IMMO: underwriting from days to hours, sourcing across 3 countries
IMMO is a pan-European residential property investor. Sourcing depended on brokers, and underwriting meant broker calls, site visits and days of data gathering. As Senior Product Manager, I launched an AI-assisted sourcing product across Germany, Spain and the UK and led product for an underwriting workflow that took the first pass from about 3 days to 2 hours.
- My role
- Senior Product Manager
- Company
- IMMO
- Timeline
- Feb 2023 – Feb 2024
- Markets
- Germany · Spain · UK
At a glance
The challenge
Opportunities arrived when a broker called, and every deal needed calls, site visits and days of data gathering before an analyst could judge it.
What I led
An AI-assisted sourcing product across 3 countries, GPT-3 property-data extraction, and an underwriting workflow driven by property descriptions and images.
Result
Underwriting from about 3 days to 2 hours. 98% less data-ingestion time. 1 to 5 data sources in 3 months. Sourcing efficiency +30%.
Context
For a residential investor, two steps decide how many good deals you can do: finding the right assets, and judging them quickly enough to act.
At IMMO, both were manual. Sourcing depended on brokers and relationships, so opportunities arrived when someone called. Underwriting meant broker calls, site visits and days of data gathering before an analyst could apply judgment.
My role
As Senior Product Manager, I owned the AI-assisted sourcing product and led product for the underwriting workflow. I also owned an AI document-processing product and orchestrated the IMMOCore platform migration. I led this work from Chennai.
Sourcing: from broker calls to data
I launched an AI-assisted asset-sourcing product and scaled it from 1 to 5 data sources across Germany, Spain and the UK in 3 months.
In 2023, I integrated OpenAI GPT-3 into property-data extraction, improving sourcing efficiency by 30%.
Underwriting: from days to hours
Simplified view of the underwriting first pass
I led the effort to move underwriting from manual broker calls and site visits to automated analysis of property descriptions and images. The workflow auto-populates 300+ data points an underwriter needs, cut data-ingestion time by 98%, and took underwriting from about 3 days to 2 hours.
The point wasn't to remove judgment. It was to make sure only the right assets got a site visit.
Key decisions
1. Automate the first pass, keep the judgment
Analysis of descriptions and images handles the data gathering, and analysts and site visits focus on assets that clear the first pass.
Why it mattered: underwriting capacity goes to the deals worth doing.
2. An LLM for property-data extraction, in 2023
I integrated OpenAI GPT-3 into property-data extraction while LLMs were still new to production use. Sourcing efficiency improved by 30%.
Why it mattered: it tied an early LLM integration to a measurable business result.
3. Coverage across three markets
The sourcing product reached 5 data sources across Germany, Spain and the UK within 3 months.
Why it mattered: a sourcing product is only as useful as the share of the market it can see.
Also shipped
- AI document processing: 100% user adoption across 5 departments.
- IMMOCore platform migration: a 30% operational efficiency gain.
How I measured it
Each metric answers one question: did we see more of the market, and did we judge it faster?
| Metric | Before | After |
|---|---|---|
| Underwriting time | about 3 days | 2 hours |
| Data points auto-populated | — | 300+ |
| Data-ingestion time | — | −98% |
| Data sources | 1 | 5 in 3 months |
| Sourcing efficiency (GPT-3) | — | +30% |
| Document AI adoption | — | 100% across 5 departments |
What I'd do now
Lessons
Automate the gathering, not the judgment. The biggest time saving came from removing data collection, not from replacing the analyst.
Distance from the market isn't a barrier. I led products for three European markets while based in Chennai.
Automating a judgment-heavy workflow?
I've taken underwriting from days to hours without taking the analyst out of the loop. Happy to compare notes.