>> Narrator: From Berlin, Germany, it's theCUBE! Covering Dataworks Summit Europe 2018. Brought to you by Hortonworks. >> Well, hello. Welcome to theCUBE. I'm James Kobielus. I'm lead analyst at Wikibon for big data analytics. Wikibon, of course, is the analyst team inside of SiliconANGLE Media. One of our core offerings is theCUBE and I'm here with Joe Morrissey. Joe is the VP for International at Hortonworks and Hortonworks is the host of Dataworks Summit. We happen to be at Dataworks Summit 2018 in Berlin! Berlin, Germany. And so, Joe, it's great to have you. >> Great to be here! >> We had a number of conversations today with Scott Gnau and others from Hortonworks and also from your customer and partners. Now, you're International, you're VP for International. We've had a partner of yours from South Africa on theCUBE today. We've had a customer of yours from Uruguay. So there's been a fair amount of international presence. We had Munich Re from Munich, Germany. Clearly Hortonworks is, you've been in business as a company for seven years now, I think it is, and you've established quite a presence worldwide, I'm looking at your financials in terms of your customer acquisition, it just keeps going up and up so you're clearly doing a great job of bringing the business in throughout the world. Now, you've told me before the camera went live that you focus on both Europe and Asia PACS, so I'd like to open it up to you, Joe. Tell us how Hortonworks is doing worldwide and the kinds of opportunities you're selling into. >> Absolutely. 2017 was a record year for us. We grew revenues by over 40% globally. I joined to lead the internationalization of the business and you know, not a lot of people know that Hortonworks is actually one of the fastest growing software companies in history. We were the fastest to get to $100 million. Also, now the fastest to get to $200 million but the majority of that revenue contribution was coming from the United States. When I joined, it was about 15% of international contribution. By the end of 2017, we'd grown that to 31%, so that's a significant improvement in contribution overall from our international customer base even though the company was growing globally at a very fast rate. >> And that's also not only fast by any stretch of the imagination in terms of growth, some have said," Oh well, maybe Hortonworks, "just like Cloudera, maybe they're going to plateau off "because the bloom is off the rose of Hadoop." But really, Hadoop is just getting going as a market segment or as a platform but you guys have diversified well beyond that. So give us a sense for going forward. What are your customers? What kind of projects are you positioning and selling Hortonworks solutions into now? Is it a different, well you've only been there 18 months, but is it shifting towards more things to do with streaming, NiFi and so forth? Does it shift into more data science related projects? Coz this is worldwide. >> Yeah. That's a great question. This company was founded on the premise that data volumes and diversity of data is continuing to explode and we believe that it was necessary for us to come and bring enterprise-grade security and management and governance to the core Hadoop platform to make it really ready for the enterprise, and that's what the first evolution of our journey was really all about. A number of years ago, we acquired a company called Onyara, and the logic behind that acquisition was we believe companies now wanted to go out to the point of origin, of creation of data, and manage data throughout its entire life cycle and derive pre-event as well as post-event analytical insight into their data. So what we've seen as our customers are moving beyond just unifying data in the data lake and deriving post-transaction inside of their data. They're now going all the way out to the edge. They're deriving insight from their data in real time all the way from the point of creation and getting pre-transaction insight into data as well so-- >> Pre-transaction data, can you define what you mean by pre-transaction data. >> Well, I think if you look at it, it's really the difference between data in motion and data at rest, right? >> Oh, yes. >> A specific example would be if a customer walks into the store and they've interacted in the store maybe on social before they come in or in some other fashion, before they've actually made the purchase. >> Engagement data, interaction data, yes. >> Engagement, exactly. Exactly. Right. So that's one example, but that also extends out to use cases in IoT as well, so data in motion and streaming data, as you mentioned earlier since become a very, very significant use case that we're seeing a lot of adoption for. Data science, I think companies are really coming to the realization that that's an essential role in the organization. If we really believe that data is the most important asset, that it's the crucial asset in the new economy, then data scientist becomes a really essential role for any company. >> How do your Asian customers' requirements differ, or do they differ from your European cause European customers clearly already have their backs against the wall. We have five weeks until GDPR goes into effect. Do many of your Asian customer, I'm sure a fair number sell into Europe, are they putting a full court, I was going to say in the U.S., a full court press on complying with GDPR, or do they have equivalent privacy mandates in various countries in Asia or a bit of both? >> I think that one of the primary drivers I see in Asia is that a lot of companies there don't have the years of legacy architecture that European companies need to contend with. In some cases, that means that they can move towards next generation data-orientated architectures much quicker than European companies have. They don't have layers of legacy tech that they need to sunset. A great example of that is Reliance. Reliance is the largest company in India, they've got a subsidiary called GO, which is the fastest growing telco in the world. They've implemented our technology to build a next-generation OSS system to improve their service delivery on their network. >> Operational support system. >> Exactly. They were able to do that from the ground up because they formed their telco division around being a data-only company and giving away voice for free. So they can in some extent, move quicker and innovate a little faster in that regards. I do see much more emphasis on regulatory compliance in Europe than I see in Asia. I do think that GDPR amongst other regulations is a big driver of that. The other factor though I think that's influencing that is Cloud and Cloud strategy in general. What we've found is that, customers are drawn to the Cloud for a number of reasons. The economics sometimes can be attractive, the ability to be able to leverage the Cloud vendors' skills in terms of implementing complex technology is attractive, but most importantly, the elasticity and scalability that the Cloud provides us, hugely important. Now, the key concern for customers as they move to the Cloud though, is how do they leverage that as a platform in the context of an overall data strategy, right? And when you think about what a data strategy is all about, it all comes down to understanding what your data assets are and ensuring that you can leverage them for a competitive advantage but do so in a regulatory compliant manner, whether that's data in motion or data at rest. Whether it's on-prem or in the Cloud or in data across multiple Clouds. That's very much a top of mind concern for European companies. >> For your customers around the globe, specifically of course, your area of Europe and Asia, what percentage of your customers that are deploying Hortonworks into a purely public Cloud environment like HDInsight and Microsoft Azure or HDP inside of AWS, in a public Cloud versus in a private on-premises deployment versus in a hybrid public-private multi Cloud. Is it mostly on-prem? >> Most of our business is still on-prem to be very candid. I think almost all of our customers are looking at migrating, some more close to the Cloud. Even those that had intended to have a Cloud for a strategy have now realized that not all workloads belong in the Cloud. Some are actually more economically viable to be on-prem, and some just won't ever be able to move to the Cloud because of regulation. In addition to that, most of our customers are telling us that they actually want Cloud optionality. They don't want to be locked in to a single vendor, so we very much view the future as hybrid Cloud, as multi Cloud, and we hear our customers telling us that rather than just have a Cloud strategy, they need a data strategy. They need a strategy to be able to manage data no matter where it lives, on which tier, to ensure that they are regulatory compliant with that data. But then to be able to understand that they can secure, govern, and manage those data assets at any tier. >> What percentage of your deals involve a partner? Like IBM is a major partner. Do you do a fair amount of co-marketing and joint sales and joint deals with IBM and other partners or are they mostly Hortonworks-led? >> No, partners are absolutely critical to our success in the international sphere. Our partner revenue contribution across EMEA in the past year grew, every region grew by over 150% in terms of channel contribution. Our total channel business was 28% of our total, right? That's a very significant contribution. The growth rate is very high. IBM are a big part of that, as are many other partners. We've got, the very significant reseller channel, we've got IHV and ISV partners that are critical to our success also. Where we're seeing the most impact with with IBM is where we go to some of these markets where we haven't had a presence previously, and they've got deep and long-standing relationships and that helps us accelerate time to value with our customers. >> Yeah, it's been a very good and solid partnership going back several years. Well, Joe, this is great, we have to wrap it up, we're at the end of our time slot. This has been Joe Morrissey who is the VP for International at Hortonworks. We're on theCUBE here at Dataworks Summit 2018 in Berlin, and want to thank you all for watching this segment and tune in tomorrow, we'll have a full slate of further discussions with Hortonworks, with IBM and others tomorrow on theCUBE. Have a good one. (upbeat music)

>> Narrator: From Berlin, Germany, it's theCUBE, covering Dataworks Summit Europe 2018. Brought to you by Hortonworks. >> Well hello, welcome to the Cube. I'm James Kobielus, I'm the lead analyst for the Wikibon team within SiliconANGLE Media. I'm your host today here at Dataworks Summit 2018 in Berlin, Germany. We have one of Hortonworks' customers in South America with us. This is Fernando Lopez of Quanam. He's based in Montevideo, Uruguay. And he has won, here at the conference, he and his company have won an award, a data science award so what I'd like to do is ask Fernando, Fernando Lopez to introduce himself, to give us his job description, to describe the project for which you won the award and take it from there, Fernando. >> Hello and thanks for the chance >> Great to have you. >> I work for Quanam, as you already explained. We are about 400 people in the whole company. And we are spread across Latin America. I come from the kind of headquarters, which is located in Montevideo, Uruguay. And there we have a business analytics business unit. Within that, we are about 70 people and we have a big data and artificial intelligence and cognitive computing group, which I lead. And yes, we also implement Hortonworks. We are actually partnering with Hortonworks. >> When you say you lead the group, are you a data scientist yourself, or do you manage a group of data scientists or a bit of both? >> Well a bit of both. You know, you have to do different stuff in this life. So yes, I lead implementation groups. Sometimes the project is more big data. Sometimes it's more data science, different flavors. But within this group, we try to cover different aspects that are related in some sense with big data. It could be artificial intelligence. It could be cognitive computing, you know. >> Yes, so describe how you're using Hortonworks and describe the project for which you won, I assume it's a one project, for which you won the award, here at this conference. >> All right, yes. We are running several projects, but this one, the one about the prize, is one that I like so much because I'm actually a bioinformatics student so I have a special interest in this one. >> James: Okay. >> It's good to clarify that this was a joint effort between Quanam and GeneLifes. >> James: Genelabs. >> GeneLifes. >> James: GeneLifes. >> Yes, it's genetics and bioinformatics company. >> Right. >> That they specialize-- >> James: Is that a Montevideo based company? >> Yes. In a line, they are a startup that was born from the Institut Pasteur, but in Montevideo and they have a lot of people, who are specialists in bioinformatics, genetics, with a long career in the subject. And we come from the other side, from big data. I was kind of in the middle because of my interest with bioinformatics. So something like one year and a half ago, we met both companies. Actually there is a research, an innovation center, ICT4V. You can visit ICT4V.org, which is a non-profit organization after an agreement between Uruguay and France, >> Oh okay. >> Both governments. >> That makes possible different private or public organizations to collaborate. We have brainstorming sessions and so on. And from one of that brainstorming sessions, this project was born. So, after that we started to discuss ideas of how to bring tools to the medical genetiticists in order to streamline his work, in order to put on the top of his desktop different tools that could make his work easier and more productive. >> Looking for genetic diseases, or what are they looking for in the data specifically? >> Correct, correct. >> I'm not a geneticist but I try to explain myself as good as I can. >> James: Okay, that's good. You have a great job. >> If I am-- >> If I am the doctor, then I will spend a lot of hours researching literature. Bear in mind that we have nearly 300 papers each day, coming up in PubMed, that could be related with genetics. That's a lot. >> These are papers in Spanish that are published in South America? >> No, just talking about, >> Or Portuguese? >> PubMed from the NIH, it's papers published in English. >> Okay. >> PubMed or MEDLINE or-- >> Different languages different countries different sources. >> Yeah but most of it or everything in PubMed is in English. There is another PubMed in Europe and we have SciELO in Latin America also. But just to give you an idea, there's only from that source, 300 papers each day that could be related to genetics. So only speaking about literature, there's a huge amount of information. If I am the doctor, it's difficult to process that. Okay, so that's part of the issue. But on the core of the solution, what we want to give is, starting from the sequence genome of one patient, what can we assert, what can we say about the different variations. It is believed that we have around, each one of us, has about four million mutations. Mutation doesn't mean disease. Mutation actually leads to variation. And variation is not necessarily something negative. We can have different color of the eyes. We can have more or less hair. Or this could represent some disease, something that we need to pay attention as doctors, okay? So this part of the solution tries to implement heuristics on what's coming from the sequencing process. And this heuristics, in short, they tell you, which is the score of each variant, variation, of being more or less pathogenic. So if I am the doctor, part of the work is done there. Then I have to decide, okay, my diagnosis is there is this disease or not. This can be used in two senses. It can be used as prevention, in order to predict, this could happen, you have this genetic risk or this could be used in order to explain some disease and find a treatment. So that's the more bioinformatics part. On the other hand we have the literature. What we do with the literature is, we ingest this 300 daily papers, well abstracts not papers. Actually we have about three million abstracts. >> You ingest text and graphics, all of it? >> No, only the abstract, which is about a few hundred words. >> James: So just text? >> Yes >> Okay. >> But from there we try to identify relevant identities, proteins, diseases, phenotypes, things like that. And then we try to infer valid relationships. This phenotype or this disease can be caused because of this protein or because of the expression of that gene which is another entity. So this builds up kind of ontology, we call it the mini-ontology because it's specific to this domain. So we have kind of mini-semantic network with millions of nodes and edges, which is quite easy to interrogate. But the point is, there you have more than just text. You have something that is already enriched. You have a series of nodes and arrows, and you can query that in terms of reasoning. What leads to what, you know? >> So the analytical tools you're using, they come from, well Hortonworks doesn't make those tools. Are they coming from another partner in South America? Or another partner of Hortonworks' like an IBM or where does that come from? >> That's a nice question. Actually, we have an architecture. The core of the architecture is Hortonworks because we have scalability topics >> James: Yeah, HDP? >> Yes, HDFS, High-von-tessa, Spark. We have a number of items that need to be easily, ultra-escalated because when we talk about genome, it's easy to think about one terrabyte per patient of work. So that's one thing regarding storage and computing. On the other hand, we use a graph database. We use Neo4j for that. >> James: Okay the Neo4j for graph. The Neo4j, you have Hortonworks. >> Yes and we also use, in order to process natural language processing, we use Nine, which is based here in Berlin, actually. So we do part of the machine learning with Nine. Then we have Neo4j for the graph, for building this semantic network. And for the whole processing we have Hortonworks, for running this analysis and heuristics, and scoring the variance. We also use Solr for enterprise search, on top of the documents, or the conclusions of the documents that come from the ontology. >> Wow, that's a very complex and intricate deployment. So, great, in terms of the takeaways from this event, we only just have a little bit more time, what of all the discussions, the breakouts and the keynotes did you find most interesting so far about this show? Data stewardship was a theme of Scott Knowles, with that new solution, you know, in terms of what you're describing as operational application, have you built out something that can be deployed, is being deployed by your customers on an ongoing basis? It wasn't a one-time project, right? This is an ongoing application they can use internally. Is there a need in Uruguay or among your customers to provide privacy protections on this data? >> Sure. >> Will you be using these solutions like the data studio to enable a degree of privacy, protection of data equivalent to what, say, GDPR requires in Europe? Is that something? >> Yes actually we are running other projects in Uruguay. We are helping the, with other companies, we are helping the National Telecommunications Company. So there are security and privacy topics over there. And we are also starting these days a new project, again with ICT4V, another French company. We are in charge of their big data part, for an education program, which is based on the one laptop per child initiative, from the times of Nicholas Negroponte. Well, that initiative has already 10 years >> James: Oh from MIT, yes. >> Yes, from MIT, right. That initiative has already 10 years old in Uruguay, and now it has evolved also to retired people. So it's a kind of going towards the digital society. >> Excellent, I have to wrap it up Fernando, that's great you have a lot of follow on work. This is great, so clearly a lot of very advanced research is being done all over the world. I had the previous guest from South Africa. You from Uruguay so really south of the Equator. There's far more activity in big data than, we, here in the northern hemisphere, Europe and North America realize so I'm very impressed. And I look forward to hearing more from Quanam and through your provider, Hortonworks. Well, thank you very much. >> Thank you and thanks for the chance. >> It was great to have you here on theCUBE. I'm James Kobielus, we're here at DataWorks Summit, in Berlin and we'll be talking to another guest fairly soon. (mood music)

(upbeat music) >> Announcer: Live from the Javits Center in New York City. It's the Cube. Covering Inforum 2017. Brought to by Infor. >> Welcome back to the cube's coverage of Inforum here at the Javits center in New York City. I'm your host Rebecca Knight along with my co-host Dave Vellante, and Jim Kobielus who is the lead analyst for Wikibon in AI. So guys we're wrapping up day one of this conference. What do we think? What did we learn? Jim you've been, we've been here at the desk, interviewing people, and we've certainly learned a lot from them, but you've been out there talking to people, and off the record I should say. >> Yeah. >> So give us. >> I'm going to name names. >> Yes. >> If I may, I want to clarify something. >> Yeah, okay, sorry. >> I said this morning that the implied valuation was like three point seven, three point eight billion. >> Rebecca: Okay. >> Charles Phillips indicated to us off camera actually it was more like 10 and a half billion. >> Yeah, yeah. >> But I still can't make the math work. So I'm working on that. >> Okay. >> I suspect what's happened, was that a pre debt number. Remember they have a lot of debt. >> Yes. >> So I will figure it out, find out, and report back, okay. >> You do. >> So I just wanted to clarify that. >> Run those numbers okay. >> I'll call George. >> Kay, right, but Jim back to you. What do think is the biggest impression you have of the day in terms of where Infor is? >> Yeah, I've had the better part of this day to absorb the Coleman announcement which of course, ya know AI is one my core focus areas at Wikibon, and it really seems to me that, well Infor's direct competitors are the ERP space of all in cloud it's SAP, it's Oracle, it's Microsoft. They all have AI investments strategies going for in their ERP portfolios. So I was going back, and doing my own research today, just to get my head around where does Coleman put Infor in the race, cause it's a very competitive race. I referred to it this morning maybe a little bit extremely as a war of attrition, but what I think is that Coleman represents a milestone in the development of the ERP cloud, ERP market. Where with SAP, Oracle, and Microsoft, they're all going deep on AI and ERP, but none of them has the comprehensive framework or strategy to AI enable their suites for human augmentation, ya know, natural language processing, conversational UI's, Ya know, recommenders in line to the whole experience of ya know inventory management, and so forth. What infor has done with Coleman is laid out a, more than just a framework and a strategy, but they've got a lot of other assets behind the whole AI first strategy, that I think will put in them in good steady terms of innovating within their portfolio going forward. One of which is they've got this substantial infusion of capital from coke industries of course, and coke is very much as we've heard today at this show very much behind where the infor team under Charles is going with AI enabling everything, but also the Burst team is now on board with it, and the acquisition closed last month Brad Peters spoke this morning, and of course he spoke yesterday at the analyst pre-brief, and so David and I have more than 24 hours to absorb, what they're saying about where Burst fits into this. Burst has AI assets all ready. That, ya know Infor is very much committed to converging the best of what Burst has with where Coleman is going throughout their portfolio. What Infor announced this morning is all of that. Plus the fact that they've already got some Colemanize it's a term I'm using, applications in their current portfolio. So it's not just a future statement of direction. It's all that they've already done. Significant development and productization of Coleman, and they've also announced a commitment Infor with in the coming year, to bring, to introduce Coleman features throughout each of the industry vertical suite, cloud suites, like I said, human augmentation, plus automation, plus assistants, that are ya know, chat bots sort of inline. In other words, Infor has a far more ambitious and I think, potentially revolutionary strategy to really make ERP, to take ERP away from the legacy of protecters that have all been based on deterministic business rules, that a thicket, a rickety thicket of business rules that need to be maintained. Bringing it closer to the future of cognitive applications, where the logic will be in predictive, and deterministic, predictive, data driven algorithms that are continually learning, continually adapting, continually optimizing all interactions and transactions that's the statement of direction that I think that Infor is on the path to making it happen in the next couple of years in a way that will probably force SAP, Oracle, Microsoft to step up their game, and bring their cognitive or AI strategies in portfolios. >> So I want to talk some more about the horse in the track, but I want to still understand what it is. >> Jim: Yes. >> So the competitors are going to say is oh. It's Alexa. Okay, okay it is partially. >> Jim: Yeah sure. It's very reductive that's their job to reduce. >> Yeah you're right, you've lived that world for a while. Actually that was not your job, so. >> If you don't understand technology, you're just some very smart guy who talks a good talk. >> Yeah, okay. >> So, yeah. >> So, okay, so what we heard yesterday in the analyst meeting, and maybe you found this out today, was is conversational UX. >> Yes. >> It's chat wired into the APIs, and that's table stakes. It augments, it automates, an example is early payments versus by cash on hand. Should I take the early payment deal, and take the discount, or, and so it helps decide those decisions, and which can, if you have a lot of volume could be complex, and it advises it uncovers insights. Now what I don't know is how much of the IP is ya know, We'em defense essentially from Amazon, and how much is actual Infor IP, ya know. >> Good question, good question, whether it's all organically developed so far, or whether they've sourced it from partners, is an open issue. >> Question for Duncan Demarro. >> Duncan Demarra, exactly. >> Okay, so who are the horses in the track. I mean obviously there's Google, there's Amazon, there's I guess Facebook, even though they're not competing in the enterprise, there's IMB Watson, and then you mentioned Oracle, and SAP. >> Well, here's the thing. You named at least one of those solution providers, IBM for example, provides obviously a really sophisticated, cognitive AI suite under Watson that is not imbedded however, within an ERP application suite from that vendor. >> No it's purpose built for whatever. >> It's purpose built for stand alone deployment into all manner of applications. What Infor is not doing with Coleman, and they make that very clear, they're not building a stand alone AI platform. >> Which strategy do you like better. >> Do I like? They're both valid strategies. First of all, Infor is very much a sass vendor, going forward in that they don't they haven't given any indications of going into past. I mean that's why they've partnered with Amazon, for example. So it's clear for a sass vendor like Infor going forward to do what they've done which is that they're not going to allow their customers apparently to decouple the Coleman infrastructure from everything else that ya know, Infor makes money on. >> Which for them is the right strategy. >> Yeah, that's the right strategy for them, and I'm not saying it's a bad strategy for anybody who wants to be in Infor's market. >> So what is in Oracle, or in a SAP, or for that matter, a work day do, I mean service now made some AI announcements at their knowledge event. So they're spending money on that. I think that was organic IP, or I don't know maybe they're open swamps AI compenents. >> Sure, sure, A they need to have a cloud data platform that provides the data upon which to build and train the algorithm. Clearly Infor has cast a slot with AWS, ya know, SAP, Microsoft, Orcale, IBM they all have their own cloud platform. So >> And GT Nexus plays into that data corpus or? >> Yeah, cause GT Nexus is very much a commerce network, ya know, and there is EDI for this century, that is a continual free flowing, ever replenishing, pool of data. Upon which to build and train. >> Okay, but I interrupted you. You said number one, you need the cloud platform with data. >> Ya need the conversational UI, you know, the user reductive term chat bots, ya know, digital assistant. You need that technology, and it ya know, it's very much a technology in the works, its' not like. Everybody's building chat bots, doesn't mean that every customer is using them, or that they perform well, but chat bots are at the very heart of a new generation of application development conversational interfaces. Which is why Wikibon, why are are doing a study, on the art of building, and training, and tuning chat bots. Cause they are so fundamental to the UX of every product category in the cloud. >> Rebecca: And only getting more so. >> IOT, right, desk top applications. Everything's going with , moving towards more of a conversational interface, ya know. For starters, so you need a big data cloud platform. You need a chat bot framework, for building and ya know, the engagement, and ya know, the UI and all of that. You need obviously, machine learning, and deep learning capabilities. Ya know, open source. We are looking at a completely open source stack in the middle there for all the data. Ya know, you need obviously things like tenserflow for deep learning. Which is becoming the standard there. Things like Spark, ya know, for machine learning, streaming analytics and so forth. You need all that plumbing to make it happen, but you need in terms of ERP of course, you need business applications, and you need to have a business application stacked to infuse with this capability, and there's only a hardcore of really dominant vendors in that space. >> But the precious commodity seems to be data. >> Yeah. >> Right. >> Precious commodity is data both to build the algorithms, and an ongoing basis to train them. Ya see, the thing is training is just as important as building the algorithms cause training makes all the difference in the world between whether a predictive analytics, ya know ML algorithm actually predicts what it's supposed to predict or doesn't. So without continual retraining of the algorithms, they'll lose their ability to do predictions, and classifications and pattern recognitions. So, ya know, the vendors in the cloud arena who are in a good place are the Googles and the Facebooks, and others who generate this data organically as part of their services. Google's got YouTube, and YouTube is mother load of video and audio and so forth for training all the video analytics, all the speech recognition, everything else that you might want to do, but also very much, ya know, you look at natural language processing, ya know, text data, social media data. I mean everybody is tapping into the social media fire hose to tune all the NLP, ongoing. That's very, very important. So the vendor that can assemble a complete solution portfolio that provides all the data, and also very much this something people often overlook, training the data involves increasingly labeling the data, and labeling needs a hardcore of resources increasingly crowdsource to do that training. That's why companies like Crowd Flower, and Mighty AI, and of course Amazon with mechanical terf are becoming evermore important. They are the go to solution providers in the cloud for training these algorithms to keep them fit for purpose. >> Mmm, alright Rebecca, what are your thoughts as a sort of newbie to Infor. >> I'm a newbie yes, and well to be honest, yes I'm a newbie, and I have only an inch wide, an inch deep understanding of the technology, but one thing that has really resonated with me. >> You fake it really well. >> Well, thank you, I appreciate that, thank you. That I've really taken away from this is the difficulties of implementing this stuff, and this what you hear time and time again. Is that the technology is tough, but it's the change management piece that is what trips up these companies because of personalities who are resistant to it, and just the entrenched ways of doing things. It is so hard. >> Yes, change management, yes I agree, there's so many moving parts in these stacks, it's incredible. >> Rebecca: Yeah. >> If you we just focus on the moving parts that represent the business logic that's driving all of this AI, that's a governance mess in it's own right. Because what you're governing, I mean version controls and so forth, are both traditional business rules that drive all of these applications, application code, plus all of these predictive algorithms, model governance, and so forth, and so on. I mean just making sure that all of that is, you're controlling versions of that. You've got stewards, who are managing the quality of all that. Then it moves in lock step with each other so. >> Rebecca: Exactly. >> So when you change the underlying coding of a chat bot, for example, you're also making sure to continue to refresh and train, and verify that the algorithms that were built along with that code are doing their job, so forth. I'm just giving sort of this meta data, and all of that other stuff that needs to be managed in a unified way within, what I call, a business logic governance framework for cloud data driven applications like AI. >> And in companies that are so big, and where people are so disparately located, these are the biggest challenges that companies are facing. >> Yeah, you're going to get your data scientists in lets say China to build the deep learning algorithms, probably to train them, your probably going to get coders in Poland, or in Uruguay or somewhere else to build the code, and over time, there'll be different pockets of development all around the world, collaborating within a unified like dev ops environment for data science. Another focus for us by the way, dev ops for data science, over time these applications like any application, it'll be year after year, after year of change and change. The people who are building and tuning and tweaking This stuff now probably weren't the people five years ago, as this stuff gets older, who built the original. So you're going to need to manage the end to end life cycle, ya know like documentation, and change control, and all that. It's a dev ops challenge ongoing within a broader development initiative to keep this stuff from flying apart from the sheer complexity. >> Rebecca: Yes. >> So, just I don't Jim, if you can help me answer this, this might be more of a foyer sort of issue, but when we heard from the analyst meeting yesterday, Soma, their chief technical guy, who's been on the Cube before in New Orleans, very sharp dude, Two things that stood out. Remember that architecture slide, they showed? They showed a slide of the XI and the architecture, and obviously they're building on AWS cloud. So their greatest strengths are in my view, any way the achilles heel is here, and one is edge. Let's talk about edge. So edge to cloud. >> Jim : Yes. >> Very expensive to move data into the cloud, and that's where ya know, we heard today that all the analysis is going to be done, we know that, but you're really only going to be moving the needles, presumably, into the cloud. The haystacks going to stay at the edge, and the processing going to be done at the edge, it's going to be interesting to see how Amazon plays there. We've seen Amazon make some moves to the edge with snowball, and greenfield and things like that, and but it just seems that analytics are going to happen at the edge, otherwise it's going to be too expensive. The economic model doesn't favor edge to cloud. One sort of caveat. The second was the complexity of the data pipeline. So we saw a lot of AWS in that slide yesterday. I mean I wrote down dynamo DB, kineses, S3 redshift, I'm sure there's some EC2. These are all discreet sort of one trick pony platforms with a proprietary API, and that data pipeline is going to get very, very complex. >> Flywheel platforms I think when you were talking to Charles Phillips. >> But when you talk to Andy Jasse, he says look we want to have access to primitive access to those APIs. Cause we don't know what the markets going to do. So we have to have control. It's all about control, but that said, it's this burgeoning collection of at least 10 to 15 data services. So the end to end, the question I have is Oracle threw down the gauntlet in cloud. They said they'll be able to service any user request in a 150 milliseconds. What is the end to end performance going to be as that data pipeline gets more robust, and more complicated. I don't know the answer to that, but I think it's something to watch. Can you deliver that in under 150 milliseconds, can Oracle even do that, who knows? >> Well, you can if you deliver more of the actual logic, ya know, machine learning and code to the edge, I mean close the user, close to the point of decision, yes. Keep in mind that the term pipeline is ambiguous here. One one hand, it refers, in many people's minds to the late ya know, the end to end path of a packet for example, from source to target application, but in the context of development or dev ops it refers to the end to end life cycle of a given asset, ya know, code or machine learning, modeling and so forth. In context of data science in the pipeline for data science much of the training the whole notion of training, and machine learning models, say for predictive analysis that doesn't happen in real time in line to actual executing, that happens, Ya know, it happens, but it doesn't need it's not inline in a critical path of the performance of the application much of that will stay in the cloud cause that's massively parallel processing, of ya know, of tensorflow, graphs and so forth. Doesn't need to happen in real time. What needs to happen in real time is that the algorithms like tensorflow that are trained will be pushed to the edge, and they'll execute in increasingly nanoscopic platforms like your smartphone and like smart sensors imbedded in your smart car and so forth. So the most of the application logic, probabilistic ya know, machine learning, will execute at the edge. More of the pipeline functions like model building, model training and so forth, data ingest, and data discovery. That will not happen in real time, but it'll happen in the cloud. It need not happen in the edge. >> Kind of geeky topics, but still one that I wanted to just sort of bring up, and riff on a little bit, but let's bring it back up, and back into sort of. >> And this is the thing there's going to be a lot more to talk about. >> Geeking out Rebecca, we apologize. >> You do indeed, it's okay, it's okay. >> Dave indulges me. >> No, you love it too. >> Of course, no I learn every time I try to describe these things, and get smart people like Jim to help unpack it, and so. >> And we'll do more unpacking tomorrow at two day of Inforum 2017. Well, we will all return. Jim Kobielus, Dave Vellante, I'm Rebecca Knight. We will see you back here tomorrow for day two. (upbeat music)