Transformation by technology and the human touch

How KPMG re-engineered a system involving hundreds of human analysts and millions of pages of data annually so that judgments became consistent, response time came way down, and reliability was restored.

Thomas R. Frame

Thomas R. Frame

Director, KPMG US

+1 703-286-6888




 

Video transcript

Tom Frame:

(00:08):

Good afternoon, everyone. So this is a story about having way too much on your plate. I read about a dozen books a year. Some for work, some for fun. Fiction, nonfiction novels. A few I read closely, but not too many. Usually I just read before bed. A dozen books, that's about 2,500 pages or about three quarters of a million words. It takes me about, I don't know, about 50 hours to read or an extra long work week, or for many in the room, not such a long work week at all. If I get something out of it, that's great, but if not, it's okay. I'm really just doing it for fun.

(00:40):

Now what if I told you that you had to read more than a dozen books? A lot more. Instead of 2,500 pages, what if I told you you had to read 30,000 pages? And instead of novels, what if you were reading technical manuals? With numbers and details? And instead of doing it for fun, what if you were analyzing it? Testing it? Making sure every number and detail was right and accurate? And to make it crazier, what if you made a mistake?

(01:13)

Financial markets might take notice. They may even start to panic. That's a lot of responsibility on someone whose job it is to read. A crazy scenario, one that surely could never happen. Yet, in 2016 the Federal Housing Administration, or FHA came to KPMG with just this problem. Nine million pages to review annually with details on over 30,000 loans to be reviewed by just 300 people. That's millions of numbers to verify, thousands of situations to analyze, and behind it all, it's all based on a thousand pages of rules, so you have to know those, too. All with major financial consequences if someone makes a mistake or comes to a different conclusion than someone else reviewing exactly the same thing. All resting on the judgment of 300 overworked reviewers across the country, all going in different directions and doing it their own way.

(02:05):

Like I said, that's a lot on your plate. But we cleaned up that plate, so to speak, with new ideas and new technologies that didn't just patch the current solution, but really changed the way that FHA did business. We identified the problem at its core and created a better way to do things from the ground up.

(02:27):

I want to walk you through a little bit about how we solved FHA's problem and how our approach solves a whole host of problems for doing business in the 21st century. But first, a little background. If you’ve ever had a mortgage loan, then you may or may not have had mortgage insurance. Mortgage insurance is an insurance policy that protects lenders and investors from losses due to the default on a mortgage. So similar to car insurance, you get into a car accident your insurance provider pays for any losses due to the damages in the car accident. If you default on your mortgage, your mortgage insurance pays for any of the losses due to that default.

(03:00):

Now, our client FHA is an issuer of that insurance. One of their main goals is to provide and make housing more attainable to Americans around the country, and in order to do that, they issue mortgage insurance on middle to high risk loans that conventional banks may not underwrite, at least not at a reasonable interest rate. Now to do this, they rely very heavily on the banks themselves to make sure that these loans comply with FHA policy. But just to be safe every year, FHA selects about 30,000 loans or nine million pages, that's where I got my number from before, nine million number filled pages that FHA has to review in detail to make sure that loan complies with FHA policy.

(03:44):

See, this is where our work began. You may be thinking to yourself, okay, he's up there, he's about to talk to me about a big data problem. Well, it actually turns out that this was a more of a problem in change management. Not process, not finance or technology or personnel management, but really change management. Because it wasn't always nine million pages. It used to be a whole lot less. In 1997, FHA processed about 900,000 loans. Yet in 2017, they processed 1.16 million loans or a 33% increase over those 20 years. That's more pages to review. A lot more. Simple as that. Not only had the volume of work increased, but the value of that work is increased as well. It's increased from 77 billion to over $200 billion over those 20 years. Which means a greater risk exposure for FHA, and if a collapse, a much greater risk exposure for the market in which it lives.

(04:48):

We had to figure out a way to handle this change at scale. We needed to find a way to focus the reviewers on where in those 300 pages to look. Reviewers across the country all had different ways of reviewing a loan. And we wanted to standardize the way a loan was reviewed while also letting these reviewers' individual review styles to remain. And the 4,000 lenders that did business with HUD or that do business with HUD, they all also had their own unique way of responding to HUD's inquiries about these loan reviews. All with one major goal: to protect their loan portfolio.

(05:23):

That's where our agile approach to solution design and development began and the loan review system was really born. We knew reviewers and lenders across the country were going to be very interested and very anxious to see what this new process was going to be and what the solution that they were going to be spending most of their time in was going to look like.

(05:43):

So I want to walk you through three steps in our process in delivering the loan review system for FHA. Step one – Design. We designed the entire loan review system, every screen, pixel for pixel, in the first two months of the project. Using our proprietary design tool cycle, we uploaded screen designs that then lenders and reviewers across the country could comment on and give feedback on to us. This made the lenders and the reviewers feel as though they were part of the process in actually designing and developing the tool, not that they were just handed a tool.

(06:18):

It also helped HUD with change management in that HUD could then actually train these reviewers and these lenders after just 60 days and before KPMG had ever written a line of code. So in the end, the cycle tool allowed us to really engage those lenders and those reviewers in those first 60 days and from the beginning of the project, all the way to the end. Cycle currently sits in our gov cloud environment and we use it today to help a whole host of our clients solve similar design challenges.

(06:46):

Step two – Q&A trees. We created a process that put everyone on the same page. A reviewer that started with the same data as another reviewer ended up with the same answer. Whether they'd been there for 10 days or 10 years. We created a Turbo Tax style question and answer tree, a piece of our application where questions appeared and disappeared based on actual data that came from the loan itself. Those questions also appeared and disappeared based on answers to previous questions. The answers to the questions also drove reviewers to similar places in the system where they were supposed to take a specific action. Thus actually standardizing the review. We took about 300 pages that our reviewer had to go through with a fine-toothed comb and boiled it down to about a dozen.

(07:34):

Step three – Standardization. We took a mostly reactionary process and created a proactive, planning process that allowed FHA to see what tomorrow looks like, today. We created an algorithm to select loans based on risk indicators that FHA wanted to target. Additionally, we made it easier to match a loan to a reviewer based on what that reviewer does best. We also made it a lot easier to manage those reviewers' vacations, so when those reviewers went on vacation, the loan review system didn't have to go on vacation itself.

(08:09):

Finally, the benefits. In about 20 sprints or about 10 months, KPMG designed, developed, and deployed, a solution that is currently in operations at HUD today that has reduced the amount of time it takes to do a single loan review from 51 to 33 days. That's an over 35% reduction in time that can be spent reviewing additional loans for risk and fraud and helping protect the mutual mortgage insurance fund. To date, we've collected over 250,000 documents that would have been mailed via snail mail to HUD in the old process. At about five pages of document, that's over 1.25 million pieces of paper saved or about 150 trees. We've created a process and a system that allows HUD to be more flexible to respond to changes in regulation as well as industry input.

(09:05):

If you have a drip of water, you want to clean it up with washcloths. But if that drip turns into a flood, well, you don't need more washcloths. You need to get yourself a pump. That's what was happening here. HUD had 300 reviewers, a set amount of loans to review, and a system to do it. When the volume of that work exploded, they didn't need more reviewers, they needed a new system. They didn't need more washcloths. They needed to get themselves a pump. Something for a different problem. We created just that system. We fixed it by looking at what the problem was today, not what it was 20 years ago. We solved it using new technology and old-fashioned critical thinking. There's a whole world of problems out there that come down to change management, and the best solution for them will always be this. Think about what the problem is, not what it used to be. The benefits will be amazing.

Thank you.