Skip to main content

A Message That Was Ahead of Its Time

As AI-focused technology makes the headlines, we take a step back in time to replay a 2018 BankOnPurpose presentation by Carl Ryden, Q2 executive fellow and former CEO of PrecisionLender, who shares where he believes artificial intelligence, financial technology, and the banking industry are headed based on what he was seeing and hearing in the industry across the world.

Listen

 

Watch

 

Subscribe

  

 

Related Links

[Blog] FNBO Drives Deeper, More Meaningful Customer Conversations

[Blog] Start ’23 With No-Code, Custom Andi Skills

BankOnPurpose

[Article] How ChatGPT Managed to Grow Faster Than TikTok or Instagram 

 

Transcript

Alex Habet

Hi, and welcome to The Purposeful Banker, the leading commercial banking podcast brought to you by Q2 PrecisionLender, where we discuss the big topics on the minds of today's best bankers. I'm your host, Alex Habet. Today, we're going to both simultaneously be reacting to some tech headlines that you're all probably seeing out there, but we're also going to contextualize it with a framework that was actually shared a few years ago.

We can all probably relate or remember a certain presentation or a talk or a lecture that inspired us to think about a problem a little differently, a little more creatively, and finding the associated solution. In seeing the headlines and some of the breakthroughs today, one particular talk from the past stood out in our minds. We're going to rewind to that moment in this episode for the important contextual element for why it matters. It goes without saying that this rewind was remarkably ahead of its time.

But before we get to that, what are the headlines I'm talking about? I know I'm not the only one who's regularly bummed out by what's in the news. But recently, there have been some stories, decidedly exciting, on the science and technology front. Of course, I'm talking about the news around ChatGPT built by the company OpenAI on top of their GPT-3.5 model, getting a little technical there. But this one is all about what you experience when you actually try it out.

It truly shows you how that underlying model in the technology is uniting a vast amount of knowledge and facts, but more importantly, language to deliver that mind-blowing access to the valuable insight across pretty much any topic. You try and you're like, "Whoa, where did this come from?" Then you look around and you suddenly realize that all of the mainstream is all of a sudden aware of other incredible AI tools, like the ones that help you write better or those image generators. In fact, I actually recently tried out DALL E 2, which is also by OpenAI, and I was blown away by that as well. It's also pretty hilarious if you try some funny things you ask it to generate for you.

But these experiences are waking up a lot of people in a way that really feels generational. There's something next level about it. I've seen this moment quoted several times as an iPhone moment, at least for the field of advanced computational science or predictive models and that sort of thing. We're just scratching the surface on what this can be and how it can be applied with today's problems.

I know this is also a banking podcast, but it's impossible to ignore the other big consequential breakthrough. Felt like it dropped the same weekend as ChatGPT. Of course, I'm talking about the news around the breakthrough for the quest for commercial fusion reactors, which that concept always felt like science fiction to me, like that possibly is something that I'm not going to see in my lifetime, let's say.

But now, the story's changed. I watched all three hours of Lex Fridman's interview with Dennis Whyte, who's one of the lead scientists on the project, specifically around the designs and magnets that work to eventually heat up the plasma to cause the reaction. But forget about the technical marvels here, just the applied uses with this kind of technology once it's achieved and economically viable to do so is astounding. You would hear this conversation, and everything just feels logarithmic and potential. Say what you want about 2023. These are definitely exciting times for the science and technology front.

Let's bring it back to the main event for this episode, and specifically we'll bring it back to the realm of computer science in this case. Look, all of a sudden, AI is roaring back to life, at least the interest. It's always been around, it's always been worked on then. The attention it's getting, all of a sudden, is extraordinary. I think a big reason for that is a lot of companies out there had trouble harnessing the power of what machine learning and AI can bring.

With that in mind, we went back and listened to a talk that Carl Ryden gave at BankOnPurpose just a couple years ago, was actually right before the pandemic. What really struck me as really interesting was how approachable even the more advanced stuff he was talking about in there now suddenly felt right. Don't get me wrong, a lot of the talk, Carl emphasizes the practical uses for intelligence and even starting with the basic heuristics and then graduating to the next-level stuff. But it's all about building a framework around coaching, and that's where it could be really advantageous, specifically when it comes to banking.

But the difference watching this talk two years later is many of the high barriers of entry for the really advanced stuff now suddenly seem completely possible, thanks to what the inspiration behind things like ChatGPT are bringing in and, of course, many others that are underway. The underlying technology that is powering experiences like ChatGPT is continuing to improve. It's evolving, it's being put to uses beyond just chat, it's showing people practical ways and tools to get to that next level across a lot of different topics. This is going to naturally mean things are going to accelerate.

All I'll say is our Slack channels here internally, specifically with our Product team, they're going off in ways I have never seen before. It's really exciting, but stay tuned on that. For now, I'd like to leave it to Carl to share his incredible perspective on how the power of AI can shift the future of banking, or any industry for that matter. Sit back and enjoy something that's ahead of its time.

[replay of Artificial Intelligence: The Power of AI and the Future of Banking]

Carl Ryden

The title of this talk, I actually didn't come up with it. I'm not sure who did. I found out the title about maybe a week ago when I paid attention. Brandi sent me a slide deck with that title slide and another slide with one bullet on it, and that bullet was, "Put magic here." My choice was I can go deep on this rather ambitious topic here, or I can just tell you some really good stories and distract you from this ambitious topic, but you'll still walk away happy. I'm going to take the second route. I think we actually might come back around to some of the first, because I think some of the stories are a good way of illuminating where I see the world going.

I bet absolutely no one in the audience knows who this is. This is a gentleman named Doug Engelbart. In 1968, December 8, 1968, he walked into a room of 1,000 engineers in an auditorium in San Francisco. He's no Steve Jobs. He's a humble engineer. He was getting ready to do a presentation to this group to show them how computers could be far more than just calculators, how they could be ways of expanding the collective intelligence of all human beings. How did he come to this point? He was born in the Depression era on a farm just outside of Portland. He then went to Oregon State University, where he began studying electrical engineering.

He did this because he knew the war was pending, and if he studied electrical engineering, he could be an electronics technician or radar technician, and they don't get shot at as much if he gets drafted. He did get drafted. He was on a boat heading to the war in Japan, thinking he was going to have to invade Japan. As the boat sails out under the Bay Bridge, it's heading for the Golden Gate Bridge, he hears San Francisco erupt in cheers. He's wondering what happened. The captain of the boat comes out and says, "The war's over. Japan has surrendered." He says, all the guys on the boat said, "Turn the boat around." They didn't. They went for 38 days at 10 miles an hour and got to the Philippines.

He served his time in the Navy as a radar technician, as an electronics technician, came back, finished his degree at Oregon State, went to work for the predecessor of NASA. It was called NACA at the Ames Research Center in Stanford. Got engaged to his wife. He's driving to work one day, he says, "I've now achieved what I wanted to achieve. My goals were to have a steady job, a stable family, live happily ever after." But he had read a book in the library while he was in the Philippines, in the Red Cross library about setting goals for your career, setting goals for your life.

He decided that what he wanted to do was he would focus his career on making the world a better place. That's what he wanted to achieve. He reasoned the way he would go about it was that any serious effort would require some kind of organized effort that harnessed the collective intellect of all people to contribute effective solutions to the world. He thought if he could dramatically improve that, if he could boost every person on Earth, boost their intelligence to solve problems, the sooner, the better, and he thought computers were the best vehicle to dramatically improving this.

In that presentation, in that demo to 1,000 engineers in San Francisco, he opened it with this statement. "If in your office, you, as an intellectual worker, were supplied with a computer display backed by a computer that was alive for you all day, instantly responds to your every action, how much value could you derive?" He went on to do what's widely known now around the world as the mother of all demos. In it, Engelbart sat down at a computer keyboard. Now, what you need to know is this is 1968. Computers were only owned by governments and certain universities and large corporations, and they were truly seen as large calculators. You'd take punch cards. You'd feed them in. They'd do a bunch of calculations. They'd spit out these bifold reports that people would roll around, drop off at folks' desks, then come back at the end of the day, pick them up, having not been read.

He thought there was so much more that they could do, and he saw them as a tool to augment human capability, to expand what folks could do, and actually transform what people are. He displayed the keyboard, the mouse, the graphical user interface, the display. He walked on stage with a headset, video conference to his lab back at Stanford, hypertext, word processing, the first file system. This was 1968. This was 16 years before the Apple Macintosh. This was 35 years before Skype and 40 years before Google Docs. He thought, "If I show this to these 1,000 engineers in the audience," that he would immediately have hundreds of them line up after the presentation that want to help him change the world. At the end, they were all awestruck. They stood up, gave him a standing ovation, and then filed out, because their world was punch cards and calculators and reports. It was too big of a leap for them.

One guy in the audience was a guy named Alan Kay. Alan Kay went on to run Xerox PARC, which took a lot of the engineers from Engelbart's group and further developed them. That's where Steve Jobs later toured Xerox PARC, and saw the mouse and saw the keyboard, saw the graphical user interface, and that transformed into the Apple computer. Doug Engelbart, almost no one's heard of him. Alan Kay said that presentation was life-changing to him. Alan Kay is largely known as one of the fathers of Silicon Valley. He was the chief scientist at Atari, at Apple, and HP, and all these places. He actually invented object-oriented programming. Alan Kay makes the joke, "I wonder what Silicon Valley will do when they finally run out of Doug's ideas."

It is amazing, as I was researching for this talk, I found my way to Doug and what he was trying to do in terms of intelligence augmentation, and it really captivated me. He once actually taped a brick to a pencil to prove a point. In 1962, he was investigating how tools shape who we are and how they augment our human abilities. He actually taped a brick to a pencil. Now, augmenting our capabilities is nothing new to humans. We've done it forever. We don't have fangs, we don't have claws, so we invented spears and arrows. Once we invent spears and arrows, now we can actually accomplish more. It not only shapes what we can do, but it changes who we are and it actually changes the way we develop as humans.

One of the early technologies, humans, like they don't have fangs and claws, they also don't have long-term memories, so we created language. They can't transmit language through generations and across time and space, so we created written language. Once you create those tools, those tools allow for bigger tools. Once you have written language, now you can have literature, mathematics, poetry, everything that we build upon. That's why he tied a brick to a pencil, to prove a point.

Of all the tools that we created to augment human intelligence, perhaps writing is the most important. Actually, one of the tools that you guys deal with every day, one of the technologies that you deal with, is we created money. Money is actually a technology. Particularly, it was coinage and then paper money, flat money. That was a technology that allowed us to transfer value across time and space, just like we transfer knowledge across time and space with written language.

But what he found with the brick is when you de-augment the pencil, when you make it hard to do the lower-level tool, written language, that it actually destroys all the higher-level things. When you tie a brick to the pencil and make it hard for folks to just write letters, they can't form more coherent thoughts. They can't form higher-level thinking. It breaks everything down. Now, you look at this and you might laugh. You say, "A brick tied to the pencil, that's silly."

I've talked to many banks around the world, and I've had several of them tell me that for an RM to actually get a deal done, they have to touch 100 different systems. Of those systems, none of them are actually valuable to them to making progress in their life. They're entering data into a system to run through a calculator, generate a report, that maybe not someone reads. They're bricks, not pencils. It's a tax, not a tool. The same fight that Doug Engelbart was fighting back then about it's not just a calculator, this is a way of expanding what we can do and what we can accomplish, and how we interact with each other and the world around us, I think is just as true today as it was then.

Digital transformation. As I dug into this, how does a guy like Doug Engelbart invent the future in 1968, really see such a clear vision of this? How does he understand truly what innovation is? Actually, if you google Doug Engelbart, some of you may have now, there's actually a rule, a Wiki page, that describes Engelbart's Law. Engelbart's Law predated Gordon Moore—Moore's Law. Moore's Law, some of you may have heard of that, it's the capability of microprocessors double about every 10 years and remain true forever. Engelbart's Law was the rate of human improvement is intrinsically exponential.

If you go back and look and plot GDP over time, over any time period, worldwide GDP, you will see there's blips in there. What you'll see is exactly what Engelbart said. Why he believed this to be true is he broke the world into three processes. He's an engineer, so he wasn't really good at naming things. One of the great ironies there, he called the system he showed folks the online logical system or something like that. You would think that'd be OLS, but he called it NLS, because he used the N in online, so it was called the NLS. He had a bad name that he transformed into an even worse acronym, and that proves he's an awesome engineer.

He developed a process for improvement he called the ABC process. He talked about process A systems. Process A are the human and tool systems that we use to do our everyday work. It's how we do what we do. Human systems, culture, language, society, procedures, policies, those things are how we do what we do. Process B systems, the human and tool systems. Process B systems are systems that help us get better at A. It's business process reengineering, Lean Six Sigma, knowledge management, all these things that help us get better at A.

But he says, "Really what makes humans unique and allows for this intrinsic exponential growth of capability is Process C." Process C are the systems, the human systems and the tool systems, that allow us to get better at B. It's getting better at getting better. If you think about a race car, its position, velocity, acceleration, when you get to the third level and you can out-accelerate everybody, you will catch them, you will pass them, and you will win. That's what drives the curve this way.

Now, if you think back to digital transformation, I'm going to talk about maps. I could take this all the way back to the Vikings. The Vikings, when they used to navigate, they would say it was all language and storytelling. "You go here and there's a dragon, then you turn left, and maybe there's a dragon, maybe there's not." But that's how they would figure out how to get around. Then they developed written language. Written language, it lent itself to make maps. You start drawing pictures, and you put language on the map.

I'll take it to our time. Remember when you were a kid? Who knows what the slot on the back of the seat behind the passenger seat's for? It was for maps. That was the map slot, when you'd put the big Rand McNally thing in there and you'd stick in there. It was so you could reach back and pull it up and go, "Where the hell are we?" While you're refusing, the wife's saying, "Pull over, ask for directions." "I'm not going to do it." That was probably the predecessor of "Texting and driving is dangerous." Wow, Rand McNally map.

You started with maps, but then we had the technology of process B innovation, which is the GPS. The GPS allows you to see where you are on the map, and you see you can correlate it. We'll say, "Well, shucks, that's no good. Let's make it better. Let's put the map inside the GPS, and we digitize the map. Well, shucks, once we know that, let's start doing turn-by-turn navigation, then you get a Garmin thing that tells you to turn here and turn there and move there."

All those were Process B innovations. They made it better. But then you go to Waze. Now, Waze doesn't just give you information. Waze understands your goals. "I want to get from here to there." Waze has this wonderful Process C network effect that every user of Waze makes it better for every other user of Waze. It becomes a cycle, a flywheel that feeds on itself. You can see how this has come true in quite a lot of areas.

One of the things we talk about with digital transformation ... I was at an event in Ireland with Fortune 100 companies talking about digital transformation. I was mostly listening. One of them, what they talked about was transforming from a project focus to a product focus. It was a big part of that. I don't think many people know what that actually means, but I'll give you a hint. If you look at Amazon or any of these other companies, and you see how they think about their business, it's a flywheel. Amazon is low prices, wide selection. That allows us to serve more customers that allow us to get lower prices and wider selection. That allows us to serve more customers, and you work around that flywheel. Waze, we help customers navigate and make it useful. More customers come on. That makes it more valuable to the next customer, and it's a flywheel.

Product companies typically have a circular path to how they want to deliver value. Project companies have a Gantt chart. It's linear. We're going to start here, we're going to do all this stuff, and then we plant a flag, hang a "mission accomplished" banner, and say, "We did it," and it's over. We measure the ROI on that project. We don't think about how we build the machine that's going to accelerate value creation that's going to get us to Process C. I would say the difference between project focus and product focus is if we map out your value, if it's linear, you're doing a project. If you map it out and it's circular, you're probably doing a product. That's a simple heuristic, I think.

I want to talk about our journey to Andi. Where we started, when we started PrecisionLender ... Actually, a long time ago, I started working with Mitchell Epstein in 2004 helping banks with pricing. Mitchell had a set of Excel sheets, and he wanted me to help him build it into products. I started working on this, and first, I was like, "This can't be a thing. Surely, folks know how to do this." Then, "Well, it can only be small bank thing." Then, "Well, maybe it's only a medium bank thing." Then, "Oh, crap, it's a big bank thing," and "Oh, wow, it's the same thing on the other side of the Earth."

But what happened is I started thinking this, "What is it we're trying to do here?" It was really similar to that Doug Engelbart's insight. It's not a calculator. This is a tool that expands what the RM can do. This is a tool that helps them have better conversations with their customers. This is a language for the back of the bank to communicate with the front of the bank, so we can all stop this shooting war where the customers get caught in the crossfire. "Here's what's important to us. Here's what's important to you. Let's put that together and have a better conversation with this customer." That was where we led.

One of my guiding principles when I said, "What am I doing here?" wasn't my Doug Engelbart moment, but we figured out that what I was trying to do is I wanted to build a system. Oops, I'll go back before I show Andi. Sorry, I previewed that. The way I say it is we want to allow the machine to do everything that a machine can do well, so that, and the so that's the most important part, the humans can do what humans can uniquely do well even better. We're not trying to replace humans. We're trying to augment humans. We're trying to make them better. We're trying to expand what we're building, tools. We're building spears and arrows. This is as old as time. This fixation on calculators, feed a punch card in, do calculations, get a report out, is something that I think has led us astray.

The story of Andy and the magic dots. This is Andy Max. He's in the room. When we first launched PrecisionLender, we had these little dots on the screen that would help guide a user. Where that came from is I hated the systems that I saw that folks would build, where it was typically the IT folks would take over from the business folks who had built an Excel sheet, a citizen developer who has built an Excel sheet of some sort. They would take over this. Then what you really need is a forms-based application, because that's what we build, because that's the modern-day punch card. You fill out a form. It goes into a database.

You need a forms-based application where they fill all this stuff out, hit Next, fill out a screen, hit Next, fill out another screen, hit Next. You get to the last screen, it says, "Sorry, that didn't work." They go, "Well, this isn't helping me. This is a brick. It's not a pencil." We said, "Well, we have a computer here. We'll just tell them the answers. This is not really complicated. We'll just give them the answers in the back of the book, and here's how you could make the deal work." We put these little dots on the screen.

We were lucky enough to partner with First National Bank of Omaha and Andy Max. Andy Max did have a sweet model. He talked about this sweet model the other day. He understood that problem. He understood how we were trying to help the RMs be better. We're giving them a tool, not a system. Nobody actually wants a system, they want a tool. Think about if you're digging a ditch, and you got a broken shovel and some guy shows up, "Here's what you need. You need a system that every 10 feet, I want you to fill in this form and hit Enter, and then I'll generate a report, and I'll come back and tell you why you're not meeting the benchmark of ditch digging. I'll give you a better shovel. I'll give you a Ditch Witch. We can dig the ditch a lot better, and we'll focus on that."

Andy was really good at understanding that, and he would actually coach the RMs. He would sit with Joe and work this deal. He'd help him find a way to make it work. Then, later he was talking with Jim. He said, "You know, Joe was just doing this. Maybe you ought to try this." He was just cross-pollinating, taking what's best and spreading it to the rest. He would call us up and he would say, "Hey, Carl. Could you put a magic dot on the screen that when you see him doing this, why don't you tell him to do this?" Man, that's really smart, so we did that. We were doing a lot of those.

Then one day, I was sitting with a developer, this was about 2 1/2 years ago, sitting with a developer. We were going through the list of the magic dot requests from Andy, and we were talking about how Andy made our product probably 100 times more valuable to his bank because he was there. She said, "Wouldn't it be wonderful if every bank had an Andy?" I said, "Well, let's see if we can do that. Let's see if we can build a way of finding what's best, spreading to the rest, see if we can find a way of scaling that out, and augmenting every lender's intelligence, improving the collective IQ, helping them make progress in their lives."

Two years ago today, I came to our CAB (Client Advisory Board) and I said, "I'm going to try to do this. What do you think, guys?" I had a slide that says, "You might think I'm crazy, but I think this will work." Then they said, "Well, you're not crazy, and we think it'll work, too." I said, "Well, OK, we're going to do it." Then January 2017, we launched Andi, so six months later, seven months later. We said, "Every PrecisionLender user now has their own pricing analyst. Her name's Andi. She's new on the job, but she learns really quickly. She works 24/7. She never sleeps. She sees every deal. She learns what's working, what's not. She helps you price each new opportunity, she monitors, and she has a framework where you can build your own skills, where you can put your own magic dots to make it better."

Now, what have we learned by doing this? By the way, when we did this ... If I back up, I'll give you the general theme. I thought this was going to be cool, a lot of machine learning stuff, to see all this stuff, and it does provide a capability for that. But the first, most valuable set of things that folks asked us to do were exactly what Andy had been asking us to do, simple heuristics. If you see him doing this, tap him on the shoulder and say, "Consider this." There's no machine learning in that. There's no AI. It's just I. It's just intelligence. At the end of the day, the other thing is the RM who's getting that suggestion doesn't care. They don't care if you cracked open a fortune cookie and typed in the results. If it helped them make progress in their life, it's magical, and they like it and they appreciate it.

What have we learned? One is context is king. Narrowing the context makes everything easier and allows you to, instead of having 1,000-page user manual that you give to every single user, you can actually target it to specific behaviors at a specific time, and it becomes on-demand, kind of just training and coaching. Even how to use the application, I don't have to do the shotgun approach where I get everybody in one room. I don't do the shotgun approach of giving everyone a manual and read the whole thing, that they don't. When they don't and I find a problem, I got to add one more page to the manual, which makes that problem only worse. It's a negative flywheel, by the way. It's going the other way. It continues to get worse.

User experience matters immensely, and it's not merely from an ease of use, it's not merely a saving mouse clicks. There is a lot of that. It was interesting to hear Kristen talk yesterday. In PrecisionLender, there are no empty boxes, because when you have empty boxes, what do people do? They give up. We don't want them to give up. We want to lead them to a place that helps them be better. Allowing Andi to see what's going on in the application and allowing to see the moves the RM makes and make it easy for them to make the move so you can see it, is really important.

Not all data is equal. Behavioral data is the best. Most of the data within the bank is not this, and no amount of big data projects is going to change that. If you focus on a narrow context and look at behavioral data, particularly behavioral data grounded to outcomes, when you do this, it results in this and that's good. We can label it as being good. That's what labeled outcome data means.

Consider the Waze example. Every interaction with the system, every time somebody reports, "Hey, there's a police officer here," and somebody else says, "Yup, I saw it, too," every interaction makes it better for every other user. Every deal you price ought to make it better for the next deal you price. Every relationship you bring home ought to make you better at serving the next relationship. Every interaction you have with your customers ought to make you better for that next interaction. How do you get to that Process C flywheel going? I think a lot of what we talked about yesterday was exactly that.

Other thing I said already, no one actually wants artificial intelligence. They just want intelligence, the artificial kind, the unnatural kind. Artificial means unnatural. Natural means natural. You have a lot of natural intelligence that lives within your bank. Who are the best RMs? What do they do? What are the things that they do where there's a lot of friction? Figure out who's best, share it with the rest. Use the machines to amplify the humans. I say build Iron Man suits, not Terminators. Amplify the human-in-the-loop. Arbitrage what's best, spread it to the rest.

One of the things that surprised me is the personification of Andi. I'm an engineer. I look at the NLS system that was this magical thing that he called the NLS. The magic dots, we called it the profitability wizard. We weren't even bold enough to put a TM by it if somebody would steal that stupid name. Nobody actually called it that. They used the heck out of it, but they called it the magic dots because that's how they experienced it. The personification of Andi has been amazing, because the responses are, "I so need it." I could go walk into a room of RMs and talk to them about machine learning and data and label, but when I show them what Andi does for them, they go, "I so need an Andi." It gives them a nice handle to understand it. We talk about this as humanizing the machine. Don't mechanize the human.

It guides the development. We focus on jobs to be done. Jobs to be done is featured in Clay Christensen's recent book, Competing Against Luck. It's how you understand the job folks are fundamentally trying to do. The quintessential example of that is no one actually wants a quarter-inch drill, they want a quarter-inch hole. They hire the drill to give them a better hole, so the drills compete on, "Ours is more voltage and more power, and red and blue and green." They're competing on things that no one cares about. "I don't want a drill. I just want a hole, and so I hire the drill to give me a hole." 

But when you actually try to uncover the job to be done, you say, "Well, why do you want a hole? Why do you want a drill?" "I want a hole." "Why do you want a hole?" "I want to put a fixture there, a light fixture." "Why do you want a light fixture?" "I want to be able to read in bed." "How about I give you a Kindle with a backlight?" "Oh, OK." Now all of a sudden, a Kindle with a backlight is competing with a drill. You get fixed to the thing that you have, which is making better drills, but not fixed on the job that they're hiring the drill to do. 

Last year, we had Bob Moesta speak, who's a pioneer in this. The title of his talk, you ought to go back and watch it, was No One Actually Wants a Loan. They want what the loan gives them. They'll tell you they want it cheaper and they want it faster. That's evidence that they just don't want it. They want what it gives them, what it allows them to do. You have to be able to connect to that job that's being done.

When we rolled out Andi, we used to get a lot of requests from folks. "Can you put a button here? Can you give me a report that does this?" I'd go, "Well, OK, why do you want the button?" You go through the whys, and you really don't need a button. What you need is this, this little nudge here, this little nudge there. You really don't need the report. My favorite question to ask in reports is, "If I gave you that report perfectly, what would you do differently Monday morning once you have it?" "OK, I'd tell Joe that he needs to do this differently. I'd tell Joe he should have done this differently," which is the worst. Monday morning quarterbacking. "Why don't we detect what caused that and coach Joe to do better, and then you don't ever need the report? Because now the report comes in your inbox, and I'm depending on you to go beat people over the head and tell them to make it better." That never works. 

Then the extensibility through the skills has opened up the floodgates of, "Could Andi do X for me? Could Andi do this for me?" One of the ways we talk to folks is don't write a spec. Don't write a spec of put a button here, whatever. Write a memo to Andi. Andi is on the shoulder of every RM. They're with them helping them. You say, "Hey, Andi. If you see they're pricing a commercial real estate deal, I'd like you to go call the API, the S&P. Pull down all the lean data that also for that borrower, and what other properties do they have with which other banks at which vintage," and sometimes it even shows the rate, "and deliver that to the RM right there."  

Now, what we've done there is we've transformed a chore to be assigned into a gift to be received, which is really enormously powerful in terms of the human experience it creates. The last thing I want to talk about is one of the things we learned is there is an intelligence value chain that I'll talk about, and when you have to implement the entire chain. In most solutions, most AI solutions that are being deployed in banks or anywhere, they die at the last mile. Let's talk about what that is. 

Ajay Agrawal is a professor at the University of Toronto. He's an economist. He studies AI and the economic impacts of AI and how it matters. He has this thing he calls the anatomy of a task. You start with data. You use the data to make a prediction. That prediction is presented to a human, human-in-the-loop type stuff. The human says, "Yes, I'm going to do it," to take this action. That action leads to an outcome. You then have the machine that looks at the actions taken and the outcome generated, and uses that to feed back to make better predictions. A couple things you'll notice. There's a circular path here. There's a feedback loop. There's a Process C taking place here, that we can get better at getting better all the time, which leads to exponential improvement, which gets the acceleration going in the car.

Gordon Ritter. Gordon Ritter's a guy at Emergence Capital Partners out on the West Coast. Gordon, 20 years ago, he's a thesis-driven investor, a really sharp guy. Twenty years ago, his thesis was enterprise SaaS software was going to win. He invested a million dollars, the first million dollars, into a little company called Salesforce.com, and that proved to be true. He later invested in Box and Workday and all these other things along that thesis. His second thesis 10 years ago was that the first wave of SaaS software was horizontal, and the next wave will be vertical, and it'll be bigger than the last.

The first wave of SaaS software was horizontal stuff, sales force automation, Salesforce.com. You had all the marketing tools that were horizontal. You had NetSuite, which was kind of ERP for businesses that was horizontal. The next wave was going to be vertical. In that thesis, there was a company called Veeva, Veeva Systems. You may have never heard of it. They're a public company now. He put $6 million on a $20 million pre-money valuation, so he owned about a fourth of it. They never touched the $6 million. Veeva went public eight years later with a $6 billion valuation. They're approaching half a billion dollars of recurring revenue now. Veeva builds systems for the life sciences industry, and how they run their business. It was targeted at a vertical. There's many more examples of verticals.

His current thesis is called coaching networks. His thesis here is that forms-based applications are going to go away, the punch card, calculator, report. He's an ultra-marathon runner. He says, "When I run a marathon," he says, "I don't stop every mile and fill out a form, check the report, then run another mile, fill out a form." He said, "I have a Fitbit. It sees what I do. It knows my goals. It's coaching me. Here's a point where you can pick up some pace. Here's a point where you might want to slow down. It's coaching me to get there the whole time."

He says, "Where the world's going to go to is you do what you do as a human. The machine sees what you do, and then coaches you how to do it better." His loop is gather, compare, and coach. You see what the humans do, you gather that. You compare it. What do the best humans do who have the same goals as you? Then coach. Coach those other folks to do it better. This should also feel familiar to Waze, this circle.

One of his companies is a really interesting one, I'll just mention it later, but I'll tell you now, is Chorus.ai. Has anybody ever heard of this company? You probably couldn't use it in a bank just yet, but what Chorus.ai does is it listens in on a sales conversation. It records. The machine listens. Your salespeople have their conversation. Chorus.ai listens, understands the natural language, extracts out the to-dos and the follow-ups and the next actions, fills out Salesforce for them, creating the follow-up tasks, then compares their conversation to the conversation other salespeople have had, and coaches them on what they could have done differently. Think about that as the experience benchmark of where the world's heading. You do what you do. The machine sees what you do and then coaches you to do it better.

His other sub-thesis of this is that, in fact, enterprise organizations like you guys are actually in a really good position to do this. His sub-thesis, if you think about like Facebook and Googles of the world, they have to harvest your data to sell to advertisers to deliver their... Their economic model is based upon it. One of his sub-theses is help, not harvest. You guys aren't in the harvest business. You're not gathering information on your RMs to target them with ads. You're gathering it to help them be better. Help, not harvest is a big theme I think you're going to see in enterprise software of we're not harvesting. The harvest stuff, this is why you probably don't want to hire a bunch of guys from Facebook with hoodies to build your stuff. They're harvesters. You want the folks who know how to build that feedback loop and generate help.

Systems of record, systems of intelligence, SoR, SoI. Oh, my. The three types of systems, this I got from Jerry Chen at Greylock. Systems of record are the typical things, the three-letter guys, CRM, ERP, SCP, all those things. You guys probably have a ton of these. They capture data around what happened. They deal in facts. This is binary. Either this happened or it didn't. Systems of intelligence, DataRobot's one of those, TensorFlow, all these machine learning tools, is we take all of those set of facts, and we translate them into beliefs. Belief's nonbinary. I'm 80% certain this customer's going to churn. I'm 90% certain that we could close this deal. I'm 75% certain that this is a really good prospect. Predictions are just beliefs about the future. They have some confidence in the outcome.

Systems of engagement are the ones that ride shotgun with the user. They see what the user does. They translate the facts and beliefs into discrete measurable actions. They deal in actions. PrecisionLender is, I think, one of those systems of engagement. That's what we try to build it to be. Hemingway App is another one where you can actually ... If you go to look up Hemingway App, you can actually write a paragraph, and it'll coach you to write more like Hemingway, which is fun. Chorus.ai is another one.

We're in the early innings of this, make no mistakes, and we're still working on this. When you're in the early innings of something, everybody seems to see the same thing, but call it different things, because we haven't yet settled on the language. Greylock's Chen, systems of record, intelligence, engagement. Microsoft's Satya Nadella, who has basically retooled the entire company around AI, machine learning, and in fact, augmentation of human intelligence. If you read his book and hear what he says, he calls it perception, cognition, and action. Gordon Ritter called it gathering, comparing, coaching. Agrawal called it data, prediction, and action. These are all the same things. We just haven't settled on the language of what to call them yet, but that is truly that AI value chain.

I want to talk about centaur chess. This is Garry Kasparov. In 1997, Deep Blue became the first computer to beat a human at chess. In 20 moves, Garry Kasparov resigned, and this is him walking away from ESPN. It was really very funny. Right now, you can actually download a chess-playing AI for your laptop that's better than Deep Blue, that will beat Deep Blue, just to give you an idea from 10 years later. He demanded a rematch. IBM said no, because the marketing benefit of this was now done and they had nothing to gain. They actually shut down Deep Blue.

Garry Kasparov couldn't help but imagine, what would it be like if computers actually worked with humans? He created this game he called centaur chess, some people call it freestyle chess, where instead of a computer playing a human, you had three types. You had the best computer in the world playing chess, you had the best human grandmasters in the world, and then you had these folks in the middle, centaurs, the mythical beast, half-horse, half-human, where it's a human with a computer. There's no surprise that the human with a computer beats the human.

What the big surprise was is that the human with the computer beat the best computer every time. What's even more surprising is it wasn't the best human with the best computer. It was a reasonably good amateur chess player who had three computers, reasonably good computers, PCs, laptops, and he had an awesome process for deciding when to use which. He would actually have each of them suggest their moves. If they all agreed, that's what he would do. If they didn't agree, then he would actually have them explore different options using his judgment of which things might matter, and then come back. He had a really clear process of how those worked together.

The human plus the machine beat both the human and the machine. This is where I say, Iron Man suits beat Terminators every time. I can't prove that, but I've watched enough Marvel movies with my 11-year-old son that ... He actually wanted me to show the last scene from Iron Man 2 where all the autonomous robots, that the Iron Man suit killed them. I said, "I think I got copyright issues with that," but use your imagination.

The key part about this is when you create a human plus AI system, the hard part isn't the AI. There's plenty of tools out there. DataRobot's fantastic at helping you with the AI and the machine learning. The hard part isn't even the human. The really hard part is the plus. How do you put them together? How do you decide when to use which? What you find is computers are ... Pablo Picasso said, "Computers are useless. They only give you answers." But they're really good at giving you answers. Computers are great at answers. Humans are great at asking the right questions. When you put those two together is how you actually turbocharge that improvement and actually build that Process C innovation.

Again, there's many ways of saying the same thing. If you look at that entire value chain of AI, Gartner calls this Cognitive Expert Advisors, which is a very Gartner-type way of saying it. It sounds like this robo-consultant, which that doesn't seem very good. Accenture calls it Citizen AI, which is how do you actually put the humans together with the machines and augment them? I like Gordon Ritter's coaching networks. Deloitte calls it cognitive collaboration. But all of them have reached the same conclusion that it's really the combination. It's the centaur chess play. It's the freestyle chess play. How do we combine machines with humans?

I just like coaching because, going back to what I saw Andy Max doing and what we built into our Andi, what really tends to drive value is coaching, and coaching to be effective. This is also why people have talked about my hatred of reports. A report, you put data in the system, you get out a report, you come back way after the fact, and you say, "Here's what you did. It's not up to the benchmark. Do better." That's not coaching. That's kind of yelling at the kids from the sideline, "Score more points." You're not telling them what to do.

I can do an analysis of your golf scores over the years and say, "You suck." What have I told you? "You're two deviations from the mean golfer." "Yes, I know, I stink." Then you give them some generic manual to read on how to be a good golfer. That doesn't accomplish ... If I go play with them, and I see them hit a shot, when they hit that shot, "Hey, look, on that one, your elbow was a little bit high. On this one, change your stance a little bit, change your elbow, and try it again." That's how they learn. That's how they get better. We want to do that. Coaching has to be contextual. It has to understand, have a deep knowledge of the context. I'm saying if you're going to coach people how to do better loans, you better understand how loans are profitable. You better understand how these conversations take place.

It has to be contemporaneous, in the moment. It can't be after the fact. It has to be constructive. You'll notice in PrecisionLender, when we pop up all the green dots of here's how to make the deal work, we don't pop them up and slam them in the face. You have to click, and Andi then gives you this ... You have to request it. I've actually had some bankers say, "Why do you do that? I want you to put it all caps, in their face, and put it red, and tell them, 'Here's the reasons you suck.'" It's like functionally, this functional spec is exactly the same. The behavior outcome is vastly different. This is why the user experience matters. It has to be constructive. Individualized.

If Phil Jackson's coaching the Chicago Bulls with Michael Jordan and Scottie Pippen, how he coaches those folks is if he's coaching the Pee Wee League team of nine-year-olds will be radically different. We have to individualize it to each RM to make sure it meets them where they are. It has to be actionable. It can't just be Monday morning quarterbacking. "Here's what I want you to do differently right now." Then it has to be attributable. When you actually build these skills in Andi or you coach, you need a feedback loop to say, "I told him to raise his elbow, and oh, man, his shot straightened out. That works for this person." You have to tie that feedback loop in.

That's a little bit about how we think about this. I want to finish up with the same quote I began with, which I think Doug Engelbart is unmistakably a genius. If in your office, as an intellectual worker, you were supplied with a computer display backed by a computer that was alive for you all day, instantly responding to every action, how much value would it derive? I would add what he saw, in terms of the way, the mechanism for making people better, augmenting human intelligence, is coaching, is how do you coach them in the moment to be better.