On Process Automation in Mortgage Financing

Recently we had the chance to discuss mortgage financing and the different process steps that it involves with a group of bankers. We had a nice exchange in which we learned a lot about the mortgage financing process and gathered ideas of how AI and automation could help improve efficiency where it is needed the most. Inspired by this conversation, we feel like blogging about what we learned. As a note, they explained that some of the details are being handled differently by different banks, and many of what they do has to do with the regulations, they have to follow. 

Process Automation in Mortgage Financing by midjourney

Here is how we understood the process: first an appointment is set up with potential customer. During this step the customer is given a url where they can upload all the documents that will be needed during the process. Alternatively, customers can bring the documents to the appointment where the mortgage consultant will scan and upload them. In the case of this bank, all the documents are stored in this one folder, where they have to be manually sorted and classified by the consultant. 

They have to check if the documents are usable (readable scanned, not out of date,...),  they have to see if all the necessary information is there (salary slips, land register, ...) and lastly they have to check that the documents are not faked. This one is a bit more tricky, some banks use software for this, others have long time employees that "just know what to look for".

Once all this is done, the information is transferred to the valuation department, which uses special certified software, based on current data, to figure out the fair market value of the object. Ideally, this step just involves typing info into a software mask, but finding the relevant data (square footage, year of construction, ...) may take up some time. And once this is also done, the value is reported back to the consultant, who can prepare financing options and finally a contract.

Different steps in the mortgage financing process. There are more steps further down the line, but we focused on the once shown here.

One major issue that is making this process slow is the inefficient document management in this upload folder. The bank employees have to go through the documents and decide what is what, and sort the files, find the needed information and keep it somewhere for later use. In cases where eg two properties are described, one that is to be bought and one that is to be used as security, the valuation department has to understand what is to be used where before they can process the data and return back the results. We were somewhat surprised that the document management is seen as such a major hurdle by the bankers. Using modern vision and language models, the documents could be assessed, classified and entities extracted in near real time. As interface for such an automated handling of documents we imagined a chatbot, that the employee (not the customer) could use to trigger the processing. The insertion of data into the special valuation software could be done with RPA (as this software has no APIs). Meaning many of the straightforward cases could be done completely automated. A MortgageBot so to speak.

List of different technologies and where we feel they can be utilized to advance the mortgage financing process. We had Power Platform in mind when coming up with that list, but obviously other solutions are viable too.

From what we learned this sounded feasible, and we even got around to make initial estimates for how long it would take to develop such a bot. Since this bank already made use of UiPath and is to introduce Power Automate and Copilot soon too, we felt those would be the perfect tools to implement the automation. The bankers appeared very interested as well. Obviously, a closer inspection of the process would be needed to actually implement this (like we'd need to see this software and some example documents) but if they decided to actually try this out, we'd be happy to take part.