Shownotes
Imagine if you were to receive 500 loan applications one day and they just they showed up at your doorstep. That may be one, two, three boxes of paper, depending on all the supporting documentation that goes with that. You can imagine what this looks like. A loan package, on average, is around 600 pages. That’s what Charlie Weidman aims to solve with his company, Buddha Logic, an enterprise content management (ECM) solutions provider that helps companies streamline digital document capture and management. Folks get intimidated by the word robot or automated process and there’s some level of concern about them replacing people, but Charlie says it’s an education and an awareness that helps you eliminate unnecessary work when you start automating more.
—
Buddha Logic: Simplifying Life with Charlie Weidman
We have Charlie Weidman as our guest. He’s the CTO and President of Buddha Logic. We’re going to do a deep dive in robots, digital transformation, and financial process automation. If that doesn’t take in and make your toes curl, we’ll dig into it. Charlie, thanks so much for taking the time.
Bob, it’s a pleasure to be here.
You and I chatted in a previous podcast and we talked about what you do at Buddha Logic. I was sufficiently interested in what I thought the possibilities were that I wanted to come back and do a deep dive because I think folks don’t quite get what it is that you do.
I’m happy to share what I can and give you some insight into what robots are, what digital transformation is, and what we can do with financial processes.
We were talking before we started, you’ve been working with a Housing Authority here in Denver.
It takes weeks to process a paper application.
That’s correct. We’ve seen them through quite a transition. If I were to look back, we were processing truckloads of paper that would show up and be captured in one way or feature or fashion or another. Scanning, sorting, separating, and batching. It would take a week to process a paper application.
I think about that from the end-user, from the customer standpoint, “What’s going on with my app?” It’s somewhere between the loading dock and the fourth floor or whatever. What I thought would be useful is to go back and go “This is where we were years ago.” They didn’t just jump off the cliff and started doing robots. I don’t even know that “robots” were available at that time. Let’s go sequentially what were they doing.
If you really want to start from scratch, you’re an entity that has not even thought about moving away from paper, getting into the digital transformation of paper. This Housing Authority was pretty much right at that space back then. They made a decision to say, “We can’t process paper anymore and the paper we get in, we need to digitize and move it into repository and make it available for our loan processors to take care of our customers.”
Phase one, let’s define a backbone, put together something that allows us to capture that information, and understand what we have and get it into the processors hands within a week’s time. Let’s say seven business days, that would be a huge goal. At that time, it was all over the map, how quickly something got processed. Sneakernet, paper from one desk to another in inboxes and outboxes. You can imagine what a mess that that was.
I had this vision what people’s desks look like.
Just think of the storage space you need to handle that paper because you’ve got to keep these documents for a certain period of time to make sure that you captured everything and then you got to go find it if there’s an issue. Not only do you have paper coming in and you’re capturing it, you have the storage of the paper.
Whose desk is it on? If line two is wrong, who does it go back to?
It was quite a challenge. Phase one was how do we streamline and centralize where the paper shows up. We have a mailroom. We have this concept of a mailroom and we’re going to automate the mailroom. It’s a robot, if you want to look at it. I’m going to start automating how I capture my paper, that was phase one.
In phase one, when the document came in, what automated it? What did you capture?
Buddha Logic: Automation was really just an understanding of content and how the content needs to be separated and fed it into the digital process.
You’d have a handful of people that would look at the document and they would start sorting the document types. This is beginning the capture, beginning scanning. We’re going to segregate these, then I’m going to scan in batch applications. Here’s a scan and a batch of titles and some insurance documents. That was phase one. Let’s understand the documents that we’re ingesting. They created a job aid for all these people. I have this huge board that says, “When you get this document, this is the pile that it goes into and that’s what we scan it into.”
That automation was really an understanding of content and how the content needs to be separated and then how you feed it into this digital process. Once that was accomplished and they felt good about that, I think that took them about a year, year and a half to get comfortable with. They’ve got this beautiful job aid that represents 160 plus document types. People who they bring into train, start understanding what they’ve got and how to process it.
For our audience, they may not visualize the document mass.
A good way to look at that would be, imagine if you were to receive 500 loan applications in one day. You can imagine that may be one, two, three boxes of paper, depending on all the supporting documentation that goes with that. You’ve got everything from your insurance information, bank information, and appraisal reports. You can imagine what this looks like. I would say an average loan package is around 600 pages. That’s a lot per document. The volume is important to think about because then they have to store it for a period of time.
You have to have a room or a floor in your building. If you’ve captured them, wonderful. They are electronic, but you still have to hang on the source for 30 days or 60 days, depending on what your requirements are for your particular business. That’s the volume. Phase one was, we understand what we’re getting, we have a good idea how to separate it and put it into a process that allows the people who need to review and decide, “This is a good risk. Let’s go ahead and follow through and make this loan.”
We’re at about five to seven days from who knows how long it took before, maybe it was a couple, three weeks. Like you said, the customer is going, “Where’s my app?” I was like, “Let me go find it.” That takes even more people, “Who’s got this loan?” They have to go to the file room and check to see if it came in. You can imagine all the steps that are going on to track where something is at in any particular point in time in a paper process. Even when it’s digital, that first phase, unless it was in front of the processors. What they were learning is now we have to have a way to say, “Once we’ve digitized it, how do we retrieve that information quickly when we have somebody who’s interested in what the status of their loan is?”
By doing phase one, you start identifying those things that you need to think of in phase two. How do we improve our automation process? Part two is, “Can we stop sorting all the paper? If we have to keep doing paper, can I put it in a scanner in a stack and let the software take care of it?” Phase two was to bring in different toolsets that said, “I understand what a mortgage document is. We’re going to create a model that this software does and says, ‘I just scanned 500 pages and I know that there’s a loan app in there. I know there’s an appraisal report. I know there are titles and deeds. I know there’s a verification of employment, all sorts of certifications.” The robot or the machine has to decide, “Here’s what I found, now I’m going to present it to the same mailroom people who have learned what these document types are from phase one and say, ‘Did I get it right?”
Now, the robot is starting to take that approach or the machine, the software saying, “I’ve done my best to understand what you sent me. I’m presenting it to you to tell me if I’m right or wrong. When you tell me I’m right, I’m going to remember that information. When you tell me I’m wrong, I’m also going to remember that, and you tell me what it should’ve been.” We’re talking about machine learning. How does the machine understand when it makes mistakes? The human has to tell it. The robot has to understand, “Next time I see this, I know it should be this document type, not the one I thought it was.” There’s some auto correction that goes on.
The automation process takes some of the work out of the human and lets the machine or the robot do it.
Phase two took about a year and a half. It was to not only continue the automation process, but now take less of the work out of the human and let the machine or the robot do it. I’m going to let the robot tell me what I’ve got and I’m just going to review. I’ve changed how I do my job. I’m hired to review documents, not sort them and batch them and go look for them. With that piece was how do we grab information about that document so when we store it into a repository that can be searched and retrieved by the processors? What’s the information around it so it makes it easy for that loan operator to go, “Here’s the application to co-borrower and borrower name. Put that in, show me all the documents associated with those two. Maybe they have a loan number, let’s grab the loan number.” Every document that’s in the repository comes back.
What was the reaction from the customer, the people that were submitting the apps as this was going through?
First of all, to say five to seven days, it was a big win. It could have been a couple of weeks before they heard anything back, maybe three. The service level agreement or SLA of people coming into the Housing Authority says, “Within five to seven days, we’ll get you a response back.” They were very pleased about that. It was a big win for them.
How did the folks that started at the beginning with handling the paperwork, what was their response to this change?
There was a lot of relief. That’s how I would describe it. What used to be when they find out there’s a problem, somebody’s complaining, we have a customer that’s not happy, everybody’s in scramble mode to go find all the documents associated with it. With the transition into digitizing that information, there’s one place to look. Occasionally, they still had to go reference to paper because maybe we missed something. Maybe we didn’t get everything that we should have scanned in. In general it’s like, “At least we have one place to go look for a data now.”
For the housing agency, did they start to do notice the cost reduction at that point?
What happened was they started to get more customers. Since they could offer the five to seven-day turnaround, now more vendors were interested in using them. The volume of paper went up. It’s a wonderful problem to have. I’m getting twice as much paper as I used to, which is okay because they set up phase one to handle that. Storage grew, so now they have to ship some of this stuff off site, so you’ve got the paper moving off site. Customer participation, you got more customers. They are thrilled that they have a five to seven-day SLA versus the two to three weeks. For the second phase which is how do we do our automation and make it smarter, continue to let the machine do as much work as possible and let the humans review.
Phase three was how do we reduce paper? How do we grow our customer base and reduce the amount of paper we’re gathering? Paper costs money to store. It takes a lot of time to process even if you have fast scanners. This Housing Authority had four scanners running all the time. That’s a big investment. That’s quite a crew to run those and keep things flowing. How do we shrink the paper? What the Housing Authority did was offer incentives to say, “If you email us or drop this into an FTP site as a PDF or an image, you don’t have to send us the paper anymore.”
These vendors who were shipping boxes of paper daily just saved anywhere from $10,000 a month depending on the vendor to even more for some of these larger vendors. They’ve just saved themselves shipping. A lot of times they want it done as quickly as possible so they FedEx stuff. Imagine the cost savings for their customers. The software has to take the next step, “How do I ingest these electronic files?”
You have to have bandwidth. There were some investments that the Housing Authority had to make. We’re going to set up an FTP site that allows people to put documents there. We have to take another tool set that says, “We have this wonderful model that we’re building that recognizes mortgage documents, understands what they do, and we’re going to now feed it without paper. We need to grab it electronically, do the same thing, split it up, and present it to somebody who understands what the document is.” They say, “You got it right,” off to the processor. That was really the last piece of this automation process to get them to the point where paper started dropping. It took about six months to go from 98% paper and 2%electronic to the opposite, which is 98% electronic to 2% paper.
As soon as the larger vendors understood they could do this, they switched almost immediately. What happened is like, “Let them do the scanning.” Now, I’ve taken a part of what I used to be as a customer service, I’m providing you the service, I’m letting you do that for me. What I’ve offered is you don’t have to send me the paper and we’re going to shrink our SLA down to four days. It’s a huge win for everybody. The customers are actually happy about it. You’ve shifted the burden of the scanning to the customer. There are some little stars you put that because quality of image drives everything with machine learning and understanding what you have. If the image is poor, I’m not going to do a very good job as a robot or a machine and say, “I understand what this is. I’m going to give it my best guess.” Encouraging the vendors and the customers to do a good job, that gets a gray area. It’s hard to enforce that.
Buddha Logic: The model continues to grow and get better in understanding the documents coming in.
The opposite side of that is, “It was really poor. Yours instead of being two days, if the document had been really clean, we’d probably could have gotten back to you in two days, it took four.” SLAs, it basically represents, especially when you’re doing digital transformation, what it takes at the extreme end. If you send me really poor stuff, it’s going to take me my maximum SLA. If you send me really good stuff, I might turn that around in hours depending on how clean it is. What we’ve done is we’ve refined the model. The model continues to grow and get better to understand the documents coming in.
Now, we’re extracting data off of the actual application. We started with classifying and separating, and putting some minimal data around that so it could be retrieved quickly. What we’re going to do is say, “Let’s extract the information off of the document so nobody has to enter it.” Processors have to enter a number of the data fields into their systems. They were happy to do it because before they were working off a paper or whatever. That wasn’t an issue.
The next step is let’s let the machine do the next step. Extract the data, try to determine through the databases that are available if the data that I extracted matches. I know this is good data, I don’t even need to present it to somebody. I can take that data and put it into the system. If it’s not very confident, now I can present it to an operator. It’s that same concept. The machine thinks it’s this, “I need a human to review it, yes or no.” It starts learning.
What happens in the next phase is our model not only understands what it’s looking at, it understands where the information is on it. It extracts it and puts it into this system that helps them make the decision whether or not to make the loan to this customer. It took us about six years to get to that point where we’re extracting data, we’re very comfortable with how the model works and we’re starting to look at how do we expose the workflow portion of it? How do we move this information in such a way where the customer has visibility into the process? No longer is it, “I fed you my documents, I’m waiting for a call or I’m calling you.” I need to show. How do we make it visible?
There’s been a couple of ways that they’ve accomplished that. They set up a portal. The customer has a place to go to put their documents in and because they used that portal, and there’s information that we enforced. The robot doesn’t have to be as smart in many ways because now we’re controlling what the data looks like when it comes in. It has to be a PDF, it has to have certain attributes associated with it. You’ve got to tell me the documents that you’re sending. We’re shifting some of that work into the customer’s hands. It becomes more incentives because now we’re dropping down our SLA to less than a day. Here’s how you incent the customer to start using the tools that allow you to do a better job.
We’ve gone from seven to a day?
When I received that document in my world, in a day, I’m going to get a response back to you.
That’s in year six?
The robotic processes do the manual repetitive tasks that can complement what we're doing.
Six to seven and now, we have a day turnaround time. We’ve got a way to do a better job of understanding the data coming in. That speeds up our process, it helps with separation, classification, and extracting data. The customer knows within a day I’m going to get a response. Transparency starts. They can’t dial in and look, but when they feed the portal, the document set that get fed has these were pending and within a day, we update that...
