What are the challenges of the long and complex path to financing a property project?
Sébastien Perrigault: Obtaining a mortgage is one of the first steps. It can be a source of stress and delay, as well as dissatisfaction for our customers, who expect us to deal with their requests quickly. In order to obtain a mortgage, prospective buyers have to gather a file full of documents such as tax returns, pay slips, purchase agreements and bank statements. For our advisors, entering, reviewing and checking all these documents is time consuming and can lead to errors. It currently takes an average of 11 days to issue a digital offer.
What solution have you implemented to meet these challenges?
Sébastien Perrigault: In order to overcome the obstacles faced by our customers and advisors, since November 2024 we have implemented an artificial intelligence solution to streamline the entire process and shorten the loan approval time. By integrating IDP or Intelligent Document Processing (only for basic mortgage loans and excluding more complex ones, such as bridging loans or financing of SCIs - real estate partnerships), the AI will check the consistency between the information in the documents and that entered in the advisor's credit planning tool.
The solution reads the documents provided by the customer and extracts the information needed to verify the contents of their property finance file. For example, the AI verifies that the file accurately reflects the information in the purchase contract (address, purchase price, property tax date, etc.) and that the income information provided by the borrower matches the pay slip and tax statement. The AI also checks the authenticity of the tax notices.
Can you give us some technical details about the solution?
Sébastien Perrigault: We use classic AI, which we have combined with an internally developed IDP (Intelligent Data Processing) software called Doc Factory. This IDP solution is already widely used within the Group and is based on NLP (Natural Language Processing) technologies. NLP models are capable of processing large volumes of text, in this case all the text in mortgage application documents and all customer data. The AI has been trained on thousands of documents to identify the right data and then perform the appropriate checks.
What does this artificial intelligence solution offer bank advisors?
Sébastien Perrigault: For our advisors, IDP - or Intelligent Document Processing - acts as a “super assistant” at several stages of file processing. More than 80% of the document data is automatically read by AI and 60% of data is automatically matched and validated by AI. The end result is a significant time saving of around 30 minutes per file.
What's more, in addition to this assistance, the IDP helps reduce risks by checking the consistency of documents or their authenticity. On the end-to-end “real estate” path, the IDP therefore makes it possible to save 5 days and grant a mortgage in less than 30 days. This is only the first step.
For the simplest offers, i.e. around 20% of them, we aim to issue the contractual document for the mortgage offer in digital format directly in the branch. This will further reduce the average time between the customer's decision and the issuance of the offer by about 3 additional days.
The time saved in processing files will allow advisors to support our customers in their purchase, for example, by advising them on energy renovation solutions both to reduce their bills and to increase the value of their property, while reducing CO2 emissions.
For our advisors, IDP acts as a ‘super assistant’ at several stages of file processing. This AI was integrated directly into the existing loan planning tool, which greatly facilitated its adoption by our advisors who took ownership of it smoothly and quickly.
of our customers' real estate files benefit from IDP*
*for the 4 main documents
How is the customer experience of the “real estate path” optimized?
Sébastien Perrigault: The use of AI reduces the risk of error and increases the reliability of the data entered into our mortgage files. It is a virtuous circle since by improving the quality of the file, we optimize the deadlines and therefore customer satisfaction. The “real estate” path is highly valued by our customers and these technological developments allow us to further improve their satisfaction by offering an increasingly simple and fast experience.
For me, there are two key factors behind the success of this project:
- the use of AI to solve a problem that has allowed us to redesign our mortgage process;
- the close collaboration combined with the clear and shared vision between the different multidisciplinary teams mobilized.”
What are the next steps and ambitions for the tool’s roll-out?
Sébastien Perrigault: The project went live in November 2024 throughout metropolitan France, including for our Hello bank! digital unit and for private banking. Today, 80% of our customers' real estate files benefit from IDP for the 4 main documents: purchase agreement, property tax, pay slip and tax notice. We are developing new AI models that, from 2025, will focus on other documents: the energy performance certificate or EPD and savings account statements, for example. They will strengthen consistency checks and improve risk management.
The next big step is planned for the beginning of 2026, with the possibility of being able to issue around 20% of offers (the simplest ones) electronically, without having to go through the back office, further significantly reducing the timeframes for our customers, not to mention the costs.