marketplaces; commercial applications are broader.
How accurate is AI property valuation compared with an appraiser?
For standard properties in standard neighbourhoods, a well-trained AVM comes close to an appraiser, usually within a few percent, comparable to the spread between two appraisers. For special properties (listed buildings, leisure, offices with an unusual tenant, retail in shrinking areas), the spread grows and an appraiser is indispensable. Our models give an explicit confidence score for each valuation, so you know when you can rely on the model and when you need a physical appraisal. For supervisory purposes, we also document the feature importance for each outcome so that a deviating valuation remains explainable.
Will AI replace my appraisers, or does it complement them?
It complements them. AVMs and CV pipelines scale valuation and inspection to volumes that are not feasible manually, such as portfolio-wide revaluation or a roof inventory across a thousand properties. For complex individual appraisals, legal reports and disputes, an NRVT/RICS appraiser remains in the lead. We usually see the appraisal team moving higher up the value chain: review, complex casework and advisory work. The NRVT and RICS Red Book explicitly permit AVMs as a supporting tool, provided the methodology and limitations are documented, which we include as standard.
Can you integrate with BAG, Kadaster and RVO?
Yes, that is a standard part of our stack. We work with BAG (property characteristics), Kadaster (BRK ownership and transaction history), BGT (geometry), BRT (topography) and the RVO energy label database. For commercial real estate AVM data, we also integrate with Brainbay, Calcasa or Altum AI. You arrange the subscription yourself, and we build the pipeline. For climate risk features, we also use the Climate Impact Atlas and, depending on the scope, flooding or subsidence data from the water authority or TNO.
What does the EU AI Act mean for our AVM?
An AVM that contributes to a decision on a consumer mortgage falls under Annex III of the AI Act and is classed as high-risk. That brings requirements around data governance, bias monitoring, technical documentation, human oversight and transparency towards the end user. For commercial real estate AVMs (B2B) the bar is lower, but the same principles are good practice. We include an AI Act compliance package as standard. See also our page on custom LLM integrations for the broader context.
Do you work with drone and satellite data?
Yes. For roof condition, solar panel inventory and maintenance planning, we use drone imagery combined with aerial photo archives (Cyclomedia, Beeldmateriaal Nederland) and, for large-scale applications, satellite data (Sentinel, Planet). For street and facade analysis, we use public street-view data and BAG integration. The CV pipeline is largely reusable between clients: training the detectors for roof tile condition, moss growth, solar panels and roof window types is a one-off job, and deploying it per portfolio is configuration.
Can we build a tenant or buyer portal with AI Q&A?
Yes. A customer portal with retrieval-augmented generation on your tenancy agreements, service documents and homeowners' association (VvE) documents is a common follow-up to an AVM project. We build it with the same compliance requirements as the valuation tools: audit log, human escalation, and a clear boundary around what the model may and may not make statements about. For housing associations with large tenant volumes, this is often the most visible AI product to the outside world, while AVM and CV do the heaviest work internally.
What determines the cost of a real estate AI project?
Mainly: the quality and volume of your own data, the number and complexity of integrations (BAG/Kadaster/RVO/Brainbay/your own acquisition system), explainability and compliance requirements (AI Act, NRVT), and whether we build an MVP AVM or a full platform with portal and monitoring. After the data audit, we give you a concrete estimate for the entire build, with no open-ended costs. We work in sprints; each sprint delivers something that works, so you can steer on scope rather than on budget.
Do you sell a product or build custom?
Custom. We are not an AVM vendor like Brainbay or Calcasa — we build for your organisation, on your data, with your methodology. That also means the model and the codebase are yours, not ours. For components we come across more often (BAG/Kadaster connectors, RVO pipeline, basic CV detectors) we do use reusable building blocks, but these are under an open licence so you do not have to depend on us for further development.

Ready to deploy AI on your real estate portfolio?

A half-hour introduction with an asset manager, risk officer or proptech lead. We listen, ask questions about your data and your use case, and give you an initial direction. No obligation.

Edit content