Utilizing Tech - Season 7: AI Data Infrastructure Presented by Solidigm - 2x30: Bringing AI to the Executive Suite with Josh Epstein of AtScale

Episode Date: July 27, 2021

Businesses have long tried to use data to drive decisions, but over the last few years new big data and AI capabilities have appeared. In this episode, Josh Epstein of AtScale discusses the opportunities that enterprise AI brings to drive business decisions. Although executives might not know the details of AI models, they can certainly benefit from the forecasts and recommendations these tools deliver. One benefit of these systems is that they can bring in more diverse data to uncover real value from areas typically outside the sight of executives. Three Questions Are there any jobs that will be completely eliminated by AI in the next five years? Will we ever see a Hollywood-style “artificial mind” like Mr. Data or other characters? Can you think of an application for ML that has not yet been rolled out but will make a major impact in the future? Guests and Hosts Josh Epstein, Chief Marketing Officer at AtScale. Connect with Josh on LinkedIn. Frederic Van Haren, Founder at HighFens Inc., Consultancy & Services. Connect with Frederic on Highfens.com or on Twitter at @FredericVHaren. Stephen Foskett, Publisher of Gestalt IT and Organizer of Tech Field Day. Find Stephen’s writing at GestaltIT.com and on Twitter at @SFoskett.         Date: 7/27/2021 Tags: @AtScale, @SFoskett, @FredericVHaren

Welcome to Utilizing AI, the podcast about enterprise applications for machine learning, deep learning, and other artificial intelligence topics. Each episode brings in experts in enterprise infrastructure together to discuss applications of AI in today's business. Today, we're discussing bringing AI to the executive suite to make better business decisions. First, let's meet our guest, Josh Epstein. Hi, I'm Josh Epstein. I'm the Chief Marketing Officer here at AtScale, based in Boston and at AtScale's world headquarters. I've been here about six months, joined after being a serial CMO in the Boston area covering all sorts of enterprise tech. Frederick and I crossed paths in the storage world. I've also worked in networking and user computing.

I'm really thrilled to be at AtScale working in big data and analytics in AI. Hi, my name is Frederick Van Heren. I am the founder of Hyphens, which is a consulting and services company active in the HPC and AI market. And you can find me on LinkedIn and Twitter under Frederick V. Heren. And as always, I'm Stephen Foskett, your host here, organizer of Tech Field Day, including AI Field Day events, and publisher of Gestalt IT. You can find me on most social media channels, including Twitter at S Foskett. So we've spoken quite a lot on this podcast about the various applications of AI, everything from industrial IoT and automation to autonomous vehicles. But one thing we haven't

really talked about is driving business intelligence with AI. And so when we got a chance to sit down with Josh, we decided, you know, this is a topic I think that we would really want to bring to our audience. So Josh, talk to us a little bit about this. You know, how exactly can executives and, you know,-level management of companies, how exactly can they leverage AI? Yeah, so I'm a longtime data-driven marketer, and I think that marketers and go-to-market executives in general have been able to certainly leverage big data and analytics in their day-to-day and increasingly leverage ML and AI-driven models. We see it now integrated more and more into Salesforce. We certainly have our ops teams leveraging this, have some models and coming up with new ways that we can run our business.

You know, more broadly speaking, you know, AtScale is focused on really bringing both big data and AI to the business intelligence community and the executives that they support. So increasingly we're looking to bridge this massive world of data, all the opportunities that enterprise AI brings and really able to publish it more broadly through BI processes, through the SaaS applications we use to run our business. Yeah, so Joshua, you're talking a little bit about tools and applications and integrating in tools like Salesforce.com. Do you feel that the executives today understand AI enough

to make decisions to help AI move forward in organizations? So I think what is happening is that you have executives and line of business managers, more broadly starting to really accept things like a prediction or a forecast or a suggestion in their day to day. So you know, obviously, every industry is a little bit different, but things like forecasting inventory stocking values, right, of how, you know, how much inventory should you have in place at a given spot to respond to demand. And that's driven by, you know, basic ML models, regression, or what have you

to help there. In other instances, you have, you know, even down to the sales executive level, who's looking at their Salesforce dashboard, maybe getting suggestions on activities that would maximize the probability of close is another common example. So if you ask the average executive what their acceptance of AI is, maybe they might not know how to understand. But in terms of the suggestions, the predictions, the prescriptive analytics that come as part of business processes today, I think that they're more and more familiar with these concepts. Yeah, it really reminds me of some of the discussions we have about, well, many different aspects of AI, where people may not really

understand what the system is or how it's being used or how it functions. But yeah, to be able to say, you know, our tool suggests that we should stock up on widgets before the next quarter. I'm sure they can understand a recommendation like that. Is that the kind of recommendation they're getting out of these tools? Yeah. And, you know, what's really fascinating is you start to see this, you know, for a long time, AI and ML and sort of data science has been the realm of data scientists, right? You have these really smart folks in the room kind of running models, playing with models, ingesting massive amounts of data. But really the art and kind of bringing that to the boardroom or the executive suite, it's about how do you integrate those insights into the business processes, into the tools that support those business processes for day-to-day business processes and the tools that support those business processes for day-to-day business processes. So how do you bring it into the CRM system? How do you bring it

into the dashboard or the reporting application that folks are consuming data with? So it's really interesting. It's both a combination of the innovation coming out of the data science communities and obviously the cool tech that supports them, the AutoML platforms and the like. But then also thinking kind of strategically about how you bring that into business processes. So you're starting to bridge, you know, a lot of different worlds beyond just classic data science or AI experts. So do you feel that the AI initiatives are top-down or the other way around? Meaning, you know, you're talking about bringing data to the boardroom, but is the boardroom asking for the data or are people pushing that data to the boardroom so that the executives kind of understand what the data is telling them?

You know, I do think it comes from both directions. I think that, you know, certainly most leading organizations appreciate the power of AI, really of AI supported insights. And so the organization and a support to the organization in the form of budget and investment and priorities to go out and find new creative ways to bring AI and ML and kind of the broader spectrum of data science into the way organizations make decisions. So I do think it comes from both directions. Yeah, you saw you talking about budgets and so on. Do you feel that enterprises are spending enough money on AI and AI tools, so to speak, in order to help them out? Or do you feel that the budgets are staying the same in the hope that AI can do more with less, so to speak? You know, I do think that the budgets are increasing. I think, though, you know, like all emerging techs, this is really past the level of emerging. You know, the real effort is on showing return, right?

And so how do you actually scale and bring, you know, these insights to a broader audience? So I think the subject of bringing AI to the executive suite, to the boardroom, to the broader line of business is really important because that is going to be where value is delivered. I think we've gone through this initial phase of AI and data science where, again,

you have some really smart people able to show some really cool results on a sort of an experimental basis, on a proof of concept basis. But when you think about really scaling and kind of bringing these insights, publishing these insights in real time to a much broader community, both executives and line of business, that's really when the value starts happening. And so I think we're just starting to be in that stage. You see a lot of folks in that scale included, focused on not only making AI more accessible,

but more publishable, more shareable, more discoverable. I think that you have this situation where there's lots of AI generated insights out there. Now there's lots of predictions, lots of recommendations. It's about making them available when people go searching for them. So one of the things that we at AtScale really focus on is this idea of discoverability,

this idea of sharing both data driven insights, but also data augmented with AI generated insights out to the community. So when they're looking for a prediction, and when they're looking for a projection, when they're looking for support by a more sophisticated model, folks can find it, right? Whether they know the technical terms to use and searching in their data catalog, or they're more, you know, thinking more natural language, or they're more thinking, you know, more business oriented, and they're able to review and see what's available and really kind of get in touch with the AI model or the AI driven resource that best fits their

need. I'd like to follow that thread a little bit, because I think that that's actually an interesting opportunity. One of the challenges for business managers has long been sort of their mindset and limited exposure to understanding of the various aspects of the business. In other words, you know, maybe somebody was the CFO and they're going to come in with a CFO perspective. And sometimes that's good, you know, but sometimes that's a challenge because they may have a little bit of blinders on to other parts of the business.

And similarly, you see this in tech a lot where you've got a real techie founder who may not understand some of the production and customer service and even financial aspects of the business. In a way, AI could help break down some of these barriers by pulling in a real, you know, holistic data set and presenting answers that take into idea of creating a semantic layer or semantic model for both historical data and data assets that a company has data looks like in tables, data field names, it's extraordinarily confusing, right? And you have to be a data person to really be able to abstract the business significance out of raw data. So first step is really creating this common language, this common way to look at key enterprise business metrics. For instance, the right way to look at revenue, right. It seems obvious, but anyone who's worked in the data fields knows that feeling of when they show this analysis and then the person

in finance says, well, that's not actually a revenue. Because the fact is it's not simple to actually do that kind of analysis. So this idea of a semantic layer that brings business context to key metrics is key. And the same actually applies when you start thinking about how do you create a level of semantics for AI driven insights, or really kind of these models that could be predicted. So when a executive comes in and says, I want to know what the forecast is for sales next quarter in this region, you know, the semantics of next quarter in this region. The semantics of next quarter in this region are key in actually delivering the right kind of results.

So it's both a challenge of organizing that data and modeling that data, but then also providing the right infrastructure for the AI to come and do what it does best, which is make predictions based on valid inputs in the way. Yeah, you said something very interesting earlier about sharing. So typically, if you look at AI, the idea exists today is because a lot of people decided to share their data, right?

So as opposed to looking at enterprises being in silos, a lot of organizations can benefit from kind of going horizontal and sharing data. So when you talk about key metrics and having data and models, do you see enterprises having the willingness to share with other enterprises, even if it's their competitors,

just to come up with an AI model that suits better the industry. Yeah, I think, you know, we're growing. We talk a lot at scale about kind of first party, second party and third party data. First party data being obviously being the data from our own enterprise applications, right? Whether it's stored out in a data lake

or organized in some type of data warehouse, either way, that's, you know, your first party data set, which by the way, just sharing across a large enterprise is no simple feat, right? There's lots of different applications. So making that all available to the right folks is step number one. Second party data becomes really trading partners, right? So certainly if you're a large retailer, sharing information with your suppliers is key in order to make better supply chain decisions. So that's come a long way. And certainly some of the services provided by the large cloud

data providers make it easier to share this data in a secure but cost-effective and scalable manner. A really exciting piece, though, becomes in the world of third-party data, right? If, you know, these large data providers providing anything from, you know, point-of-party data, right? If, you know, these large data providers providing anything from, you know, point of interest data, foot traffic data on the retail side, certainly web statistics, certainly financial statistics. We're all familiar with the big Bloombergs and the S&Ps of the world,

which have been, you know, providing data for a long time. But more and more of this data is available on very consumable platforms. So an Amazon data exchange, Snowflake data marketplace. And so what you see is this, we're in this golden age of both being able to access data, first-party data and share first-party data, but they really scale up to second and third-party data. And the more data available, the better the models, the more inputs there are to these AI

driven models. So, you know, we're seeing a huge interest in expanding the type of data that can go into feature discovery, into feature engineering on the ML model side of things, and really having the capability expand both within the hardcore data science community that's, you know, sort of doing experimental models and looking for interesting correlations, but then also as you look into productizing those models or bringing those models from a development experimental phase into kind of a routine operational model so that if you find a statistic, it could be something as simple as census data, right? As you have new census data coming in, those models get updated and that might update

the forecast that's feeding the executives decision-making process. So absolutely a huge shift and really excitement in leveraging these third-party data sets. As you see enterprises putting a lot of focus on having the right data, do you feel that executives are willing to pay for the data? Is that even a business, buying data in order to have an advantage over your competitors? I mean, it seems like, and I talk from my own experience, is that large organizations that are behind

are more than willing to pay a lot of money to buy data from their competitors in the hope that they can leapfrog them. Yeah, I think it's both a willingness to invest in the data assets, but I think it's more the willingness and the excitement in harnessing the invest in the data assets, but I think it's more the willingness and the excitement and harnessing the value of the data. And that harnessing is, is both,

it's both related to the resources, the people that you have in place, the data scientists, and the, you know, say the sophisticated sales ops, marketing ops, business ops, folks that can kind of translate that data and those outputs of those models into business insights. But it's also the whole data pipeline or the whole data flow from raw data to the point of integrating it with other data sets

to then making it available and discoverable to the broader audience. So there's no shortage of investment. I think you can see from a lot of the valuations of whether it's data integration, data virtualization technologies, certainly the data platforms out there like Snowflake, lots of growth expected and continue to see a lot of both investment, but also experimentation. How do you actually optimize these infrastructures for insights to support the business? There are also some drawbacks here, and I think that it would be remiss of me not to point out some challenges of using machine learning to drive business decisions. One of the things that we've come up with in the podcast here

over the last couple of years that we've been doing this now is, of course, the challenges of biased models and the ethics of using AI when, you know, it's fed with models that are not truly representative of the entire population. You know, and this also does give concern. I mean, obviously, people are concerned when, you know, AI driven cars could drive off the road or something. But, you know, AI driven businesses could metaphorically drive off the road, couldn't they? I mean, if we're if we're feeding them with, you know, let's say you had a business intelligence system, and you fed it only the data from American consumers, and then you tried to apply that globally, that could really drive the business off the road

if the habits and buying power and buying decisions of American consumers are not the same as people in third world countries or in just other countries. And I can see all sorts of other, you know, kind of challenges that could happen if businesses were making decisions based on incomplete data. Yeah, that's a really interesting point. I think what I would say is we're not ready to take humans out of the equation, first and foremost, right? And autonomous vehicles is probably a good, you know, analogy.

You know, we're not quite at the point where a human can just totally close their eyes and start, you know, watching TV. Although they do, and you hear lots of horror stories about them running off the road. We frankly might even be closer there than we are in the real world of data and of enterprise insights being driven by NML. I think that the role of both data scientists and the folks that are building these models can't be understated. You need a person that is thinking really critically about data inputs, obviously, and applying their own intuition on the potential biases that might be coming from the inputs, obviously, and applying their own intuition on the potential biases that might be coming from the inputs, and then really looking critically at the outputs. And I think data scientists and kind of the technical folks running the models and executing the analyses are one

layer of defense there, kind of having them being sensitive and aware of the biases that can come from models fed with incomplete data. But even more important, we start talking about creating savvy executives that know how to consume good AI, having them ability to, number one, gut check the number. Any good executive does this anyway when they see whether it's a prediction or frankly even a statistic defining historical performance, they first check it with a gut right in the good executive good manager says okay doesn't quite feel right let me sort of probe a little more to understand what went into it. really think about businesses that are running big pieces of their business on AI-driven insights. I think it is on executives to think very critically about the biases that could creep in. You know, there could be biases that, you know, when you say the word bias, you think immediately in terms of socioeconomic biases or gender biases or things that are kind of morally or ethically

questionable and they need to watch out for. But even more generically, just the biases that might creep in that might tend to overstate results or understate results or miss certain suggestions in certain categories. Thinking very critically about data-driven insights and testing it with your gut, I think is more and more important as we have more and more AI-driven insights and data-driven insights, not less so. Does that make sense? Yeah, I would suspect that in the industry, there would be AI tools to validate AI data. In other words, AI basically relies on statistical applications and they do

sampling. I would not be surprised if there are tools being in the works that actually do exactly that, which take incoming data, sample it, and test it against other data sets that don't have the bias. And I presume that's more like on the consumer side where such a data exists. And I would also presume based on my previous question that organizations actually would sell you data that has been validated against bias. I mean, it's a real problem. But my point is, I would suspect that the AI industry and that AI executives should have the right attitude and making sure that where they're getting their data from is not from, you know, the guy on the corner of the street, but more from an organization that

sells you data for a particular market. Do you see some of those tools being developed? Well, I was just going to say too, that it seems like people are getting a little more savvy about this and understanding that the challenges of models and biased data, and hopefully those people are involved in this rollout. Well, and I think two observations about the topic, because it is really an interesting topic. One, it's becoming easier and easier to consume or find additional sources of third-party data. And again, if you take a look on Amazon Data Exchange, it's not like data marketplace, you can see competitive data sources, right? So the different ways of looking at, say, retail patterns or things like that. So with more data providers, with more

competition, there's the ability to both trial different sources, compare different sources, even tap into the kind of the idea or the innovation behind those data sources and make some decisions on which ones are best representative. And then certainly being able to try them and test them versus, you know versus results to find more accurate sources. But I still think more and more as we think about data savvy executives and AI savvy executives, it is about the ability to sort of think critically about the input and ask the right questions. You know, I've been, I grew up as a data analyst. I, you know, grew up as a playing with ML models and the like. And when you are a data scientist heads down playing with the data, it is sometimes difficult to sort of see the bigger picture.

So I think it's all the more important that we look towards executives and managers to be better AI consumers and being able to ask those questions to potentially spot those potential challenges and obstacles that come with poor data inputs. You know, Frederick, back to your original point of, you know, is there a play for AI-driven resource to kind of go out and validate that, absolutely. I think a large organizations that operationalize some of these AI driven insights, in fact, would be investing in their own or third party type tools to go out

and help validate and cleanse and spot potential sources of biases in their data. Because increasingly you're gonna see big organizations really rely tremendously on these insights. So having kind of a check mechanism in place to find potential issues before they bubble up into true bias that can hurt the competitive position of a company or even more concerning, really inject some damaging results or actions from a more ethical standpoint is key.

Yeah, and I think additionally to bias data, I think aging data is also key. An example of that is how people look and do searches on Google is majorly different than what people used to do four or five years ago. It's not biased, but should you keep the data forever? So it might look like great data at some point, but eventually the data will expire, right? Or should have an expiration date. I'm not sure if it fits in the whole bias, but it definitely kind of derails the statistical analysis for certain data.

And I think, you know, in the subject of what the AI models themselves can get better at doing, it can be supported by making a broader set of data available and a well-organized set of data available. These models can, in effect, you know, they can internally test, right? So if they see a certain feature that is no longer valid or no longer predictive, they can swap out in favor of another feature that perhaps has. So looking at both data sources as well as time periods that are relevant, I think, is on the AI model. What's phenomenal, if you look at the amount of ML power out there and classic enterprise AI platform,

they're extremely easy to consume. You know, there's a huge opportunity to tap into platforms like, you know, those from DataRobot or H2O or other kind of enterprise AI platforms. So lots of compute and AI horsepower out there, lots of data to feed into it. It's a matter of now having the right folks in the data science community to help build those models and think strategically about how they can create value for their company.

And then as we're talking about having executives and know how to consume those insights in a way and give feedback and ask questions on those insights in a way that you can create this flywheel effect and just have this ever-increasing focus on both the data, on the AI, and on the actions that come out of analyzing. So do you have a feeling on what barriers there are

for executives dealing with AI? Do you see resistance? Do you see acceptance? Where do they stand? What can we do to help those executives? Well, I don't see resistance per se, but I do think that everyone has only so much information they can take in.

So it starts with how do you actually organize these insights and the outputs of these models in a way that folks can consume? You have to think about the method and the channel of how to distribute this information. Do executives want to consume it in their BI dashboards? They want to pick it up in their application, in their Salesforce.com dashboards, they want to have it recommended to them in PowerPoint. So I think thinking about the mechanism of delivering the information is important. And I do think that AI consumption is really just the same as BI consumption here at Atscale. We talk a lot about this convergence of AI and BI. And really there it's about how do you consume this information in a way that,

in the timing that aligns to when you're making a decision, right? So how do you actually distribute these insights, these forecasts, these predictions, these suggestions in the same forums that you would distribute analysis of historical data, right? And so that can be in a BI dashboard, that can be in an application, that can be honestly in the presentation, the PowerPoint presentation being delivered by, you know, an analyst. So we talk a lot about data storytelling here, about how do you actually wrap a story and explanation around data points?

That data is both historical data as well as the outputs of these AI models. How do you actually wrap these statistics in a story that an executive can internalize, consume, and then make decisions on? So do you feel that if i would ask an executive what is ai you know what are the chances that i get a reasonable answer i mean it that's that's my gut feeling is it's there's a technology and then there is a messaging i without a doubt and i think the uh you know the term ai is certainly quite broad, right? It covers everything from autonomous vehicles to natural language processing, and as well as the more classic ML driven predictions, suggestions that come out of the classic data science models. So does the average executive know how AI is being integrated into their day to day? Maybe it depends, but I would bet that every single, you know,

Fortune 1000 executive out there has, looks at a forecast, looks at a suggestion and understands the power of getting more accurate in both their predictions, as well as the more micro predictions that their leaders and their business run their day-to-day operations on. Yeah, I think it's pretty unlikely that a lot of executives are really going to understand the nuances of, you know, oh, is this machine learning or is this, you know, is this deep learning or, but at the same time, you know, I think that they might appreciate just better

insights if that's possible. You know, for me, I think that's the big takeaway from this discussion is that idea that perhaps we can be integrating more sources of information, more clever information that helps them break out of their, you know, traditional mindset. I think that's, to me, the exciting power of AI in the executive suite. So before we move on, though, we do have a typical way that we like to finish our episodes. Since we've been going here for a little while, I think it's time to turn the page and ask

you a few unexpected questions. Are you ready for that, Josh? I think so. Bring it on. All right. the page and ask you a few unexpected questions. Are you ready for that, Josh? I think so. Bring it on. All right. So those of you listening, as always, we ask three questions that we have not prepared our guests for in hopes to get a little bit of special insight from them on the entire field of AI. I do try to tailor them to our guests and not ask you something that you're not the right person for, but let's try some of these, OK? So first of all, Josh, you've had a lot of jobs.

Are there any jobs that you know of that will be completely eliminated by AI? What is AI going to do? Nobody's going to be doing that thing anymore. You know, I'd like to think that AI and the utilities that sort of surround AI ML could really do away with data prep. I mean, I grew up as a data analyst kind of scaling up and down with SQL Server queries and Excel pivot tables and futzing around just to really prep my data, get my data into my laptop, prep it, and then feed it into a model or into some type of analysis. I'd like to think that AI can really do away with all that data prep and data wrangling.

Cool. Next up, will we ever see a Hollywood-style AI, an artificial mind like Mr. Data or any of these other guys? I am a huge sci-fi fan. I've seen, I think, anything from Singularity or I'm trying to think of the other good ones out there. I hope not. I do think it could be a little bit scary if we see a true artificial mind out there, particularly given the state of things and how much uncertainty there is in the world today. So I think that'll be pretty nerve wracking. But you know what? Who knows? Lots of nerve wracking things happen. So I think we all have to be eyes wide open with where this all goes. And finally, just think about all the different ways that you interact with the world, all the different people you come into contact with, all the different businesses that you're exposed to. Can you think of an application for machine learning that you've noticed that hasn't been revealed yet,

but you're like, yeah, right there, that's where they're going to use AI to do something clever. That is a good one. So I think we're already seeing signs of this. I mean, you know, and I know Frederick's background in natural language processing, and he must just be really with how far it's come. But really how much we integrate Alexa in our day to day lives or in, you know, Siri or, or whatever, you know, I, and this is not novel, but I really do think

that people will increasingly interact with data in a way that touches on natural language processing, whether it's truly just asking your Alexa for what's my sales going to be next quarter, or more accurately kind of leveraging all of the improved kind of semantics understanding that we have of all the data and AI out there to allow us to more naturally interact with these predictions is happening and is, I believe, is going to be the next kind of big, the next big thing that combines both AI, NLP, and really broader enterprise analytics. Well, thank you so much for that. It's always fun to throw a curveball at our guests and see what they come up with on the spot. So thanks for being a good sport. And thanks for great discussion.

You've brought up some really interesting ideas about bringing AI to the executive suite. Where can people connect with you and follow your thoughts on enterprise AI and other topics? Yeah, I regularly blog for AtScale. You know, I'm on LinkedIn, Josh Epstein, and I tweet at Josh M. Epstein. How about you, Fred? What's going on lately with you?

Yeah, so my background is AI and I'm focusing more and more on data management. And it's more specifically, it's helping enterprises understand the data they have and organizing and managing their data in a scalable fashion. So you can imagine, you know, people with petabytes of data where they need a little bit of help. I'm on LinkedIn under Frederick Van Heren and on their Twitter as Frederick V. Heren. And as for me, you can find me here every week on Utilizing AI. You can also find me every week on Wednesdays

with the Gestalt IT Rundown, where we discuss the news of the week. So please do tune in for that as well. If you enjoyed this discussion, please do give us a rating on iTunes. It really does help. And please do share this show with your friends. This podcast is brought to you by gestaltit.com, your home for IT coverage from across the enterprise. For show notes and more episodes, go to utilizing-ai.com or you can find us on Twitter at utilizing underscore AI. Thanks for joining us and listening in and we'll see you next time.