AI and OCR is being used to speed up mortgage applications, here’s how.

The last time I applied for a mortgage, I had forgotten how many different pieces of paper and documents I needed to find and hand over to the bank.

At the time, I couldn’t help but think how irritating this was, and how that surely they had all this data to hand already – especially as I was an existing customer with the bank.

What I hadn’t really thought about was the process and technology behind the application and reams of paper I was relinquishing. Not until I recently met a company based in the Moorgate WeWork, London.

TechCo

Having written a few of these blogs already I realise that I use terms like “the company” or “that company” a lot, so from now on I will simplify this by using “TechCo” (short hand for “Tech Company”).

The Chief Financial Officer (ok, let’s throw in another TLA acronym here, “CFO”) of TechCo is a very experienced chap, very pleasant to speak with and is also a board advisor of another interesting company I met that is revolutionising how startups, investors and advisors work together – but I can cover that another time.

The CFO describes to me the effort that goes on behind the scenes of a mortgage application to simply extract and process the data from the forms and letters that get sent in, and the particular example he uses is payslips.

AI for payslip interpretation.

A recent Google search for ‘payslips’ shows how AI could help identify formatting

This example is about payslips, because it’s probably the most well understood formal document, but the same ideas can apply to invoices, quotations, product sheets, academic papers, basically anything that has some structured layout but could be different in each instance.

For example, when I look at the last payslips I had from three companies (that takes quite a bit of work because I’ve been running my own businesses or a contract for many years now), I see that indeed, there’s information about

• Me, name, address, date of birth, national insurance number etc
• Current pay period, how much I got paid this month, benefits and deductions
• Tax, how much tax I paid this period
• Year to date, How much I’ve earned this tax year, how much tax I paid etc.

When I look at the three examples all this information exists but in different places on the forms. One has the “me” data in the top left, another in the top right. One has the “year to date” in a column on the right, another has it across the bottom.

Simple but complicated 

This kind of problem is very easy for a human to solve. We easily recognise particular shapes or layouts of information before we even read it, then we give it a quick glance to confirm that, yes, that’s the “me” section.

For computers this can be harder and where AI comes in.

OCR was AI back then

Courtesy of DropBox, click to visit their blog on OCR in their product

Thirty years ago it was probably very difficult to think that a computer could look at a picture and determine what was in it. And then optical character recognition came along (“OCR”).

A loose paraphrase of the philosophy behind artificial intelligence (AI) is the quest to enable computers to do something humans can but computers can’t. As such, OCR was originally an AI accomplishment, yet today seems quite mundane and not particularly “magical”.

OCR has its limits as many of you will likely have experienced, but ask OCR to deal with information the challenge I mentioned above and it’s a whole level of complexity harder.

TechCo’s technology is specifically designed to find where in a digital image of a invoice, payslip or other semi-structured form the information that needs to be captured can be found. Once that is accomplished, good-old-fashioned OCR can come in to play to analyse and extract the data.

Turn weeks into hours

So what’s the benefit?

Today when a bank sends off your forms to be checked, the major part of that process is handing the digital images to humans to extract the information. This currently takes days and snowballs into a week or two of delay as the information is moved around, processed and sent back.

Automating the whole process from digitising the image to identifying the information blocks, extracting that data and feeding that data into the decision-making algorithms helps reduce the process from days and weeks into mere hours.

TechCo has already delivered their technology into a major global bank and has real-world examples of this process improvement to share.

So next time you’re in a bank applying for a loan or a mortgage, you could be in for a surprise, things might have just got a whole lot better for you.

How could this help you?

Maybe you work in a bank, and are looking for ways to improve your processes, maybe you work in a company where you deal with invoices or quotations from multiple vendors, or maybe you’re in a governmental body that deals with all sorts of identity information. Maybe you read my previous article on digitisation and now have lots of digital documents to deal with? If you’d like an intro to this company to explore it further? I’m happy to do that for you – no cost, no obligation, just ask.