Listen: Fintech chief operating officer shares 3 barriers to bank automation
Hyperscience’s French reveals key performance indicators for measuring automation success
Legacy systems are one of the financial industry’s top three barriers to automation. That’s why digital banks, which are unencumbered by legacy applications, are often able to do what larger banks can’t — and at a faster pace.
In this podcast, Charlie Newark French, chief operating officer for the automation fintech Hyperscience, discusses with Bank Automation News the three main barriers that banks face, but that neobanks do not. The company uses machine learning to automate account opening, the know your customer process and anti-money laundering verification. TD Ameritrade and ONE Insurance are Hyerscience clients.
After legacy systems, the next most significant barrier is how banks are required to support customer interactions. These interactions can even include, French added incredulously, faxes.
Finally, there is the “systemic issue” of delays that even financial institution executives don’t clearly understand, French said. “It takes six weeks to file a mortgage. … Nobody understands fully — even within the banks and especially the higher and higher you go up — what is causing all that delay.”
French also discusses with BAN the three key performance indicators that financial institutions use to measure success when automating.
Hyperscience has raised $188.9 million over seven funding rounds, according to Crunchbase.
Bank Automation Ignite, on April 13-14, is the event for inspiring automation initiatives and investment in financial services. At the virtual event, financial services professionals can discover new use cases and technologies that are accelerating automation in banking. Learn more and register at www.BankAutomationIgnite.com.
The following is a transcript generated by AI technology that has been lightly edited but still contains errors.
Loraine Lawson 0:01
Good day. This is Loraine Lawson with Bank Automation News. Recently I spoke with hyper sciences Chief Operating Officer Charlie Newark, French Hypersciences, an automation company that leverages machine learning instead of bots to automate opening accounts, the know your customer process and anti money laundering verification. I asked French what he sees as the biggest barriers to automation, and what key performance indicators banks and other companies use to measure success in automation.Charlie Newark French
I think that I would really say three things, and most certainly, mostly the first. The first is the legacy text app, I gave this example of better mortgage, and so far, and there’s other smaller companies popping up being able to do do very nice things a lot, lot faster. And it’s like when the new airlines popped up in the sort of 1980s 1990s. And were able to be a lot more cost efficient, versus sort of some of the incumbents who are all filing for bankruptcy at the time. They were just unencumbered by a whole series of stuff. And so for this space, it is until unencumbered by legacy tech stack, like small changes for large banks have very, very tough. And you can have a sort of sofi come along and sort of say, if you want to borrow 100, up to $100,000, and you live in this zip code, and you earn this amount, and you sort of really, really narrow it down, we can process that for you in 24 hours. But we certainly start expanding out of that things become difficult, very fast. And the legacy tech stack, as a result has got has become very complex, and fragile and hard to operate. So allow the bigger the bank, the worse it is. Because the more differences, the more nuanced they’ve had to deal with. Our aim is to make change possible, right? So of the 100 different software or systems or databases or data entry. software that’s used, or databases are used. You want to swap out one without model, it shouldn’t be possible, that’s out a way that we can see that. Like I think the second is that people want to communicate with a bank in different ways and can communicate with the bank in different ways. And I think a lot of again, the likes have a better mortgage is able to say right, if you’re able to put all of this in online and or communicate to, to us via this one form. That’s great. A big bank, I mean, 2% of what we process is faxes, right, I say that for the just pure absurdity of it. I didn’t even know faxes existed still until about two or three years ago. But some people want to physically mail stuff in some people want to take a picture on their iPhone, it’s kind of skewed and doesn’t doesn’t go into an OCR engine at all. Some people want to submit part of their data via an email and some part of it via calling somebody up and submitting it that way. And you have this sort of diverse set of documents and data coming in. And the sort of harder that is, the faster the likes of a box breakdown. And they’d like Center, a large bank will only accept or the customers of a large bank will only accept data in this exact format. It just won’t work. So that’s the second. And again, machine learning is the only way to do that. I mean, if you think of if you think of doing even proof of identification, the number of documents that can be submitted for proof of verification, from passports that were issued in different years and have things in slightly different places to driving licenses are different by every state by every year. The number of sort of custom code you need to write to say look at the top right and take the last seven digits with this exact pixel set just makes bots and kind of manageable ways to try to solve that problem. So the first is legacy tech stack. The first is customers want more and more freedom on how to communicate with customers. And the third is is really a sort of systemic issue where everyone knows the problem, right? All three of us on this call everyone in a bank. Everyone in the world knows the problem. It takes six weeks to file a mortgage. It takes two years to file a Social Security Disability claim it takes up to three One of the largest insurance companies to file an insurance claim, see the problem. Nobody understands fully even within the banks, and especially the higher and higher you go up, what is causing all that delay.And we try to, we try to expose that issue a little bit more, and make it make it more tangible for or more possible for people to do change. So when we go in, we don’t rip everything out straight away. We slow we slowly take on more and more. We’re sort of saying, well, there’s a little bit that we can do a little bit better. And then once we proved ourselves there are well, there’s also this module of our software that you can take. So isn’t this daunting? over the whole approach. Okay.Loraine Lawson
Thank you those who could answer. also wondered, I’m writing a story about KPIs for automation, do you have any recommendations for what kind of KPIs companies banks should be using to judge their automation?Charlie Newark French
Yeah, we look at really three KPIs. And then we’ve done one of them, there’s a little bit of nuance, the first KPI for us is accuracy. But one thing we do, we don’t have an OCR engine, every piece of software that we’ve ever built is proprietary. We use machine learning for almost everything we do, and certainly for extracting data. But with our machine learning software, we put an emphasis on accuracy. And some of our competitors will say 100% automation. And for us 100%. Automation really means 0% automation, because there’s no way in this is a super specific target, super specific task, in which case, it is just a very, very custom piece of code which has existed for 30 or 40 years. But there’s no way that automation software can do everything with 100% accuracy. In fact, people don’t we know that on average, and most tasks that we see, there is about a 1% 2% sometimes failure rate or clerical error or inaccuracy. So what we do, so if you’re saying 100% automation, you have to be you basically have to go and check everything. So you probably see it sped it up a little bit, but it’s not any true automation. What we do is we assign a confidence interval. To every single piece of work we do, we say like, I’m 99, and this is a machine that machine has 99.5 to 7% confidence, this name is Loraine. And it’s spelled like this. And if that reaches our customers tolerance or errors, which will never will never be 0%, right? It just doesn’t exist not even in human life. But if, if you’re sort of one to 99.5% accuracy rate, yes, this passes that we’re very confident for No, this is actually only 99% confident and as nordnet sort of an accuracy, that rate that is fair, or total tolerable. We then hand that off to a person to check if there’s a discrepancy between what we think and the person we hand off to another person’s check in some instances, or depending on how much accuracy you want. And once there is an answer, consensus based answer, we submit that answer. And all the time the machine is watching what people do and learning. So next time you present that same problem to machine, it’s hopefully a slightly better idea, maybe still not good enough, because we always prioritize accuracy over over anything else. So I think accuracy matters in automation, automation for automation sake, will backfire on large enterprises. The second one is really speed. Now speed is the one I say breaks down a little bit more than just speed, you can think of as some form of straight through processing, or STP metric, which is like some portion of the work is just fully able to be done by the software. Or you can think of sort of breaking up into discrete chunks of work, right. So if you think of a mortgage application, straight through processing is obviously 0% right now, even for the most nimble companies, but automation of various parts of that may be extremely high. So straight through processing is one average handling time. There’s another is a sort of opposite end of that spectrum as well. We’d probably prefer to sort of say this mortgage took two weeks of handling time. Some things we do straight from processing for, like paying and paying an invoice is there’s only a few steps in that and it can be relatively automated the whole way processing mortgage does not fit into that category. So the first is accuracy. The second is speed, went straight through processing, or average handling time being the sort of key metrics we use there.And then the third is cost, right? My thought is that most companies that I speak to certainly at the sort of C suite level care mostly about speed, especially in times of COVID, when they’re being forced to be more competitive, being forced to serve their customer base substantially better than competitors. Because the customer base is the sort of in tight in tough time, has more power, right, and the ways that things were done in the past, if you keep that you go under in times like this, and if you change and you become more competitive, you do a better job. So I think people care mostly about speed. Accuracy is obviously just constantly critical. And then the third area is cost. And serving customers in a cost efficient way is important. Just to give you a sense of those metrics for us, we reduce error rates by two thirds, which is a radical reduction. We increase throughput time, by 10x. And we can save up to although it’s a bit of a trade off between speed and between speed and cost, that we can save up to 90% of costs. Now, that all depends on, if you’d rather just things go really, really fast, then you leave more people processing for that. To maximize speed. But that’s sort of a trade off the companies we work with may conduct case by case basis.
Good day. This is Loraine Lawson with Bank Automation News. Recently I spoke with hyper sciences Chief Operating Officer Charlie Newark, French Hypersciences, an automation company that leverages machine learning instead of bots to automate opening accounts, the know your customer process and anti money laundering verification. I asked French what he sees as the biggest barriers to automation, and what key performance indicators banks and other companies use to measure success in automation.Charlie Newark French
I think that I would really say three things, and most certainly, mostly the first. The first is the legacy text app, I gave this example of better mortgage, and so far, and there’s other smaller companies popping up being able to do do very nice things a lot, lot faster. And it’s like when the new airlines popped up in the sort of 1980s 1990s. And were able to be a lot more cost efficient, versus sort of some of the incumbents who are all filing for bankruptcy at the time. They were just unencumbered by a whole series of stuff. And so for this space, it is until unencumbered by legacy tech stack, like small changes for large banks have very, very tough. And you can have a sort of sofi come along and sort of say, if you want to borrow 100, up to $100,000, and you live in this zip code, and you earn this amount, and you sort of really, really narrow it down, we can process that for you in 24 hours. But we certainly start expanding out of that things become difficult, very fast. And the legacy tech stack, as a result has got has become very complex, and fragile and hard to operate. So allow the bigger the bank, the worse it is. Because the more differences, the more nuanced they’ve had to deal with. Our aim is to make change possible, right? So of the 100 different software or systems or databases or data entry. software that’s used, or databases are used. You want to swap out one without model, it shouldn’t be possible, that’s out a way that we can see that. Like I think the second is that people want to communicate with a bank in different ways and can communicate with the bank in different ways. And I think a lot of again, the likes have a better mortgage is able to say right, if you’re able to put all of this in online and or communicate to, to us via this one form. That’s great. A big bank, I mean, 2% of what we process is faxes, right, I say that for the just pure absurdity of it. I didn’t even know faxes existed still until about two or three years ago. But some people want to physically mail stuff in some people want to take a picture on their iPhone, it’s kind of skewed and doesn’t doesn’t go into an OCR engine at all. Some people want to submit part of their data via an email and some part of it via calling somebody up and submitting it that way. And you have this sort of diverse set of documents and data coming in. And the sort of harder that is, the faster the likes of a box breakdown. And they’d like Center, a large bank will only accept or the customers of a large bank will only accept data in this exact format. It just won’t work. So that’s the second. And again, machine learning is the only way to do that. I mean, if you think of if you think of doing even proof of identification, the number of documents that can be submitted for proof of verification, from passports that were issued in different years and have things in slightly different places to driving licenses are different by every state by every year. The number of sort of custom code you need to write to say look at the top right and take the last seven digits with this exact pixel set just makes bots and kind of manageable ways to try to solve that problem. So the first is legacy tech stack. The first is customers want more and more freedom on how to communicate with customers. And the third is is really a sort of systemic issue where everyone knows the problem, right? All three of us on this call everyone in a bank. Everyone in the world knows the problem. It takes six weeks to file a mortgage. It takes two years to file a Social Security Disability claim it takes up to three One of the largest insurance companies to file an insurance claim, see the problem. Nobody understands fully even within the banks, and especially the higher and higher you go up, what is causing all that delay.And we try to, we try to expose that issue a little bit more, and make it make it more tangible for or more possible for people to do change. So when we go in, we don’t rip everything out straight away. We slow we slowly take on more and more. We’re sort of saying, well, there’s a little bit that we can do a little bit better. And then once we proved ourselves there are well, there’s also this module of our software that you can take. So isn’t this daunting? over the whole approach. Okay.Loraine Lawson
Thank you those who could answer. also wondered, I’m writing a story about KPIs for automation, do you have any recommendations for what kind of KPIs companies banks should be using to judge their automation?Charlie Newark French
Yeah, we look at really three KPIs. And then we’ve done one of them, there’s a little bit of nuance, the first KPI for us is accuracy. But one thing we do, we don’t have an OCR engine, every piece of software that we’ve ever built is proprietary. We use machine learning for almost everything we do, and certainly for extracting data. But with our machine learning software, we put an emphasis on accuracy. And some of our competitors will say 100% automation. And for us 100%. Automation really means 0% automation, because there’s no way in this is a super specific target, super specific task, in which case, it is just a very, very custom piece of code which has existed for 30 or 40 years. But there’s no way that automation software can do everything with 100% accuracy. In fact, people don’t we know that on average, and most tasks that we see, there is about a 1% 2% sometimes failure rate or clerical error or inaccuracy. So what we do, so if you’re saying 100% automation, you have to be you basically have to go and check everything. So you probably see it sped it up a little bit, but it’s not any true automation. What we do is we assign a confidence interval. To every single piece of work we do, we say like, I’m 99, and this is a machine that machine has 99.5 to 7% confidence, this name is Loraine. And it’s spelled like this. And if that reaches our customers tolerance or errors, which will never will never be 0%, right? It just doesn’t exist not even in human life. But if, if you’re sort of one to 99.5% accuracy rate, yes, this passes that we’re very confident for No, this is actually only 99% confident and as nordnet sort of an accuracy, that rate that is fair, or total tolerable. We then hand that off to a person to check if there’s a discrepancy between what we think and the person we hand off to another person’s check in some instances, or depending on how much accuracy you want. And once there is an answer, consensus based answer, we submit that answer. And all the time the machine is watching what people do and learning. So next time you present that same problem to machine, it’s hopefully a slightly better idea, maybe still not good enough, because we always prioritize accuracy over over anything else. So I think accuracy matters in automation, automation for automation sake, will backfire on large enterprises. The second one is really speed. Now speed is the one I say breaks down a little bit more than just speed, you can think of as some form of straight through processing, or STP metric, which is like some portion of the work is just fully able to be done by the software. Or you can think of sort of breaking up into discrete chunks of work, right. So if you think of a mortgage application, straight through processing is obviously 0% right now, even for the most nimble companies, but automation of various parts of that may be extremely high. So straight through processing is one average handling time. There’s another is a sort of opposite end of that spectrum as well. We’d probably prefer to sort of say this mortgage took two weeks of handling time. Some things we do straight from processing for, like paying and paying an invoice is there’s only a few steps in that and it can be relatively automated the whole way processing mortgage does not fit into that category. So the first is accuracy. The second is speed, went straight through processing, or average handling time being the sort of key metrics we use there.And then the third is cost, right? My thought is that most companies that I speak to certainly at the sort of C suite level care mostly about speed, especially in times of COVID, when they’re being forced to be more competitive, being forced to serve their customer base substantially better than competitors. Because the customer base is the sort of in tight in tough time, has more power, right, and the ways that things were done in the past, if you keep that you go under in times like this, and if you change and you become more competitive, you do a better job. So I think people care mostly about speed. Accuracy is obviously just constantly critical. And then the third area is cost. And serving customers in a cost efficient way is important. Just to give you a sense of those metrics for us, we reduce error rates by two thirds, which is a radical reduction. We increase throughput time, by 10x. And we can save up to although it’s a bit of a trade off between speed and between speed and cost, that we can save up to 90% of costs. Now, that all depends on, if you’d rather just things go really, really fast, then you leave more people processing for that. To maximize speed. But that’s sort of a trade off the companies we work with may conduct case by case basis.
