Leveraging Machine Learning to Usher in a New Era of Work
Michael Canale discusses the collaboration on BRG’s Digital Workforce product, which leverages machine learning and artificial intelligence to automate a variety of return-to-office functions of business.
TRANSCRIPT
MJ 00:00 Welcome to Force Multiplier, the official podcast of BRG's Global Applied Technology team. The GAT team, as we call ourselves, is a globally distributed team of software engineers, data scientists, graphic designers, and industry experts who serve clients through our BRG DRIVE analytics platform. We're helping some of the world's largest and most innovative companies and governments transform data into actionable insights.
I'm Michael Jelen. And in these conversations, we speak with people both internal and external to BRG to discuss how technology, and specifically software, acts as a force multiplier to extend the impact of people across any kind of professional organization.
Today, I'll be speaking with Michael Canale, the head of BRG's Financial Services Technology team. Mike has over twenty years of experience across consumer finance operations, servicing, and technology. At BRG, he focuses on our financial services technology strategy, products, and strategic relationships. Mike and I have been working a lot on our Digital Workforce product, which leverages machine learning and artificial intelligence to automate lots of the back-office functions of businesses. I love talking through some of the macro trends that will impact this industry as a whole, and the way that Mike's positioning Digital Workforce is something that is always a net-positive ROI and an incredibly useful tool to usher in a new era of work. As always, thanks for joining, and please enjoy this conversation with Michael Canale.
Hey, Mike, welcome to Force Multiplier. It's good to have you back here, and I can't wait to spend some time today talking about our Digital Workforce offering.
MC 01:30 Yeah, thanks for having me, Jelen. I appreciate it.
MJ 01:33 For sure. And just to sort of set things off for everybody to frame the conversation for today, Digital Workforce is mostly focused on automation and ensuring that repetitive tasks at different clients are going to be streamlined and handled by machines rather than people. I was just wondering if you could talk to everybody a little bit about how the overall macro look of the economy, the different firms and different functions within those firms, are going to be changing over the next couple of years. Where do you see automation going and changing?
MC 02:05 Yeah. I mean, I think in terms of the macro setting—I've been talking with one of my business school professors about this for a long time, at least a decade. And his name is Edward Hess. He's written about this pretty extensively. He's citing two particular studies: one's from McKinsey, predicting about twenty-five million jobs will be automated in the near future; and another one from Oxford University predicts about 47 percent of jobs will be automated in the next fifteen years, including professional jobs.
So, this isn't really something that will apply just to manufacturing or back-office work. It's basically any repetitive task requiring data, documents. And with more advanced technology, you can get even more creative about the problems you're solving. So, we're really just at the very beginning of this. But when you start thinking about things like even building financial models, making assumptions, making predictions, these digital workers can get pretty creative.
MJ 03:04 Cool. And we're kind of at the beginning of Q2 2022 right now, where I know inflation is a concern on people's minds. How does that pertain to the cost of workers, the productivity of workers? And how do we see that impacting the choice to automate or not automate certain roles?
MC 03:22 Yeah. So, there's a couple of important things here, right? So, throughout the pandemic, employee costs have risen pretty dramatically. Everyone hears about the great resignations, individuals leaving their companies getting big paydays somewhere else. I think the war for talent's gotten expensive enough where companies really should be asking themselves whether a computer can do the task that they're hiring for.
Then you add on the productivity gap. So, what's about to happen is all about productivity. Just looking at recent history between 2000 and 2021, consumer prices rose 7 percent, while productivity grew around 1.9 percent. So, there's a pretty big gap there that's not sustainable. And that gap is probably much larger today than it was, call it a year ago, because inflation accelerated pretty considerably.
And the other piece of this is there's really only so much productivity you can squeeze out of a human being. So, we've really reached the point of diminishing returns. Yes, you can put in workflow tools, monitor productivity, manage it as closely as possible, but computers really offer the potential for limitless productivity. So, they can process as much as you can throw at them. If you're using cloud services and you reach max capacity, you can always add on more. They don't get sick. They don't sleep. It's really the next logical step for companies to adopt this sort of technology.
MJ 04:54 And all the while, it seems like although we have these overall macro trends impacting the workforce, we've also got technology itself [that] is becoming better and better at doing tasks that humans have traditionally had a competitive advantage on. And I guess, when you take these two things together and you see the speed with which technology is changing, it feels like automation, and therefore digital workers as a whole—they are coming whether you want it to or not. And this is good, because I think in a lot of ways it opens up and frees the human talent to scale up to higher value-add services.
But how do you see the role of your average white-collar worker changing as a result of this? Of course, newer jobs will be created from this as some are destroyed. But how do you see those two things reconciling?
MC 05:43 Yeah. I mean, when I think about a digital worker, they're much like a human worker. It's just an assembly line of different technologies that can perform a specific function. I'll go into a little bit about what they are, and then we'll go back to how I see humans interacting with them.
But in our toolbox at BRG, we have a combination of things like optical character recognition. So, we're transforming documents into data. We can even do handwriting. That's probably the most common use case that we've seen. But there are other things, like natural language processing. So, we can pull out customer sentiment. We can tag and label pieces of information. Robotic process automation, so we can perform tasks like pulling data and documents, prepping it for review, and then entering information into systems. So, it can also kick off things like workflow processes.
And then you have everyone's favorite buzzwords, artificial intelligence and machine learning. But those are really used to look for patterns in data and improve the functions of the digital workers over time. So, the more data you're feeding them, the smarter they get. And then you can layer on things like APIs, where you're pulling in data from external sources and systems, creating analytics models, all types of crazy stuff.
And where we excel at is taking these different pieces of technology and assembling them together to make a digital worker. The process to implement that is really to map out the current state process that any client has that they're interested in automating. And then we do what's called a proof of concept. So, we show them the tools in action. They get the ability to review the results. We then take their feedback. We update our models. So, if there are things that were missed or more complicated edge cases, we refine those things, and the worker continues to perform that function; they get better at it. Typically, we do what's called a testing phase. And we go and process what's called a control group. So, it's something that the client has already done. So, they know what the results are. Our digital worker then performs that task. And we can measure them against what humans did. It's a really good litmus test of how accurate the worker is before we deploy it into a production environment.
So back to your question on what is the relationship between the human and digital workers? The best part about having this Digital Workforce is it really is an extension of your existing team. So, automation doesn't mean you just lay off everyone in your group or in your company, but it frees your team up to focus on more high-value tasks versus doing those repetitive low-value tasks. The team still monitors how the digital workers are performing so they can do things like sample testing. There's still some oversight. But generally, they no longer have to focus on those low-priority tasks.
MJ 08:38 Gotcha. Very, very helpful to have that as a frame of mind. As a company is thinking about onboarding digital workers or taking specific tasks that they have, and they see the offering that BRG has, it looks like we have a lot of tools. And each of these different processes is completely bespoke. So, nothing is going to be built in advance that perfectly fits or solves the problem that client is looking to solve.
How do you get over the hurdle of distrust in automation or distrust in machine learning and artificial intelligence to demonstrate that there is actually so much value to these sort of digital workers and win over the stakeholders that are involved?
MC 09:18 Yeah. I mean, part of it is showing some of the use cases that we've already developed. I think that's tough, because it's like selling a piece of real estate. There are studies that show if you furnish your house and you stage it, you get a better return, versus if you sell an empty house. And this is kind of the same way, right? If you're talking to a group in cashiering, and you're showing them something from tax, it's not necessarily going to resonate. So, they do have to take a little bit of a leap of faith to get started on that bespoke journey. But I think they quickly see—once we're using their documents, and they see how quickly the digital worker can process their work and what the output is and how nicely we package it together; and DRIVE, and they have a nice dashboard that they can go in and review the results, I think it really speaks for itself. In terms of winning over other stakeholders inside the companies, I mean, it really is about just getting a couple of small wins lined up. And once they see the value, it snowballs from there.
MJ 10:22 Perfect. And so, it seems like at the beginning it might even make sense to run the two processes in parallel and be able to compare them side by side, the human on one side and the digital worker on the other. Can you talk a little bit about how we might use that human input to be able to train the digital workers? Do we just look at a set of already labeled, already finished data? Or do we have other tactics for starting from scratch with a digital worker and a specific task that we want to train it to do?
MC 10:50 Yeah. I mean, we can do both. So, if it's a task where there is a nice amount of data that's already been reviewed by human beings, and we can train our models, I think that's obviously beneficial to us, because we can probably ramp up the digital worker much quicker. But for tasks where there really isn't a history behind it, it's really about getting the digital workers stood up and then having that feedback loop where the client can actually go and review the results.
We actually have workflow mechanisms built into DRIVE so they can actually give direct feedback based on the results. Then our team takes that, refines the analytics models behind that are actually driving what the digital worker is doing, and that will improve it over time. So that human feedback loop is really important, right? Because there's only so much that you can automate right from the start. The really easy stuff you can get done relatively quickly. But then, as I discussed earlier, some of those edge cases where there's complicated data elements that need to be pulled from different data sets where the machine or our algorithms may not have the exact answer, those still get kicked out to humans for reviews. So, there's a place for them to still be doing their job, but they're going to be looking at the most complicated scenarios in a process versus the computer taking the easy, low-hanging fruit. And then over time, as that process and feedback loop gets a good amount of history behind it, some of those edge cases will become more routine. And they'll stop getting kicked out for manual review.
MJ 12:32 And for somebody who has already sold on the concept of automation and digital workers but just doesn't know where to start, I've been thinking about the way that we in most western societies have thought about outsourcing, starting kind of like the '80s and '90s, where we'd start to take certain tasks, break them down, and then move them offshore to lower-cost jurisdictions. And I sort of see the same thing happening right now.
How do we classify which tasks would be a good fit for a digital worker? And do we take a look at some of the low-value tasks or high-value tasks? How would you approach that?
MC 13:07 Yeah. So, I mean, you hit the nail right on the head. So many companies have already realized that low-value tasks can be sent offshore at a greatly reduced cost. I mean, there is some risk that comes with that, because you're sending your data overseas. I think the nice thing about automation is you can keep everything local. You can do it on-prem, or you can do it in the cloud, but you don't necessarily have to let it outside of the borders of the country that you're operating in.
What we see is a lot of companies have in-house teams that are doing these things already. They're spread pretty thin, and they're focused on the most high-value tasks. So, they're looking for that high return on investment. And they're maybe ignoring some of the low-hanging fruit that could be done quickly.
Our view is we can take some of those smaller, repetitive tasks, and we can rack up quick wins. So, if we're doing three, four, five of those, those may add up to one of those bigger focuses that the in-house teams are spending all their resources on. We're a little bit more nimble. So, we can do those things quickly.
In terms of identifying areas, it's really anywhere where you've got a large number of bodies, whether it is offshore or onshore, lots of FTE doing repetitive tasks and just working with you to map out what that process is, what the different systems of record that you're using are, the types of documents that are being reviewed, the data sets that are being pulled. And then we can quickly map out a projected roadmap for automation. And we may start with one little piece and then move through the entire thing.
MJ 14:54 Cool. And so, I see that there's a feasibility element to selecting the tasks that are best in terms of what is actually achievable from a digital worker versus a human. But I think probably one of the other things—especially as we're sitting [in] Q2 2022, its possible recession is coming down the line. As you mentioned already, employee costs have gone up quite a bit. It seems like one of the things that a lot of firms are seeking to do now, and most likely throughout the end of the year, is to focus on positive ROI tasks specifically, where they know that implementing this digital worker minus whatever human labor had previously done that job will result in net positive for the firm as a whole. Can you talk about some of the ways that you would think about measuring that and how you'd quantify that for people that are going down this journey?
MC 15:42 Yeah. So, our entire business model is built on providing positive ROI for our clients. And that is one of the ways that we're reducing the risk of going with us to take the automation journey. With the cost pressure that's already here—we talked a little bit about inflation, but now, we're layering on the possibility of a recession—there's definitely been some warning signs flashed in the past couple of weeks. The yield curve's inverted, which is a classic sign that it's coming. Lots of CEOs have been in the news talking about it. So, in general, we're at that point in the economic cycle where these things typically happen. It has been a while since the last one. I think the '08, '09 was really the big one. There was a little bit of contraction in 2020 due to the pandemic. But for the most part, we've been on a growth trajectory for an extended period of time. So many companies are starting to focus a lot more on cost reduction. And automation is a natural fit for this. So, you get the same or better productivity. You're reducing costs.
And how we measure it, it's one, very low-upfront cost, right? So sometimes we'll do a proof of concept for a nominal amount of money. If we think there is a repeatable process and a large market for it, we may do that for free depending on what the opportunity is. We will then work with the client to measure out, "This is what it currently costs you to operate this function." And we will put together a financial model, where we estimate the amount of that specific tasks that we can automate, what we believe the takeout is going to be, and then get agreement with the client on what those projected savings are. So, they can see the value before we even start the actual automation. And then as we go, we can measure it. So, if we put something in place where you're able to shift five or ten resources, that's real cost savings right there. They can go focus on something else, and you can free up the capital from that group to do other things.
MJ 17:55 And I know both you and I come from a consulting background, where typically the way that an engagement operates is by billing hours and providing services in exchange for those billable values. But in this situation, it sounds like what we're doing is often partnering with and aligning our incentives with the long-term incentives of the organization, and then probably coming up with some creative cost-sharing for that. Can you talk a little bit about why that arrangement makes sense and whether or not you see this as a larger macro trajectory of where consulting may be headed in this area?
MC 18:29 Yeah. So, it is very much a shared savings model. I think it aligns our incentives directly with our clients' incentives, right? Everyone wants to save some money. A lot of clients are in a mode where they don't really want to be spending capital without a return on investment. Technology is one of those places where everyone kind of wants it, but they don't really want to pay for it. So, when we were talking about strategy, it seemed like a natural fit for us. And it resonates with our clients, because they really have a very low-upfront cost, and they don't start paying until there is shared savings.
So, our belief is that this model is here to stay, at least for technology and automation. Whether or not the advisory side gets there, I could see that. I mean, for the past several years I have heard billable hours models have really suffered. Law firms, consulting firms, both have had reductions. Although I think there's been pockets of large amounts of work. But in general, I think if you're moving to this success fee-based model or shared savings, it aligns the incentives of both the client and the firm. And it's a really great way to do business. Everybody's happy.
MJ 19:54 Seems like there are a lot of positives. And the shared savings model makes a lot of sense, relatively low risk for clients to jump on board. What are some of the apprehensions or issues that are currently blocking clients from automating things that they're working on?
MC 20:11 Yeah. I mean, I've seen a bunch of different things. I think the biggest one is usually a client's IT department. So, they're oftentimes skeptical of heavily scrutinized third-party vendors that are touching their systems and their data. They don't necessarily want their data leaving their firewalls. So, there's a lot of hurdles to getting into a client just from that perspective.
Once you're past IT, you're generally okay. If you can start with the business and do the proof of concept and show the value first, that's always better. But it really depends on the organization that you're dealing with and what their protocols are for deploying this type of work. Then you have the combination of, "Hey, we already do this in-house." That's something we come up against all the time. I think the rebuttal on that is, "Yeah, but how much you are really getting from them? Are they focused on your particular area? If you want to do the engagement that we're talking about doing, are you on a wait list? That could be several months or several years. We can get going right now." And a lot of times, we will actually partner with those groups. So, we'll use the same tools, the same techniques. And once we're done, they can actually take it and run it themselves. So, this isn't a model where we're trying to capture subscription revenue into perpetuity like many software as a service providers do. We're just looking for a period of shared savings. And then if you want to take it and run with it yourself, you certainly can.
MJ 21:48 Another large macro trend that I feel we've seen some changes in recently has been around regulation. And I'm curious what your thoughts are, whether that be financial services or other largely regulated industries. Such a big burden comes along with compliance and making sure that firms are doing everything correct and that consumers are protected and everything like that—and there's so many positive benefits of regulation. But it feels like that's one area where as fast as you're trying to dig out from under the pile of paperwork, there's more being layered on top every single day. How do you see this current administration and the trends going into the future around regulation? And what impact would that have on the role of either hiring more people or automating certain tasks?
MC 22:35 Yeah. So, I've always thought that compliance was the number one area where automation should be deployed. No company wants to spend a lot of money on compliance. They all kind of do it because they have to. And there's varying degrees of it, right? Some companies are really hyper-focused on being compliant. Others are business first, compliant second.
I think with the current administration's view, it was really going to accelerate, especially from a CFPB [Consumer Financial Protection Bureau] enforcement perspective. It's been a little bit slower than I think a lot of people expected, but that doesn't mean that there's not a lot of activity going on. So, there may not be a large number of enforcement fines or consent orders, but there are a lot of exams going on. There's requests for information all the time. And there's definitely quite a few clients that I've heard of that are well on their way to consent orders, which will probably start to hit mid- to late summer this year, early next year. And we'll start to see that ramp. Again, as you touched on, there has been regulatory change. The CARES Act that came out during the pandemic caused lots of different issues, especially in the mortgage servicing world and the credit reporting world. There's a ton of stuff coming down the pike on crypto. So, it's only going to continue to get more complicated.
And I think truly the only way to be compliant in a way where it's not destroying your company's bottom line, it's to deploy technology to help you do this, right? So, the traditional three lines of defense model is: heavy testing in your business line, which is the first line. Then it's compliant, second line, coming and looking over their shoulder and sampling things. And then you've got internal audit coming behind both of you and checking out everything that's been done. And a lot of times, they're also hiring advisors or third parties to come in and kick the tires too, which all of that adds up to a lot of money.
And they haven't really been that effective. So, there's certainly companies that do it well, but then there's lots of banks that have spent a fortune on advisers, and they still have problems or they're getting hit up by the CFPB again for not complying. The OCC [Office of the Comptroller of the Currency], same thing.
So, I think deploying technology in those areas is really one of the only ways it's going to get better. And that's for multiple reasons, right? So one is that cost factor. You can do it—you can do more with less. You can reduce your risk, because you're going from a sampling model where you're only looking at a handful of things—you're searching for a needle in a haystack, essentially—to doing 100 percent of a population. So, you're really going to know what all your issues are.
And then, if you do the automation in the right way, not just having data analytics that tell you where your problems are and creating a ton more work—but if you set up the automation to actually remediate those things, then I think you have a really nice solution.
MJ 25:59 I know we started this discussion with some pretty staggering statistics about what the future might look like in terms of automation in the workplace. We've covered a ton of different topics throughout this conversation. I was just wondering if you were to opine and think forward five years, ten years, how do you see all these pieces coming together? And what advice would you be giving to any clients who are listening to this right now?
MC 26:23 Five to ten years from now, I think the workplace is going to be dramatically different. I think the remote work aspect of things has really changed the way humans want to interact with work. I think it's changed the way employers view their teams. I think that there is a desire for human beings to not be doing repetitive, menial, boring tasks as much as we've done in the past, because we now have all experienced what having a little bit more free time or having creativity to do other things—how that excites people.
I do think that we're going to see a much bigger shift to being more creative at work and doing less of these types of tests. I think the repetitive tests that we've discussed will be fully automated in the next, call it five to fifteen years. But that's really going to challenge—for people who cannot reinvent themselves or be creative, how do we keep them busy? Or how do we make them productive? I think that's one thing that the world really hasn't figured out yet. And that's going to be a really big challenge.
MJ 27:42 That's awesome. I completely agree. I see us being more creative, taking on the harder, more difficult tasks, and delegating the things that are straightforward to machine, so you spend more time solving problems, not executing on the ones that already exist.
This has been incredible, Mike. It's been so nice to chat with you. And as always, I'm learning so much. If people want to learn more about the Digital Workforce offering, what's the best way for them to read about it or get in touch with you?
MC 28:08 Yeah. So, we've done a couple of blogs. We've done lots of thought leadership. So, there's different pieces you can read about, but I think the best way to do it is to contact myself or you directly. And we can start talking about your particular use case and whether or not we have something that already fits, or if we want to go down that proof-of-concept journey together. But it's as easy as a thirty-minutes discussion with us.
MJ 28:37 Perfect. Well, thank you again so much. This has been great. Can't wait for the next one. I think this is such an exciting space. And it's been a pleasure working with you building out solutions here.
The views and opinions expressed in this podcast are those of the participants and do not necessarily reflect the opinion, position, or policy of Berkeley Research Group or its other employees and affiliates.