Drill to Detail - Drill to Detail Ep.30 "AI, ML & Oracle's Adaptive Intelligence Applications" with Special Guest Jack Berkowitz

Episode Date: June 13, 2017

Oracle's Jack Berkowitz joins Mark Rittman to talk about a new category of continuously adapting, self-learning applications being built-out by Oracle that use machine learning together with enterprise and third-party data to create a new generation of intelligent HR, CX, SCM and ERP SaaS apps.

So in this episode I'm joined by someone I've known for my work with Oracle for many years in the business analytics space, Jack Berkowitz. So Jack, welcome to the show and why don't you introduce yourself properly and tell us what you do at Oracle. Hey, great. Thanks Mark and it's great to be talking with you. My name's Jack Berkowitz. I'm the Vice President of Products and Data Science for Oracle's Adaptive Intelligence program. So we're all about bringing machine learning and really web-scale data into our applications to help people make decisions, to be able to give good recommendations, improve business outcomes, all those types of things.

Very interesting. So, Jack, you've been working with Oracle now for a while. I knew you from the business analytics and the BI app sort of days, but what was the route you had going into Oracle? Tell us a bit about how you kind of ended up there and kind of your past history, really. Yeah. So it's been an interesting journey, but I encourage everybody to go on a journey a bit.

So my career sort of broke into a couple of phases before I ended up at Oracle. The first phase was, for about 10 years, a series of consulting and sort of project-oriented capabilities around DARPA programs. And so I was involved in some of the very early semantic web projects, some of the early natural language understanding things, all of those types of things during the 90s. And during the 2000s, I worked in a series of startups in commercial software, but in the semantic web and in search. So I got to build one of the first scalable AL inference engines, productize that and put it into the market. I got to work on some really cool search problems. And in the course of doing that over those two periods, 20 years, got to know Oracle a little bit. And so in 2011,

I joined Oracle at first in the analytics team and now doing this. Okay. Okay. So I knew you best, I guess, from the Oracle days working on the BI applications, which was Oracle's packaged applications, packaged analytics to run on top of CRM and ERP and so on. Tell us a bit about that and the problem they were trying to solve and kind of what your role there with that was. Yeah, so the BI applications was really about trying to provide business content or business context in for analytics out of the box. So there are thousands of metrics, KPIs, that describe business.

And so it might be business about sales, or it might be business about HCM, or it might be business about procurement. And all these things are pre-calculated, pre-set, so that people can get up and running with analytics quickly. So a very successful standalone product in the data warehousing space and the classic analytics quickly. So very successful standalone product in sort of data warehousing space and the classic analytics space. But one of the cool things that Oracle did is they lifted these concepts,

these ideas, and brought them into all of the Oracle Cloud applications. So all the Oracle applications have all of these KPIs and metrics already built inside. And my role for a few years was to actually help that transition into the cloud and then work with customers to make sure that everything was working properly, that the metrics reflected their business,

and then even some of the more technical things like it performed adequately. And so that was my role in my team's role in that okay so and you talked there about moving into the cloud and i think that's a good kind of lead into kind of where you are now and obviously your background with with kind of semantic web and so on so you the apps moving moved into the cloud and your job now your role now is to try and introduce machine learning into that sort of into that space tell Tell us a bit about that and what you're trying to achieve there and the, and the kind of role of the team you're working with at the moment. Yeah.

So as we were looking at this move into the cloud of applications, it became pretty evident to us that, that even the entire world of analytics was going to change. Right. Probably the, the parallel I would have would be a pilot. In fact, my very first job out of grad school was building flight decks and designing flight decks for pilots. And if you think about it, pilots are very, very busy flying an airplane and get the right cues at the right time in order to help them fly their plane. They don't sit around particularly analyzing their speed

over time, but boy, if their airspeed drops, they get an alert immediately. And as we were starting to think about how people will use these applications and take advantage of these applications, that use case keeps coming to mind. We also start to see these things in consumer. So I use Waze every single day. Waze is this mapping capability that Google bought. I use it every single day to get through traffic in San Francisco to the office. And look at what Waze is doing. It actually knows a lot about me.

It knows a lot about what other people is doing. It's actually, it knows a lot about me. It knows a lot about what other people are doing. And it also knows a lot about sort of traffic conditions and things like that. And so it can offer me routing and it's not just the same route every day, but it actually adapts in real time to traffic conditions, to where I'm going and everything else. That notion, which is to provide that ease of use and that advice and taking advantage of data. And that was really the key, right? Waze uses, you know, tons and tons of data to make that happen. As our customers were coming to the cloud, we saw, hey, wait a second, they're going to have all their data. We can actually help them use their data more effectively, and we can provide these types of capabilities in terms of their business applications. And so this is a natural outgrowth

of people coming into the cloud, also a reason to come to the cloud, and it just made sense. So we kicked this off several years ago pretty quietly before we made announcements, but have been attacking this problem really, really diligently to try to provide, you know, these capabilities. So what you're saying there is you're helping, I suppose, the user of these applications focus on the tasks and solve the problems on the platform kind of better, really. So, you know, taking your analogy there of kind of ways, you're helping them kind of, I suppose, answer questions quicker in use of the kind of applications. So it's about focus and efficiency, really. Is that what you're saying? Yeah, it's really focus, efficiency, amplification, right?

So if I can take away, you know, sort of mundane tasks from them, right, I can let them get back to being creative and let them get back to being creative uh and let them solve problems uh creatively and that's really what we're we're trying to do with all of these embedded analytics with all of this machine learning is just amplify people's capabilities okay okay so with the so the products you're looking after the moment is the adaptive intelligent apps aren't they so what as a product line you know what kind of areas do they cover i mean i think it's cx and so i mean what kind of i suppose subject areas and verticals you're looking

at the moment yeah it's a great question so um we're actually looking across all of oracle's application footprint in the cloud and so there are things in cx and that's marketing sales and service and we made some really big announcements a few weeks ago at one of our conferences called Modern CX. But we also have solutions that we've announced around ERP, around discounting. We have things in supply chain that we're talking about, and also in human capital management, both recruiting and the balance of HCM. Probably the cool thing, and one of the things that we can do is, of course, because Oracle's not in any one of those, but across all of them, we can actually connect connected intelligence

across the different domains. So if you want to hit a sales goal, you have to have a certain number of salespeople. And how does that interplay in terms of the machine learning okay so so you've got analytics you've got machine learning in there I mean how do they I mean to define the two things there how do analytics and how to miss how does machine learning differ in these kind of platforms and what kind of different I suppose problems do they solve or assistance they give really or are they the same thing? Yeah, so, well, it's a continuum, right?

And, you know, you can get arguments all day long about where the edge is. The way we like to think about it is there's a difference between reporting and informing on a historical report and suggesting action or taking automatic action, like in a machine-to-machine sense. And so where we're applying the machine learning isn't just to be able to do projections to inform someone, but actually to either, you know, recommend actions immediately, or better yet, allow the person to say, once they get enough trust in the system, tell you what machine you take over and let the machine go ahead and do automated actions on their behalf. Okay, so in a way, this is a bit, this is a natural continuum from the work I think you were doing sort of years ago, where you're embedding analytics inside the

Fusion application. So back in the old days, we had things like, you know, in Oracle terms, Discoverer, we had kind of Oracle BI, that were kind of separate standalone reporting applications, then you were looking to embed these in the actual kind of like the application itself and the workflows. Is what you're saying now that you're taking that the next step forward and actually almost to the point in some cases of automating those decisions, but it's all about the efficiency and the focus of the person running the application. Is that kind of what you're getting at?

Yeah, that's exactly it. That's exactly it. And in fact, that was the trajectory that's exactly it and in fact that was the trajectory that the that we set out to do all along right it just takes stair steps to get there okay okay so thinking about that's an interesting sort of topic area and do you think the work you're doing does that kind of in a way negate putting say reporting inside applications and so on or is there still a need to have, you know, maybe, for example, graphs and charts in the Fusion apps? I mean, is this something that is a better version of it? Or is it still a continuum there for people?

Yeah, I don't think it negates it. I think it's a continuum, right? So, you know, at the end of the day, in fact, we've just finished our fiscal year, i was looking at sales numbers inside of our fusion application just like anybody else uh so there's there's a balance and and also you know those charts also um will come into play or those metrics will come into play maybe in a different sense now i've got a recommendation but i before i trust it i want to understand the underlying data as well and so by having both of those components together actually creates a really strong offering. Yes. So you're addressing, I suppose, the actionability of BI there as well. I mean, again, one of the, I suppose, critiques of kind of analytics in general is that often it's

very much, it's interesting and they're nice graphs and charts, but making them actionable is the challenge. Is that something you're trying to address as well. Yeah, yeah, definitely. We're trying to go after these challenges. I think the key to all of this, though, is trying to go after it in partnership with customers, right? So, and I think that's really the aspect that we can bring to bear at this moment because of the momentum that we have

about people picking up these applications in the cloud. Every time I turn around, I won't give an exact number because every time I turn around, there's another higher number of companies, but it's not even smaller companies anymore. Major companies moving their ERP systems, their HCM. We have customers with 150,000 or more employees in HCM. And so what the cool thing is, is being able to work with them very interactively because they are in the cloud and we're working with them directly there. You know, it used to be hard to do this, right? To try to do these types of projects 10 years ago or 15 years ago when I was working on them, you know, it was this massive data consolidation energy. And at the end of 12 months, then you hope to do some machine learning. Now, because the data is available,

and it's not just the company's first party data, but we can bring in outside data and augment their context, we can make progress for them very quickly. That's actually the most important part. So you mentioned third-party data there and so on. I mean, that's an area I'd like to kind of get into a bit in a bit, actually, because I know that Oracle's made quite a lot of acquisitions in that space, and I think certainly that seems to be a kind of a bit of a USP, really, for Oracle there. I mean, just before we get on to that, I mean, in terms of how how this is being developed so these are set of applications that are standalone to everything else that

oracle does are you looking to inject this kind of intelligence into into existing sort of like sas apps and how's that kind of work really yeah so this is this is a this is really interesting the way we've designed these as a set of REST APIs, right, what that means is we can make these pluggable into existing Oracle applications. So they can be exposed as widgets, they can be exposed as data feeds, depending on, you know, the use case. What that means also is that if somebody has a hybrid stack, and the hybrid stack, in that sense, having some Oracle and something from another vendor, we can provide that intelligence there as well. Because as long as they can map to our canonical model, which actually, if you go all the way back to those BI apps, turns out to be very similar, then we can have an interplay between the two.

So this REST API, the widgets that we drop into the systems or the data that can be called, that's really the key to all of this. Okay. Okay. And so we talked a bit about you making the life of the person operating the application kind of easier and simpler and more focused. Are there ways that you're also applying machine learning analytics into, for example, the offers that are being made to customers or kind of things like personalization?

Is that part of the project as well? Yeah. So when you get into the individual capabilities, for example, for CX, for CX, we have two different flavors. We have one that's around consumer-based customer experience. We have one around businesses, right? In consumer, it embodies just what people are used to, product recommendations, offers, right? So discounts or things like that. And content personalization as well. So, you know, and that could be banner images and the entire thing that happens to me when I'm on a website,

but it may also be, hey, read this document or watch this video. The interesting thing is that because it's a REST API layer, it can be inside the website, it can be in your email, we're doing work with chatbots, we're doing work with the service people that you may end up, you know, a human on the other side. So you get this continuous connected experience in terms of that personalization. Even if you go into the store, the people at the, you know, helping you inside of a store actually can have the benefits of that coordinated response okay okay so so i mean you mentioned uh offers there and so on i remember there was a product

called real-time decisions that was part of oracle bi before i mean how does this relate to that is there kind of any common technology or common kind of goals there at all i mean what's the kind of what's the situation on that yeah it's a good question. So RTD, real-time decisions, is really a development platform for people to be able to build these types of applications or other types of applications. What we wanted to do is not re-implement, but basically take the learnings out of RTD, which has been a really successful product and continues to be so.

But take those learnings and then tie it to very specific use cases and use the latest advances in terms of machine learning and all of the cloud-based technologies, the cloud architectures, so that we could build something that can scale with the businesses. And so we've taken the learnings of RTD. We've also taken the learnings of the BI apps. We've taken the learnings of the industry because our people, you know, for example, we compete our algorithm in Kaggle every so often, right? So we bring all those learnings together to this next generation of applications. Okay. Okay. So you touched a bit on third-party data and data

to the service there. And I think that it strikes me that's the kind of the additional unique thing that Oracle can bring to this. So tell us a bit about where third-party data comes in and a bit about what Oracle offer in this area in terms of products and acquisitions and so on. Yeah, so look at that personalization use case, that personalization use case about me interacting with, I don't know, a sporting goods store, things like that, right? Normally, that sporting goods store only knows about the information about me in terms of the context of my interactions with them. advertising and marketing data about consumers that people today use for ads, but we can also use for context in order to train and provide better precision to the application. And, you know, so it not only knows, you know, about me and my purchase history, but maybe some other advertising type context information. It may not be about me as a person because it's anonymous,

but sort of this class of person is doing certain things. It turns out that Oracle over the past several years has through acquisition as well as expansion of our capabilities, has the largest conglomeration and pool of this advertiser marketing data in the world. We did a series of acquisitions, starting with a company called BlueKai, and then we bought a company called Datalogix, one called AddThis, one called Crosswise, and even a fifth one now called Moat.

Yes, I heard about that. They come together to form what we call yeah the oracle data cloud and that's this capability uh advertisers use it you know so some of the biggest brands in the world use that information to to the place advertising and to target advertising we use exactly the same data through exactly the same apis as the advertisers to be able to then provide additional context to our recommendation engines or to our machine learning. So how does this, in fact there's a lot of data out there that is, some of it is probably, you've maybe got sort of different IDs for the customer, you've got

kind of challenges there in kind of linking it together. I mean, how do you link days together? I mean, I saw a product called Oracle ID Graph that's around. How does that fit into things? Yeah, so the Oracle ID Graph is interesting. So the way the data cloud works and any of these sort of third-party consumer systems work is the profiles are all anonymous. So each one of us may have five to ten. In fact, the people on this phone may have 15 or on this podcast may have 15. And it's, you know, your work email, your personal email, your Comcast account, your phone account.

All of this information comes together. The problem is actually understanding, hey, wait a second, these seven profiles are actually referring to one person, like for Jack or for Mark. So Oracle has some machine learning capabilities, some artificial intelligence capabilities that allow us to infer those matches. Now, it's probabilistic. It's not an absolute. We don't know it's exactly me, but that this unique identifier is strongly correlated with these others, and we can then pull that information that's really the id graph so you can come in from any one of those different touch

points from my phone or from my work identifier and actually get a consolidated view of this anonymous profile so so so you mentioned about these data sources going into the machine learning algorithms and so on is this all kind of transparent to the user? I mean, presumably this just happens in your models are kind of more predictive and so on. I mean, how do you take away, I suppose, the complexity of this really? You've got APIs and so on, but what work have you guys done to make this kind of easy to work with? Well, actually, again, another learning is we watched and we learned from numerous data lake and big data projects over the past several years. And this idea of each company needing to contract, each company needing to build their own algorithms, each company needing to build their own data science teams. Our customers are like, we can't even find the people, let alone hire them, let

alone keep them once we train them. And so what we've decided to do is to, for very specific use cases, remove that heavy burden. So we're investing in putting together all that infrastructure. Essentially, what we have is a massive data lake with the data science algorithms and everything ready to go. The system automatically adapts through some ensemble modeling and some dynamic, I don't want to overstep it, but dynamic machine learning to auto-tune to each customer's situation. And we sort of take away the need for a lot of that heavy lifting for those specific use cases.

Okay, okay, okay. That's interesting. So, yeah, I suppose the other part to all this is I've seen a lot of announcements being made by, say, Salesforce, for example, in areas in this way as well. So there's obviously Salesforce Einstein and so on. I mean, how does what Oracle is doing differ to that? And are you trying to achieve the same thing or different things to say what Salesforce is doing? Yeah, it's a great question. And there are more, you know, names for things out there than anybody actually understands what it is. I think we're trying to attack this slightly

differently than everybody else. We're trying to attack it really from a combination of a company's data as well as, you know, third-party data or even second-party data and bringing that together. We're also really attacking the notion of all of this intelligence needs to be connected. So it's not around the application, but actually around the person or the object that is being serviced. So I don't care that I happen to be right now on an e-commerce website versus email versus on a service desk versus in the store. It's one central connected intelligence across those. That appears to be different than our competitors. The other thing is that we come in with full knowledge of, again, getting back to the BI apps legacy, full knowledge of the domain.

So a lot of the machine learning techniques or groups will come in and say, great, describe to me your business problem from scratch, the big services engagement. We're coming in and saying, look, we actually know this business problem. We actually know everything about procurement, or we know a lot of things about transportation. We can tweak it a little bit with extensions for you, but you can get up and running with us in a matter of days, not a matter of weeks, not a matter of months, but a matter of days, you could be getting returns on those use use cases and so i think that's a little bit different um and we're pleased the reception from the market has been great and so we're happy with that approach excellent yeah i mean that's what prompted me to contact you really that i've seen

there's been a lot of press really what you're doing which is good um and you seem to be kind of getting somewhere really with this which is fantastic i mean longer term is this something i mean you guys i think you're you know it's pretty fair to say you're working as a kind of like a fairly small almost like startup team really with an oracle you know is it is the idea that this kind of in a way infuses everything oracle does in the future or or is it going to be a sort of niche thing or what yeah so you know i can't really talk too much about the future in terms of where everything is going to go, because, you know, quite frankly, we're learning the same way as everybody else is.

But I think what you'll see is that this notion of a data-driven application, this notion of a data-first or AI-first approach to the world, will permeate applications in general. Now, that doesn't alleviate the need for some things need to be transactional applications, right? There's a reason why a relational database exists, for example, in certain use cases. But what that means is that applications, whether they're from us or from anyone else, people are just going to expect that these capabilities are there and so you know is does that mean a whole re-architecture i don't think it does does that mean that this is pervasive yeah yeah you're going to see this type of uh these types of capabilities

pervasive over the next three to five years yeah fantastic i mean do you i'm just as a kind of side point do you see we're back in your old days of, I suppose, the BI tools and so on, do you see there being much of an application of machine learning in just things like BI tools in general, or is it more, the value of it is more focused when you get this kind of thing that you're doing? Well, I think in the BI world, I think there's certainly at least two or three places where machine learning is having an impact today, right? First of all, just the ability to get some advanced analytics, you know, regressions and projections and all of these types of things. Those algorithms can be applied right there. And instead of, you know, I remember trying to set one up, you know, 10 years ago, you know, one of the competitor's products, and it took me a month, right, to get a regression set up.

You know, just having those at your fingertip is important. I think you're also going to see machine learning just sort of aiding the analyst in BI, right? So, A, suggesting joins, or B, you know, suggesting proper visualizations. We actually had a capability in Oracle Business Intelligence that I launched back in 2012 with suggested visualizations for certain data sets. Remember that. Right? So you're going to see those types of things.

I think the third thing, which is really exciting, is sort of that natural interaction, whether it's to talk to the system, which is interesting to me, but more importantly, to explain what it means in terms of business analytics. That explanation stuff is, I think that's really exciting. So I don't think analytics, a classic sense goes away, right? I think it's morphing and it's changing. I think the types of things that we're doing and Salesforce is doing and others are doing in terms of embedding machine learning and recommendations and automation into applications is also going to continue. So I think they're, again, complementary. And one last thing.

Obviously, you remember the days of writing ETL routines and so on into the BI applications. The whole area of ETL is one that is still fairly, very manual, actually. Do you imagine, do you see a kind of machine learning or stats helping with that at all? I know there's a product from Amazon, Amazon Glue, that tries to do this.

Do you think that's an area that's ripe for this kind of innovation? Yeah, so there's two levels to it. You know, first of all, you know, the ETL stacks are radically changing, right? And so there's machine learning and capabilities there for getting things into, you know, sort of models and canonical models. The other thing is there's a few things that we were doing, you know, 10, 15 years ago in the semantic web with extremely loosely coupled schemas using ontologies. So where you can reinterpret data information. I think you're going to see those techniques.

I'm not sure that I would overstep and say that's machine learning or AI, although some people claim ontologies are AI, which I think flatters them. But I think you're going to see more and more loosely coupled schema approaches to these types of problems. As an end user, I shouldn't care that somebody didn't get the schema arranged properly. I still want to ask the question. And so if I can have a looser coupled approach to doing that, so much the better. Excellent. Excellent. Well, look, how would somebody get – a lot of people wouldn't have heard these applications.

How would people kind of get demos of these? Who would they speak to? Who would they approach to find out more about what you're doing and the kind of applications that you're building? Yeah, it's a great question. So obviously the Oracle salespeople are getting, you know, armed to the hilt with information about it. Oracle themselves on the website, there's, you know, a section about adaptive intelligent apps.

Modern CX, at our Modern CX conference, there were a couple of interesting keynotes given. One by Mark Hurd, our CEO, and I got a little cameo in the middle of it. And then one by Laura Ibsen, who's the head of our Oracle Marketing Cloud. She and Steve Krause, who's the VP of products for our Marketing cloud, gave some great examples of these applications in place. So good ability there. We're going to continue to push things out. So there'll be a lot of stuff on YouTube.

There'll be a lot of stuff coming out on Twitter. And then if anybody has a question, they can't get to it, find me on LinkedIn or find me on Twitter, JP Berkowitz on Twitter, and I'll get someone in touch with you. Excellent. Well, thanks, Jack. It's been great speaking to you.

It's been a while since I spoke to you last, so I think what you guys are doing is really interesting, and I think applying the kind of unique property, the unique kind of things that Oracle can bring to it, the fact that you've got the domain knowledge, the fact that you can build stuff

and it will appear on people's desktops because obviously people buy your things, and the data you've got as well, you know, and the machine learning kind of skills is, is really interesting. So, yeah,

thanks very much for coming on the show. I really appreciate that and take care and thanks very much. Thank you, Mark. It's been great and look forward to seeing you in person. Yeah. Cheers. Thanks Jack.