Real Estate Intelligence Cloud
Over the past years, real estate developers experienced a sort of “credit crunch” given the fear that banks will face again the default rates that characterised the sector after the Lehman crisis. Lending wise, there has been a decrease in the I-type error (banks stopped lending money to high-yield borrowers), but a considerable increase in the II-type error (banks stopped lending money to investment-grade borrowers with profitable opportunities).
The platform aims to correct the II-type error when granting mortgages to real estate developers. By analysing tenths of data points related to a geographical area and others related to the credit risk of the real estate developer, the AI-system is able to output the forward-looking price of a planned real estate development and the adjusted creditworthiness of the borrower. By performing the “matchmaking” between the borrower and the asset, the platform suggests adjusted pricing and LTV (Loan To Value) for the specific deal.
In the first step of our journey into the product, we focused on the vision. This is where we got to know the problem head-to-toe and created a comprehensive brief for the project. It allowed us to gather all the research needed from the different stakeholders at the table. This prevented us from being blocked by a lack of information during the following phases.
After an immersive session with all the stakeholders, we had all the information we needed to create the ultimate list of User Stories and features necessary to prove the main assumptions in a Proof of Concept (POC). Our experts in Product, UX/UI and Tech Architecture focused on finding the answers to all of the questions raised throughout the scope preparation. On the tech side we elaborated a memorandum tackling our suggested approaches for tech stack selection, database technology and orchestration. Together with Bocconi University and AlixPartners we started defining the useful datasets and the algorithm behind the lending decisions.
We developed a non-functional prototype to simulate user interaction. The experience of using the clickable prototype was very much like the final product itself, this was the adequate phase to test the information architecture, the UX and most importantly to present all involved counterparts a concrete outcome.
With the decisions closed in the previous steps, we went on developing the tool through agile cycles. Meanwhile, the data team, with knowledge on several datasets, started testing the model on the city of Milan (object of more than 10b EUR investment plan in the next decade) and fine-tuned the parameters.
For each of the developers, besides the regular scoring tools, we calculated the punctuality in delivering the past project (cadastral data stored in the municipality archive). We also built a Legal scoring because there’s a common practice among real estate developers (general contractors) to win projects by suggesting a budget “at cost”, then subcontracting the project to low-quality contractors that punctually fail to deliver what agreed, and ultimately issue a lawsuit against those contractors because of the failure. We monitor the number of given lawsuits (by the general contractors) in order to assess their propensity to “play the fair game”.
As a result of the massive data gathering about direct and indirect indicators that can potentially affect the volatility and trend of the real estate asset in time, we derive the current and future market value.
The suggested LTV and spread overrides are a direct function of the quality of the matchmaking as a result of the combination of the debtor, the asset future value and the product conditions.