You can’t not change: How IA is transforming

Well, good afternoon, everybody. In a twist of scheduling fate, just to perk you up at the end of the day we decided to save the best for last, Tax. Those of you who are sitting in the back, you may want to rush to the stage to hear the actor talk about tax here, but in all seriousness, it's a very interesting time to be a tax professional. And I know that that kind of sounds like the opening line to a bad joke or something that kind of wants you to head out the door, but it really is true. And I'd like to walk you through a story today to prove to you that that is true.

(00:43):

I think we've all experienced great change that's going on in the world today. And within tax we have some historic things that are beginning to converge with a number of new factors to really transform what it means to be a tax professional and what the role of the tax function is within most companies. And we're seeing economic pressures, tax reform, changing roles and responsibilities between tax authorities and tax payers, where the tax authority is calculating the tax instead of the taxpayer. We're seeing people enter the labor market that don't want to do data tasks anymore. And we're seeing organizational leaders asking different things of tax departments. Changing from just filing returns and making contributions to financial statements to converting to value and insight in real-time business alignment.

(01:34):

And so at the center of all of this is a whole host of very complex technologies that for a tax professional are hard to understand what some of these things are, much less how they can change what it is the tax department does or what the role of the tax professional is. And so I'd like to tell you a little bit of a story of our journey here at KPMG with innovation. And hopefully get you to think about some questions along the way and reflect on some of our insights. And so that story begins back in time, about 30 years ago where you see the images today. I didn't have as much gray hair and wrinkles back then, but I was a young professional and there were a lot of things different about that period.

(02:19):

People had landline telephones, they had cassette decks and CDs, instead of streaming music. They used VCRs instead of things like Netflix and their smartphones. We even had traders, humans on the floor of stock exchange instead of algorithms locked in computers somewhere. But like today, there was something very similar. It was about to be transformed by the introduction of a new technology. For the first time ever, the computer was being released from big box mainframes or remote facilities at far away locations and put on desktops of professionals everywhere. And as you can imagine, this had a transformative effect on what it meant to be a tax professional and a business professional in general on this.

(03:08):

And looking back from today it may be hard to imagine though that to a tax professional, this wasn't necessarily greeted with a huge amount of enthusiasm. And so if you think what it was meant to be a tax professional back then, well, you had your 10-key, adding machine that spit out some tape and you had these large spreadsheet, paper spreadsheets that organized all your numbers and grids. You have these great libraries full of tax books, code regs, cases, all sorts of information. You had a typewriter. You had a telephone to talk to people and communicate with. What more can you need?

(03:45):

And so to a tax professional you'd see this computer coming in like “ah, I don't know. I'm pretty good at what I do.” And to that person, that was a rational position to take. If you were pretty good at your job as a tax professional back then and you couldn't see what the computer was going to do, what we see today, you may say, “hmm, I don't know. This isn't really relevant to me. I don't think I want to do it.” They felt that they had a choice. But from the perspective of today, can any of us imagine going through a day without a computer in our lives? Much less if you're like me walking out the door without a smartphone or some sort of computer device on your body? There's a panic. But to them, they thought they had a choice. And so that's the question that I'd like you to contemplate as I go through the rest of the story. How can we use the computer or do we need to use the computer to transform the way we do business? And if so, how do we get that process started?

(04:49):

So a group of people at my firm about this time who were inspired by the introduction of the computer convinced our management committee to fund an artificial intelligence pilot in the early 1990s –a long time ago. And what they wanted to see is could computers help us, aid us in the analysis and application of tax law to complex business problem? And they chose an easy one here, mergers and acquisition transactions. And so what they did is they hired some outside computer consultants and got together with some of our tax SMEs and they began to document the questions that you would have to answer to determine the taxability of these transactions. To develop the logical foundation for an algorithm to predict outcomes.

(05:39):

And if you know anything about tax, tax usually starts with an easy question, a simple question, is it taxable or not? And leads to a few more questions and a few more questions, and some of those answers to those questions aren't always entirely clear. And so you have to ask more questions. And so this flow chart, this diagram got more and more and more and more complicated. To the point where we had to pause and say, “You know, can we do this? Is this something that a computer is capable of doing? Can a computer analyze and apply tax law like a human can?” And after much thought and discussion, we concluded that “no, it couldn't”. That this was just beyond the ability of computers and that it's something that was best left for humans.

(06:30):

And looking back at this time, we know that that was wrong. We know that this was wrong. But at the time again, people who were enthusiastic about computers even underestimated the ability of computers. And therefore didn't see the role that they could play in defining who we were as tax professionals. We found out some interesting things along the way, though, in trying to find out some insights of how do we change this? We found out that we weren't the only ones who got this wrong. In fact, a lot of people, some very qualified people, computer scientists, software engineers, academics, other business people –all underestimated what computers could do and the role they could play in our lives. And this is a real early example of this. This is a professor from MIT that wrote a book, a computer scientist specialist wrote a book analyzing the limitations of computers and specifically artificial intelligence.

(07:27):

And in it, he looked at a number of things, but he predicted that computers would never be able to translate languages because of a number of challenges, contextual and syntax issues and even going so far as voice recognition and all of that. And yet we know today that language translation is kind of an off-the-shelf commodity. You go into Google Translate, there's a 100 different apps that can help you with this. And so he got it wrong. And obviously here's an example of this. This is an ancient Egyptian proverb written in Arabic and I know this not because I read Arabic or frankly because I know anything about ancient Egyptian proverbs, but because of language translation capabilities that are available today. “Translate this quote into English.”

(08:24):

At the time of the test a person rises or falls. And so this supposed expert and probably expert in the area, got it wrong. Underestimated what they could do. 45 years ago, so maybe it's makes sense, but fast-forward to 2004, this is another example. Again, two academics, one from MIT and joined by one from Harvard, analyzing the effects of computerization on different aspects of the U.S. labor market. This time people were kind of a little bit more enthusiastic about what computers would do. And in it they looked at heavy industry and if you don't know, one of the main jobs in heavy industry is operating vehicles, trucks, tractors, cranes, that sort of stuff. And they said that computerization would have very little effect on heavy industry because computers would never be able to operate vehicles like humans would. That the number of real-time variables in operating a vehicle would put it always beyond the ability of a computer to do.

(09:24):

2004. Six years later Google announces the launch of its autonomous pilot fleet of vehicles. 2017 through today, every vehicle rolling off of Tesla's manufacturing line comes equipped with all of the hardware and much of the software necessary to be a driverless vehicle. A lot of people got it wrong, including us at the time. And we knew that if we could figure this out, if we could break through why humans don't see the potential of computers, we could better understand how we could get our own people to understand what computers could do for them in sort of redefining our industry and our future and frankly to help advise clients to do the same.

(10:06):

And so in looking at that, we came across some interesting facts and so maybe a story is helpful here. This is Gordon Moore, and if you don't know who Gordon Moore is, he's known for two things primarily. He's the co-founder of Intel, the chip maker, and he's also very famous for an observation that he made back in the mid-1960s that is widely regarded as a measure of progress in computer technology. Specifically, he observed that the number of transistors in a square inch of an integrated circuit doubles approximately every two years. And this doubling in kind of layperson's terms is essentially saying that computer processing capacity doubles every two years. And translating that a little bit more, it says that computer technology or the capabilities in general of computers, generally speaking, it doubles about every two years.

(10:59):

And you may have heard of this before, it's known as Moore's Law. And there are some debate including from Gordon Moore himself about the limitations of Moore's Law, but it generally is still true today in observation on this. And this really kind of gave us some insight. We thought, “okay, if we have a measure of how fast computers are moving on this, what if we compare that to how we perceive how fast computers are advancing?” And researchers have looked at this and it turns out that humans don't see computers advancing at how Moore's Law describes it, but rather sort of consistent over time. What you've experienced a year ago, you expect to see again in the next year and so on and so on. And some studies even show that humans think that computer technology is sort of slowing down. That the development is actually tapering off a little bit.

(11:52):

But what Moore's Law tells us is something very different. That it's actually exponential. And you see over here on the left side, this is from mid-1960s over to today, that there's this doubling effect that slowly and slowly and slowly advances until you get to about 2000 there over on the right and there's this big inflection point now doubling just causes computer technology to shoot straight up at almost an incomprehensible rate on that. And so when we started thinking, “why is it that people don't appreciate what computers can do? Why don't they understand how computers could redefine things?” We thought this is probably it. This is probably it here. But looking at a graph, it probably doesn't provide you that much insight. So maybe an analogy here would be helpful to kind of illustrate how much how fast computer technology is moving.

(12:47):

If we were to compare advances in computer technology, say to another technology, we take a look at transportation and compare the beginning from walking, to the other extreme, say jet travel on this. And so researchers have made this comparison and at least in terms of walking to commercial jet travel, they've looked at it and it turns out that commercial jet travel is about a hundred times as fast as walking on this. And that may feel a little bit low on that, but it's factoring in a number of things that when you go to the airport you have got to go through security and deal with your luggage and all this stuff that I'm going to have to deal with it this evening on this, but it's 100 times as fast. It's a pretty big gap on that when you compare it to walking.

(13:32):

Well, when we look at the advances in transportation compared to computers, we're not talking a factor of 100. Over the last 50 years, we're talking a factor of a million, a million. And that's not static. It keeps doubling and doubling and doubling and doubling. And it was this comparison that really gave us some insight that really has sort of formed how we approach technology today for tax and business in general on this. That the limitations of computers, they're not in there. They're not in your phone. They're up here. They're in us. They're in our ability to imagine, to create, to design, to implement and most importantly, to use. And so you may think to yourself, “okay, Moore's Law, an interesting story, but what does this mean to me? It's the end of the day. Give me something real on this.”

(14:36):

Well, so I told you back in the early 1990s that we weren't able to apply intelligent automation to the tax function. Well today that certainly isn't true. In fact, we've developed a number of different technologies and continued to develop day after day, week after week, different applications that are leveraging technologies across the spectrum that you see here. From RPA to natural language processing, machine learning, cognitive automation. All of which are transforming those important things. What it means to be a tax professional, what a tax professional is capable of doing, and what a tax department can do within their organization. To respond to those market forces and respond to those changing expectations around creating value and aligning with the business function.

(15:31):

So I want to give you, you know you've probably seen a lot of demos of these types of things. So I don't want to kill you on demos here, but I do want to give you a few examples to make it real. Some of the things that we've seen a lot of companies embrace is robotic process automation. It's rules-based automation, you tell a computer what to do. I want you to do steps one, two, three, four and five. The computer does steps one, two, three, four and five. Whether they're right or wrong, good or bad, computer just listens to you. And you can imagine that has a lot of application in the repetitive processes that were then the tax function and this is probably this example goes beyond tax.

(16:10):

This is the end of the tax compliance process where you have to assemble your records together. The returns, the work papers, the schedules, assemble them into like a complete record and load it up into a file management system. It's a little assembly line of mouse clicks. It takes a little bit of time sometimes. This one probably about 25 minutes for a professional. It's something that's absolutely necessary, but probably doesn't create a lot of value and for somebody with an advanced degree, maybe not something that they really want to do, right? And so, it's a great target for a bot, for Robotic Process Automation. It's kind of boring to look at, it’s just basically like what you're doing sitting at your computer, just done very quickly. And so it enhances and augments what we can do.

(16:59):

There's some more advanced functions as well in the tax function that begin to expand the capabilities and extend sort of the thinking processes that we have also through natural language processing and machine learning. The ability to interpret and understand language as humans would –not just the words but the context and the meaning. And to be able to learn how to make decisions and choices like humans would as well. And here's an example of both of those things put to action. This is a K-1, a partner's distributive share partner income and so I'm sure most of you know that it's also used as a source document for preparing tax returns.

(17:42):

And so for a computer, this is a little bit more difficult to take information from this than a human. A human would just read it, understand the language, say, “hey, are these items relevant to the tax return?” If yes, where do they go on the tax return? To a computer, you need things like natural language processing to read through these words and understand their meaning in context of preparing a tax return. And then once they make that determination of what it is the choice, using machine learning, of does it go on the tax return? And the choice of, if so, where does it go on the tax return?

(18:24):

That's extending the intelligence and the thought processes of our teams to transform what we do to shift this to maybe the more value creating activities. And finally, cognitive automation, to bring it all together, both of the capabilities that I talked about and several more that allow computers to ingest very large amounts of diverse information, documents, structured data, unstructured free text, and a variety of other things to interpret what it is and make predictions in different contexts. And here's a great example of this. This is an application, one of many that we've built using Watson technologies related to determining whether meals and entertainment deduction applies to expenses that employees submit.

(19:11):

And in this use case, you can imagine, most companies receive millions and millions of receipts every year that are submitted for reimbursement. Companies have to look through those and say, “okay, does a deduction apply? And if so, at what level on that?” And the challenge of this is not only one of time and effort, but it's a thought, and consistency in approach. And so, what's involved here? Well, a human has to look at a receipt copy, they have to look at the little description that you typed in saying what the expense is. You likely select a category of what the expense was and all of those things have to be read through and interpreted to say “is it a business meal? Is it a group meal? Is it some sort of entertainment?” Or is it just something that likely doesn't qualify for a deduction? And then what do they have to do?

(20:02):

They have to go out and look to the law. To the codes, the regs, the administrative interpretations, cases if applicable. And even contrary authority to make a prediction. Does my classification qualify for a deduction? And so what you see here, that little green, reverse “C” there is saying that the program at a 71% confidence level is predicting that 100% percent deduction applies to this transaction. And so it's not only reduces time, but it captures intelligence and extends our ability to do things that sometimes we can't do because we don't have the capacity and it allows us to utilize data in real time to better align with the business.

(20:48):

And so you may be thinking, “okay, I like these technologies, they're interesting, but hey, I'm a busy person. My head's down. I’m a tax professional, I'm dealing with tax reform. I've got all sorts of things I need to do. I don't have time for this.” Or you may say, “hey, I just don't see the need.” But I'd ask you to think back to the early 1990s when that computer was introduced and that tax professional with their 10-key in their library and their spreadsheets and their telephone and typewriter thought that they had a choice. They thought that they had a choice but to adopt this. And that's the question that I'd like to leave you with. Is do we have a choice but to begin to adopt these technologies, to redefine and respond to these growing demands from our organizations? And these sort of distant market forces that are transforming the tax industry? So thank you for listening to my presentation.