A brokerage that ships its own software.
Behind every broker sits a proprietary data, match and underwriting layer – built in-house, shipped weekly, used on every mandate. This is the operating system of the brokerage.
How the platform is built.
Data layer
A continuous ingestion pipeline that normalises public listings from ImmoScout24, Kleinanzeigen and ImmoWelt together with planning notices and macro indicators into one queryable dataset.
- ·Continuous ingestion from the leading German marketplaces.
- ·Deduplicated listings across portals into single assets.
- ·Time-stamped – comparables hold their history.
Match & underwriting platform
In-house scoring stack matching tenant briefs to assets, ranking micro-locations and pulling live comparables under every term. No vendor middleware.
- ·Micro-location scoring across catchment, footfall, competition.
- ·Yield, ERV and term-sheet benchmarks pulled live.
- ·Every match traceable to the underlying source.
Ship every week
Engineers, product and brokers sit at the same table. Friction surfaces in the Monday call and ships to production the same week – no quarterly vendor roadmap dependency.
- ·Continuous delivery, no maintenance windows.
- ·Brokers shape the backlog directly.
- ·Versioned changelog visible to clients on request.
What flows into the data layer.
We do not resell third-party data. Sources are used to inform our own analysis and are cited openly when surfaced to clients.
From raw data to broker-signed output.
Ingestion
Sources are polled continuously, raw records stored immutably, every field carries a source URL and a timestamp.
Normalisation
Addresses and assets are resolved against a canonical reference; conflicts are flagged for broker review, not silently merged.
Modelling
A mix of in-house scoring models and external LLMs for unstructured text. LLM outputs are always grounded in retrieved evidence – never used to invent comparables.
Broker review
Nothing leaves the platform for a client without a broker signing off. The model accelerates the broker – it does not replace them.
In-house scoring models
Asset-class-specific models for micro-location, yield bandwidth, tenant fit and rollout sequencing. Trained on the proprietary data layer; calibrated against deals our brokers actually closed.
External LLMs – grounded only
Frontier LLMs (Gemini, GPT, Claude) are used for unstructured text, briefing parsing and report drafting – always retrieval-grounded against our data layer. We do not let a model invent a comparable.
Mandate data stays mandate data.
- 01Hosted in the EU; primary data residency Frankfurt.
- 02Encryption at rest and in transit; role-based access for the team.
- 03Client mandate data isolated per workspace; no cross-mandate training.
- 04No client data is sent to public LLM training endpoints.
- 05GDPR-aligned processing; DPA available on request.
Faster shortlists. Live comparables under every term. A broker who is not the bottleneck – because the platform is doing the work the analyst used to do.