episode 203 – machine learning turns advisory relationships upside down



Doug: From Rea & Associates Studio, this is unsuitable, a management financial services podcast for entrepreneurs, tenured business leaders and others. We’re ready to look beyond the suit and tie culture for meaningful, measurable results. I’m Doug Houser.

Whether you know it or not, machine learning has changed the way we do business and it will continue to shake up the business community for many years to come. That being said, you may not be fully aware of the changes that are already taking place and how they might impact your future success. That’s where today’s guest comes in. John Dages, Solutions Director with Fusion Alliance, will explain what machine learning is, how it works, and how it can empower you to drive long-term growth and sustainability.

Welcome, John.

John: I’m happy to be here. Good morning, Doug.

Doug: It’s been awhile since we’ve been together.

John: It’s been three, four months. I know.

Doug: I know.

John: And we’ve actually had some experience together talking about machine learning in a professional setting.

Doug: We have, and I’ve always enjoyed your work and experience and when we get together for a beer as well, that’s always very entertaining.

John: I certainly appreciate and I appreciate the invitation to participate. I’m a big fan of the podcast.

Doug: Good. Glad to hear. Glad to hear. So machine learning, that sounds, if I’m a business owner, that sounds a little scary or a little big brotherish, maybe above my pay grade so to speak.

John: Sure.

Doug: What’s happened to the cost of machine learning, the cost of computing, how does that measure up for, say somebody who owns a 25 or $50 million a year business now?

John: It’s a very good question, Doug:. And I think if there’s one thing that’s complex about machine learning, it’s actually defining what it is. Right?

Doug: Okay.

John: So we’re having conversations, it can go anywhere from self-driving cars to all the crazy things we do on Facebook, to robotic surgery. Very little about that is what we do in a professional setting when it comes to machine learning. So is it valuable Doug, if I start kind of what machine learning is?

Doug: Sure. Absolutely.

John: So at its core it’s actually fairly simple. So the idea of machine learning is that we can train or teach a computer or a system to learn from data and make up prediction about something in the world. Right? So what that means is, if I have information about professional customers, economic indicators, if I have some form of data I can look at that data, apply machine learning and then use that data to make some kind of prediction in the future from the patterns and lessons learned inside machine learning. Am I making sense?

Doug: Yeah. So you’re looking for patterns in the data.

John: That’s right.

Doug: But you’re able to factor in so much data more so than you or I could individually.

John: Sure.

Doug: Correct?

John: Yes. And here is what it… So let me compare this to something a little bit different. So I imagine most of the fans of your podcast are in the financial services spaces, accounting, entrepreneurs and things like that.

Doug: Sure. Business owners. Right.

John: Business owners sound good, right?

Doug: Yeah.

John: So let me compare something to maybe a problem we would solve a little bit differently today than we would with machine learning than we did maybe 20, 30 years ago. Right?

Doug: Okay.

John: So long story short, let’s say you’re a bank, and you’re going to issue a, let’s call it an auto loan for some piece of business. Right?

Doug: Right.

John: Now this is not new business for retail bank, a credit union, what have you. There are a variety of different algorithms, patterns, researching third party data that says, “Hey, if I apply these formulas, I can understand what the risk is for this given bank loan customer.”

Doug: Sure.

John: And what machine learning says, as opposed to a somewhat biased, very technical on statistical approach to evaluate a customer for a bank loan. What machine learning does instead is we’re going to look over my banks’ history of 50 years of lending data, and find out those little bit of patterns that you would never be able to find out if it’s human-driven. Right?

Doug: Okay.

John: And as part of that, and machine learning can look at a history of similar customers that’s making the bank loan where he lives, what his income is, what his interaction is with your banking brand and so forth. And as part of that can actually say, “We think that this is at this level of risk as opposed to deploying some kind of algorithm and being a better result.”

Doug: Sure. Now devil’s advocate. What about those who say, “Well, you’re taking the human element out of it.” So I look at that as a good thing because you’re removing bias, correct?

John: That’s right.

Doug: But there’s those that say, “Well, no, I still want that human element in the decision making.”

John: So, and that’s kind of more an advanced concept, we’ll say.

Doug: Sure.

John: So much about what machine learning does and the advantage it’ll give to business owners and things like that, is it can better inform decision-making.

Doug: Mm-hmm (affirmative).

John: So what we’ve done historically, and in fact Doug, I think we’ve had this history together-

Doug: Right.

John: Is we will produce a machine learning model, which is the artifact of this process that says, given this amount of data, we have an 87% accuracy that in my previous example, this loan is going to be paid back and so forth. But nobody believes it, right?

Doug: Right.

John: So when I say that we show these statistics, everybody nods in the conference room, but like, “Well, what does that even mean to me? There’s no way I’ve been doing this the same way for 45 years. I’ve written 50,000 loans in my career. Why would I trust this arbitrary system?” Right?

Doug: Sure.

John: So your question, what’s the human element? The human element is actually the biggest… How do I put it? The biggest thing that’s holding us back in machine learning is that there is that human element, that idea that I have to understand everything-

Doug: Right.

John: And some of these ML algorithms are not going to be deployed. Am I making sense?

Doug: Yeah. So, inherently we fall back on our own biases and want to fit that into whatever the answer is.

John: It’s understandable.

Doug: It’s there. Okay. What about artificial intelligence? Now how does that differ? I’ve got clients for example, that are out on job sites, large construction projects, they’re now using wearables to track where their folks are, not to police them, but just to try to get more efficient with the movements on job site, materials, where they’re delivered, tracking how and where the materials move.

John: Mm-hmm (affirmative).

Doug: So they try to become more efficient in terms of where those materials are delivered and placed. Is that a part of it too or is that a different aspect of…

John: So it absolutely is. And just for clarification sake, so when we say machine learning, machine learning is actually a subset of artificial intelligence. Right?

Doug: Okay.

John: So to use your example, and I’m actually partnered with a company here in Columbus that does analytics machine learning on top of construction sites around demand planning, which is how much raw equipment am I going to need to order, at what time of the year, when do I actually make purchases? And from a procurement standpoint, when the prices are low, try and maximize those dollars. But that wasn’t your question. Your question is, is that machine learning? The answer is absolutely.

Doug: Okay.

John: So if we use your example in a job site, we’re going to have some IoT device or some remote device that is transmitting my activity, where I am. What we’re actually doing is something called training a machine learning model.

Doug: Okay.

John: So we can say is, “Hey, this is how these all interacted. This was the outcome.” Again, I can find patterns, another customer that a employee’s behavior and try and optimize to the output. Right?

Doug: Right.

John: So as part of that, I get that model. I can actually make better projections in how I actually allocate my staff, how to best use their time, how to best use my resources and get to a better outcome.

Doug: Okay, that’s fantastic. I mean, it sounds like if I really buy in and think about what I can do in terms of key performance indicators in my business, I mean, the possibilities literally are endless if I think about it.

John: That’s right. And I’ll give you one more interesting anecdote. So another partner, another firm that’s here in Columbus is actually using something very similar to what you described around safety.

Doug: Okay.

John: So when I say safety, what they’re doing, and this isn’t specifically around the machine learning we were talking about earlier, but we’re actually doing something called visual analytics.

Doug: Mm-hmm (affirmative).

John: So what does that mean? So that means they have cameras deployed on factory sites, on construction sites, things like that. That are in essence taking continual pictures of the staff that are out there, which I know sounds creepy.

Doug: Right.

John: There’s always a component of that for machine learning.

Doug: Right.

John: But the concept is, there’s certain things you have to do to remain safe on those job sites.

Doug: Sure.

John: It’s perhaps the most obvious is that you have to wear a hardhat. So what we do as part of that, what do we do, what machine learning is doing, is looking at a picture of all the construction staff that are out there and it can identify that, “Oh no, this person’s out of compliance. He doesn’t have the right boots. He’s not wearing the hard hat.” Send a notification over to the foreman to say, “Hey, there’s a safety issue here. Let’s make everybody safer and take an action.”

Doug: Wow. Yeah, that’s just fascinating. I mean, and I think about how it applies to our business too. And we’ve had some, certainly a lot of discussion about this as an accounting firm, if you think about our audit services for example, what we’ve historically done is take samples of the client’s data to determine if they’re recognizing things properly, et cetera, et cetera. Well, from a machine learning perspective, that really takes the need for that away, because the machine can literally test everything.

John: That’s right.

Doug: Right? And become much more detailed in terms of what it can provide us for that data. So what we’ve tried to do is think about our client’s business as a whole and shift the focus to, “Okay, now that we’ve got machine learning,” and what I’m trying to get to is if, “Okay, if I implement some type of machine learning within just perhaps a few facets of my business that allows me then to focus on things which are more valuable.”

John: That’s right.

Doug: I’m not talking about getting rid of someone’s job. Has that been your experience?

John: Well, it’s been my experience. But even more to that point, when I think, people that are listening to your podcast and all the folks that we’ve worked with in a professional capacity, when they say, “Hey, we want to do machine learning,” we really swing for the fences. Right?

Doug: Yeah.

John: Can we automate customer engagement, can we do long-term forecasting, management, product line, everything that’s actually fairly complex.

Doug: Right.

John: And these projects take a long period of time, they cost millions of dollars. What I kind of see kind of to your question is a lot of people aren’t taking advantages of some of what we call the quick wins for machine learning. So I’ll give you a couple of examples.

Doug: Sure.

John: So perhaps the number one thing that we’re talking about here and in our market for the last couple of years, is how do I better engage with my customers, my clients, what have you. Now the quintessential quick win when it comes to this machine learning component is can I actually look at my portfolio, my customer list and say, “You know, I think this person might be an attrition risk.” Right?

Doug: Right.

John: “He’s performing certain activities. I would never be able to detect that as a human being or an algorithm.” Right?

Doug: Yeah.

John: But based on the patterns that people that have left my organization, should I be concerned about this person? So looking at that as a relatively straight forward machine learning problem.

Doug: Sure.

John: When we start getting into, again, visual analytics, again to self-driving cars and things like that, that’s a little more advanced. Right?

Doug: Yeah.

John: But maybe I can look at a customer and say, “You know, I think this person is likely to leave.” Or I can say, “Hey, how much is the business I’m doing with this person a risk,” or from an audit standpoint, “Is there a pattern I see with this client.” It is very similar to some other organizations that I’ve run into problems with audits. Right?

Doug: Exactly.

John: Either just bad paperwork, malfeasance, what have you.

Doug: Yeah. And the other opportunity is, from our perspective you want to look at it and say, “Okay, well, we’ve experienced other clients that have these characteristics and these are the positive things that have happened too.” So you can get ahead of opportunities and help them prepare in their business.

John: There’s a great book and I can’t remember the name. I hope we can put it in the show notes. And it was actually a machine learning analysis on the Enron crisis. Right?

Doug: Okay. Interesting.

John: So the question becomes, and this is certainly of interest to you, how did it go this wrong? Like how do we get this far? And nobody saw what was going on. Well anyway, so, and God bless me, I can’t recall the name of the book. I’m sure we’ll put it in the notes there. But the idea is, well, let’s actually analyze the email correspondence going between in the internal executive team, their third party auditors and so on and so forth. And what’s interesting is what he publishes is that everybody knew it, whether they didn’t know it or not, that things were in trouble. Right?

Doug: Mm-hmm (affirmative).

John: The syntax has changed, the level of correspondence has changed, the detail and the accounting has changed and things like that-

Doug: Interesting.

John: Which is a perfect machine learning problem on the audit side.

Doug: Sure.

John: It says, “Boy, this is very similar to some other… Companies have gone the direction that they would get in trouble. Better take a look at this now.” Especially as a third party looking in. Am I making sense, Doug?

Doug: No, absolutely. If I’m thinking about it from a client perspective, that’s the kind of thing I want to know. I want to be prepared. I want to get ahead of what’s coming down the road and use it as predictability and things like that.

John: And it’s exact. But in many ways, the audits you’re performing and your firm is performing now, it’s very similar to the way it has been for the last 70 years.

Doug: Right.

John: This is the formulas and they’re complex and they require a high degree of education and all those pieces. Right?

Doug: Sure.

John: But wouldn’t it be more interesting if I could look at 75 years of audit data for 5 million companies, found out where they went. Did they go the bad route or they’d go the nice and clean healthy balance sheets.

Doug: Exactly.

John: And if we start to see those relationships, those scored relationships, maybe we spend a little more time with those folks.

Doug: Exactly.

John: Before you go back though, I do have the note or the name of the book. It is called Power Failure, which is the inside story of collapse of Enron by Mimi Schwartz. Thank you.

Doug: Yes. Great. I will certainly give that a read, but yeah. No, I mean, I think this is fascinating stuff because for me, you want to be able to adapt. You don’t want to just continue to do the same things over and over, so let’s take the knowledge that’s there and figure out how we can approach things a little differently and a little bit better.

John: That’s right.

Doug: And help our customers. That’s what we’re all in it for.

John: That’s right.

Doug: We all take some satisfaction in terms of helping those clients achieve their goals.

John: That’s right.

Doug: I mean, any of us in a service business. Right?

John: That’s exactly right. Which we both are. What’s interesting too as part of that concept, from the machine learning side too, which we’ve always found surprising, I think we found it surprised when we were working together, Doug, that there will always be a surprise. Right? And again, I’m not saying this in a negative way, a lot of our partners and clients that have gone through this process from nowhere to pretty sophisticated centers of excellence and advanced analytics, is that every time we were running these programs, it’s a complete surprise.

So all these folks that have been in the industry for 30, 40 years, executives at Fortune 500, people we know and you’ll say, we know what makes our customers happy.

Doug: Sure.

John: Or we can understand what risk is or we know how to do this. We’re the experts. I went to MIT, I know the right answer. And then they expect certain KPIs of how they judge the customers when we put it against the machine learning model, had nothing to do with all that research and the millions of dollars of investment. It was something completely different. So every time we do one of these things, there’s always going to be a surprise. That’s an impact. That’s an indicator. That’s a pattern that a human, again would never see.

Doug: Well, in the minute that I hear any client or business talk about the fact that, “Oh, we know most of it,” or, “we know all of it,” that’s-

John: At least a yellow flag.

Doug: Yeah. That’s always the moment where I’m thinking, “You know, we all learn something everyday.” I mean, there’s just so many different businesses out there. You can’t possibly account for all the variables. And that’s what machine learning helps you do.

John: That’s exactly right.

Doug: So now let’s talk about costs a little bit because I hear this stuff and if I’m a business owner, I think, “Boy, that sounds really cool, but that sounds expensive.”

John: Yes.

Doug: So talk about the cost of computing and machine learning and that type of thing.

John: I surely will. So just a quick background to get to your question too. So I think we’ve talked about this historically. But what’s interesting about machine learning is these algorithms that have been out there since World War II, which means some of the statistics you took in seventh grade just applied at a massive scale, like a big data scale.

Doug: Okay.

John: A little over-simplifying, but there’s nothing new that are about these algorithms. What is interesting is that things have changed from a technology perspective over the last 10 to 15 years.

Doug: Sure.

John: So of those things we’ve just got more data, we’ve got better data, we’ve got the cloud, which means these machine learning models requires some computation to get there. But once you have a model, they’re relatively straightforward to use. So having the cloud there that hey, I can go borrow 50 servers for an hour, get that model computed, then come back, it’s a huge advantage.

But to your question, is it getting expensive? So if the question was asked maybe 10 or 12 years ago, it would take a high degree of sophistication to build your own model.

Doug: Okay.

John: A lot of data science, a lot of complex math, cutting edge technology, right? That just isn’t the case anymore. So for what it’s worth, and again, as your client or your podcast listeners start to begin their journey on machine learning, there is no reason, although many will try to actually generate your own machine learning algorithms, custom neural networks, there’s just no reason for it. So in other words, now looking at more democratized products, whether they’re from AWS or other cloud providers, or there’s package software you can buy or open software you can buy, there’s very little data science to getting those initial models.

Doug: Okay.

John: What we can say is, “Hey listen, here’s the data, here’s what I’m trying to predict on, run it through that model, through a process called auto machine learning and it’ll generate the outcomes for you.”

Doug: Okay.

John: Now that’s the 95%. once you get there, you have seen a lot of value.

Doug: Right.

John: After that’s done, of course you can do the bigger investment in the data science side, try and really tune these machine learning models and things like that.

Doug: Sure.

John: But for what it’s worth, the big win is just in that first phase, right?

Doug: Sure.

John: The good news, like I say too, there’s a lot of products you can simply buy. So if you’re using Salesforce, as I’m sure a lot of your clients are, if you’re using Dynamics, almost any of these big package software products are probably using machine learning, whether you know it or not. Right?

Doug: Interesting.

John: And since the big one is Salesforce, I’ll use an example there. So with Salesforce, Einstein, all of their effort, as kind of in that piece, which is… Pardon me. So this customer relationship management software, sales software.

Doug: Yeah, absolutely.

John: Yeah. So with that in mind, what’s the right customer I should be talking to? Right?

Doug: Right.

John: Whether it’s, I think they’re in a market right now for the product I’m selling, maybe something’s changed with their relationship with your company that they’d be available for a targeted message.

Doug: Right.

John: That process is machine learning. Again, a human can predict when you’re going to be in a position to buy whatever my product or service is. So all these Salesforce components, as an option you can actually look and say, “Hey, this is exactly what your customer wants. This is exactly when he’s going to need it. And quite frankly, this is what he’s willing to pay.”

Doug: Yeah.

John: And when we talk about engaging customers a little bit better and knowing your customers a little bit more, if you get out the idea that at some statistical model you can actually use machine learning to say, “Hey, this is what my customer wants, I’ll go ahead and talk to them at this point and not burden him with some over solicitation.” Am I making sense, Doug?

Doug: Yeah, absolutely. So you’re better able to profile those opportunities. And I think of any business, the best data you have are on your existing customers.

John: That’s right.

Doug: Just in clients. So why not mind that data first and identify patterns over time, not only with your own business but others.

John: That’s right.

Doug: And figure out, okay, this is the best opportunity to get ahead of it for them. And again, better serve them and have the appropriate conversation and not go into product dump about things they don’t really care about.

John: That’s exactly right. And even taking that a little bit further, so we’re recording this podcast on September 11. So last week Apple actually announced their own retail credit card, right?

Doug: Right.

John: And we think about, “Well, is that really going to be successful?” Chances are that’s going to be a pretty big value offering to the customer. So let’s say, I mean, we detach from that example a little bit. We’re like, “How am I, if I own a medium, small, medium sized business, how am I going to compete with these conglomerates? How am I going to compete against Amazon? How am I going to compete with all the Silicon Valley companies,” and things like that. I mean those are the guys that are doing this every day. They’ve got resources I don’t have. How do I remain competitive?

John: Right.

John: Well, the reality is the one… I don’t say the one. A big advantage I have as an incumbent company is that I sit on tons of relationships, customer relationships, vendor relationships. I’ve got 70 years. That’s something that those quote unquote digital natives, those new companies, they just don’t have.

Doug: Right.

John: So you are sitting on a gold mine. And a great thing too is everybody says, “Hey, data is the most valuable thing I own as a company.” Nobody knows what to do with it. That’s the answer. It’s like, let’s find out what those patterns are. That’s really get to know our customers because nobody else doesn’t have the relations to be able to do that.

Doug: Right. Because you and I, as you said, we’ve worked together in the past, we’ve already got that trust and credibility. So if you come to me with those solutions rather than Amazon or whatever, I’m going to be much more willing to say, “Hey, John and I, we’re going to have a beer together.” So this is great stuff. So that’s awesome.

John: Appreciate that.

Doug: Speaking of beer, you had any good new ones lately?

John: This is a true story. I finally tried Pappy on the Bourbon side for the very first time in my entire life.

Doug: Where did you score that? That’s a hard find.

John: I was actually in Louisville. And what makes me very sophisticated in my job is I actually find a way to put that in the expense account.

Doug: Uh oh.

John: So I’m just putting that out there. That hasn’t been approved yet. So this might be..

Doug: Are we going to have to get Rick?

John: That’s exactly right. I have not had anything new. Oh, I’m lying to you. I did have out of Atlanta… Oh my God. I’m not going to come up with it. Crap! I’m going to turn it around to you.

Doug: It was crap?

John: No, it’s just terrific. And like before, I’m going to think of it the moment you start talking. But what have you got on your list? What’s your recommendation?

Doug: I get so tired of seeing all these IPAs-

John: I love it. I can’t get enough.

Doug: The Pumpkin beers already. So I’m still drinking the Summer Brew, the Saison style is where I’m at right now.

John: Now you got to go big or go home, Doug.

Doug: All right.

John: Sorry. So IPA, Double IPA if you have it, if you can get worse than that. Sounds good to me.

Doug: I’ll need to walk home.

John: That’s what Uber’s for.

Doug: Yes. Well, thank you very much John. And if you want to hear more tips and insight or to hear previous episodes of unsuitable, visit www.raecpa.com/podcast. Thanks for listening. You can subscribe to Unsuitable on iTunes or wherever you might like to get your podcasts, including YouTube. I’m Doug: Houser. Join us next week for another Unsuitable interview from an industry professional.

Disclaimer: The views expressed on unsuitable on Rea Radio are our own and do not necessarily reflect the views of Rea and Associates. The podcast is for informational and educational purposes only and is not intended to replace the professional advice you would receive elsewhere. Consult with a trusted advisor about your unique situation so they can expertly guide you to the best solution for your specific circumstance.


stay engaged

If you liked this episode of unsuitable on Rea Radio, let us know! Like it and share it with your professional network. You can also use #ReaRadio to join the conversation on Facebook and Twitter. You can also watch the podcast in action on the Rea & Associates YouTube channel.

John Dages | Machine Learning | Ohio Business Podcast