Watch the Webinar: Taking the Leap into AI

In this webinar, Redapt experts Amie Mason and Bryan Gilcrease discuss how your organization can unlock its data capital and adopt AI technology.

Video transcription:

Thanks for joining our webinar. Today we're going to be speaking about taking the leap into AI and unlocking data capital to solve the top challenges faced by today's executives. Why you're here: you want to learn about the current economic crisis or how the current economic crisis has impacted your need for business agility. We've got a number of customers that are just reacting to what's going on in the world and looking for new ways to service their customers and take care of their business.

You want to unlock the enterprise data to solve those problems, and then gain alignment around the right use cases. We're going to share some use cases today and actual the work we've done for our customers. And then learn how to do it yourself, like smartly and what to do and what not to do. Today presenting is going to be Bryan Gilcrease. He's our Senior Solutions Architect, specializing mainly in AI workloads for our customers. And Amie Mason, who is a Practice Lead for the Data Science and Analytics team. Amie really dives deep into what our customers’ data sets and helping them unlock new ways to use that data and has a lot of perspective across industries like healthcare manufacturing. You name it, Amie's done at retail. So anyway, I'll turn it over to Amie and Bryan.

Sure, thanks everyone. I'm Amie Mason. So Bryan and I will switch out a bit as we go through today's presentation. The first part of which I have talked to you guys a bit about what we're seeing with our clients in the wide world of AI, if you will. And then Bryan will talk about data capital and some things that you can do there. And then I will come back and talk about some of the introductory steps that you can take to get started with AI. It is a big wide array of things that you can do. And hopefully, some of the ideas that we talked about today can make you feel more comfortable about making that step yourself. So let's go ahead and move forward. Thank you.

Before we actually get into talking about some of the issues that you see listed here, I want to take a small step back and say that AI in and of itself can mean a lot of things. In its most basic form, it can just mean automating some human based processes. All the way through to what you're hearing in machine learning and deep learning, neural networks and things like that. So keep that in mind as we go through today. Some of these items that you're seeing here are going to fall more on that more complex scale. And then towards the end of our presentation we'll talk about, again, some of the introductory things that we can do.

Let's go ahead and talk about fraud first. I know the slide says fraud prevention, but I'm going to go ahead and if we can go to the next slide, please. I feel much more comfortable calling this fraud detection rather than fraud prevention for any of you that might be in the finance space. There's a lot of rules and regulations around, along actually calling something fraud detection. So when we work with our clients, we definitely want to call it fraud detection or even just anomaly detection. And so most commonly this is thought of in the finance and banking space. There's a lot of, fraud happens there in a lot of concern on behalf of the executives in those industries.

Some of the things that we've done with our customers, on the finance and also in the audit space is automating what are normally Excel based, or I'll say it again, human based processes. We've worked with some of the major auditing companies to take processes where their auditors were going through thousands and thousands of records, and maybe using a if then based or rules based process to identify records that they needed to take a deeper look into. So this was taking a lot of time. And so to reduce that time, we're able to use AI or rather machine learning to change those rules based processes into something a little smarter using machine learning. And so using anomaly detection or outlier detection, we're able to cut down on the time that takes and flagged those records so that way people aren't spinning their wheels trying to identify them.

Taking something to a true fraud prevention step, there's a bit more that goes into it. But fraud detection in and of itself is not a huge problem to solve. Another common one is churn and attrition. And so when we think about that, if you think about a lot of the turnover that's happening right now for various reasons or on the employee side, also on the customer side. And then when we look at some of the data and some of those patterns behind why employees leave or what makes a customer leave one organization and become a customer of another organization, there's a lot of data behind that. And so being able to tap into that, we're able to build out attrition or churn models, to identify who's likely to leave and even when they're likely to leave.

Building out things like a customer lifetime value model can feed into those forecasts and give you a little bit more insight. And then if we pivot this as well, the same types of approaches can be very helpful for lead targeting. So we have a lot of clients that are in various types of marketing spaces. Right from insurance to property management, imagine if you've ever been a renter and you go and you put your information into a search engine. How are people on the other end of that, deciding who's actually likely to be someone to fill out an application and rent from us or on the insurance side who's actually likely to sign up with our company.

Being able to take a lot of demographic or market data that's out there publicly, combine that with historical data that maybe we were using for some various processes to identify this. We're able to take that and just build out predictive models to help identify it. And really we're all looking for being a little bit more optimized with the data that we have, we're storing it for a reason. So a lot that we can do in that space as well. When we look at resource planning, again, these categories are a little bit more broad than maybe what's on the page. But we can look at things like inventory optimization, demand, forecasting, staffing, allocation, any of that.

It's really all the same suite of approaches to solve these types of problems. So when we're working with our clients, we're able to reuse a lot of the same model, just same general approaches to solve these types of problems. What's the next one? Is it outages? So outages. This one, if you'll work with it here, there are a lot of things out there that I'll say third parties are doing with predicting more things that have been happening with COVID. Or if you think about any of the natural disasters. So I actually don't venture into that space. I'll leave that aside. But to get you thinking about some of the things that we're actually doing in this space, it all comes back to that outlier or anomaly detection.

So the reality of what we're doing with some of our clients is using some of the same, again, same standard approaches for more of a predictive maintenance scenario. And so we've done that in the manufacturing space, we've done that in the communications space. Again, we want to address a lot of these issues that might sound familiar to you, but also give you the understanding that these aren't far fetched issues. They're all solved by the same general approaches, just using the data that's specific to your scenario. And so the final item that I'd like to talk about before we get into some more detail is in general, most companies are just wanting to keep up with the times and make sure that we are providing the solutions to our customers that are needed. And make sure we're moving in the right direction.

I mentioned this earlier. This is where really bringing in third party data can help a lot. So whether we're looking at demographic data, whether we're looking at weather data, whether we're looking at third party market competitive data. So information for NAICS, so there's a lot of different data sources that you can pull in here. But to do things like competitive analysis or maybe more specifically something like product recommendation, or market basket analysis to observe your existing customers or bring in new customers or create new opportunities to serve your customers.
I think that they're all readily solved with some of the technology that's available in the AI space. And so, again, rob high level overview of some of the types of solutions that we're seeing our clients work with, we can dive deeper into them later if you want to get into them. But for right now, let's go ahead and have Bryan talk about some of the technology and then we'll come back and talk about some of the baby steps into AI.

Yeah. Thanks, Amie. That was a great overview. I think a lot of these are real issues that I've heard from customers in the past. And summing it up with that they're all addressable I think is great. And I think we really are in a time where we have answers to a lot more questions than ever before. And using advanced techniques like machine learning or deep learning to solve some of these problems very quickly and even in real time. So it's a great topic and I love talking with customers about it. I just wanted to talk a little bit today about data capital. So what is data capital? There's always a new buzzword, big data. Data is the new oil. There's always something that marketing is putting out there and that you're hearing from a lot of vendors. And sometimes it's just a neat way to get interest behind something that really is important. And I think, that's mostly what we're trying to do here.

The fact is that although we're not going to go and create more oil, we can create data. But it just, it takes a long time to have valuable data. So I mean, if you look at Redapt, we've been around over two decades. We've seen the dot-com bust, we've seen the resurgence with Web 2.0. Customers moving to Cloud technologies, DevOps, containers, Kubernetes. All of these things are, it's been in our lifetime of our business. And then if you think about that, that's a very unique perspective. And if we had data for what all of our customers were doing from 25 years ago to now, that would be a very powerful data set. And unfortunately, it's only been the last several years that we've moved to a data driven company. But that historical data could provide insight that we may not be able to recreate in the future.
And if you look at that, that's why Amie was talking about maybe, talking to the outside firms to get their data sets. And you have to look at what's unique to your business. So what data do you have? If you're in manufacturing maybe you have a lot of machine data, a lot of sensor data. If you're in retail, maybe you have a lot of customer data that's specific to your industry into exactly what you're doing. And sometimes you have to augment that with outside data sources, but the reality is it's never going to be as good as what's specific to exactly what you're trying to do. Hopefully, you're working in a business that has some unique value. And a lot of times that can come from some data set.

So why is it important? The next step of having that data is being able to make decisions from that data. So being able to use that to solve some of these problems that Amie was talking about. Maybe it's fraud detection, maybe if your business sees a unique type of fraud, or you're able to identify any new type of fraud with your past historical business that you've been doing. And then maybe you can take that and you can say, "Hey, look we found this special case, we've got this algorithm using our data set that can now detect this data and then now you can maybe sell that to your competitors. And make a business out of that." So there's lots of uses for all of this data capital.

And I think, that's why we're having this conversation right now. It's because in the past it was very difficult to solve those problems. So one of the problems in the past was also storing data. It's traditionally been fairly expensive to store data or hoard data that that may or may not be useful. And that's no longer the case, we have a lot of very cheap ways to store data. Whether that's in a an object store or a big data warehouse or something like that. There's a lot of different ways that we can make that affordable. And I think now we're talking about where is that data most valuable and making sure that the compute resources and the storage has the flexibility you need for your business.

So we start talking a lot about Hybrid cloud solutions where you're able to store your data where you need it. Maybe that's on-premises for a some regulation or security requirements. And then but maybe, you need to do some development or testing and you want to spin up resources. And in bridging the gap between on-premise and public cloud. So it's all about where does it make sense to store the data. A lot of times, we're working with customers that are doing some image classification or something like that. And they have a lot of image data. They're storing that on-premise so that they can use high speed GPU servers, all-flash Isilon storage, stuff like that to get the most value out of that data.

To put together a data strategy, there's several pieces like Amie mentioned earlier, some of it is very complex. But to get started, the first thing you have to do is you have to realize that your business has some unique data. That the value prop that you're offering and the things that you're trying to accomplish are unique in some way. And that the data to support that has value. So one of the ways that you start with that is being able to get that in the hands of teams who can start finding value with that. And they can start augmenting that with external data sources maybe or generating reports to get that to be visible to executives. So you need to find what value that has and then start focusing on that. And then you can start creating all of these different technology based solutions.

So then you start looking at data warehousing, maybe data lakes, maybe it's just traditional relational data and you're building out a SQL Server or something like that. And you start looking at how that's going to enable you to pull data from all of your different sources. And get it into the hands of those people who can actually find the value and create that value for your business. So working with Dell EMC, there's a lot of hardware options and ways to accelerate your machine learning or AI initiatives. One of the things that you can start with is you can look at what type of data you have, is it relational data? Is it structured? Is it unstructured? And then you start to figure out what solutions you need. Like I mentioned earlier, maybe it's a bunch of image data that you're going to need to access very quickly with multiple GPUs. So you can use something like an all-flash Isilon, maybe you have a lot of unstructured data and you need to build out a data lake using something like a dupe.

There's a lot of great reference architectures from Dell and from REAP that can help put together some of the infrastructure type questions. And that helps with a lot of the things mentioned here as far as scalability, cost governance and security. Those are built into those reference architectures and they help make those decisions a lot easier for you. And then I'll pass it back over to Amie, and she'll help get started with AI.

Great, thanks. So I mentioned earlier, that there's this wide range of what is AI and if you can vision with me an egg with three layers inside of it. That first layer is all that AI is. From basic automation through deep learning. And then there's machine learning inside of that. So machine learning is a subset of AI. And that's where you get into the math and the statistics and all of that. And then within machine learning is deep learning. And that's where you get into the multi-layer neural networks and large, large, large amounts of data. And some of that can be scary for a lot of our customers or just a lot of people that I talk to at different events.

What I want to talk about in this section is some of the things that are easy first steps into AI. In using AI within your organization and getting that buy-in. And we'll come back to the role of buy-in, in a minute. But let's first talk about these. And I know that that Bryan mentioned a lot of what's, what infrastructure is available, on-prem. But I want to bring up some of the pre packaged offerings that might be available in with some of the cloud providers, like Microsoft, Google or AWS. They have prepackaged offerings or APIs that solve all of the items that we're going to be talking about here. So from automation to internal and even external communication, initial level data analysis and chat bots for customer service.

Let's go ahead and spend a bit on each of those items and talk through what they might look like. So with automation, we're looking at some logical workflows for anything from ... and I think there's some step through slides on the next one. Anything from building a form and workflow that's automated for any type of data entry process. So we've worked with, most recently a local organization here where I am in Arizona to automate a data collection process and reporting process for probation officers. So they can come in there, they're working on the Microsoft platform with it. So for those of you that are familiar with power platform, they're using Power Apps to collect the hearing data for probationers, and then Power BI to automate the creation of that hearing report that gets sent out to the courts.

So anything like that, I'm sure you can think of use cases within your organization where you're collecting data and need to automate the entry for data quality issues. And then also make it very easy to get to the consumption of that data. For natural language querying. So I would say that's where we're getting at when we say basic natural processing. There's also a lot of APIs around that for translation, keyword analysis. So if you have a lot of customer service surveys or anything like that, you get to the crux of what the comments are very quickly as well as there's APIs for sentiment analysis, so that way you can address those that are likely to be most critical.

On the internal communication side, we're thinking about things like automating internal search. There are some great search APIs. So if you think about some of the things that we've done at Redapt, we have a large amount of information that's available to our employees in an internal reference system. And so in the past, things have been all over the place and you have to go through a lot of different layers to get to what you're looking for. So if I'm looking for information on an update policy, I have to go to policies and then I have to go to vacation if I'm looking for my vacation policy. And then I have to search through a list of a list of what might be in there. And so if you're able to implement something like global search, I can just type in what I'm looking for and get taken right to a series of documents that might be related.

And some of you might think that, oh, that's straightforward. We see that all the time, but within an organization you might not have it internal to what you've done. I think sometimes when I'm talking to clients about what's possible there, they all think that Google just happened overnight or Bing just happened overnight. And implementing something yourself previously was a lot more difficult, but through some of these pre-built offerings, it is as easy as you might have thought. So data analysis is another that is pretty easy to get started with. If you think about things like some of the automated machine learning. So there's packages available and are, if you're looking for open source. I know Microsoft has an AutoML solution where you can just bring your data to the table and run this package or run this AutoML process against your data.

And it will iterate over a lot of available algorithms. And it's a great start for identifying if the data that you have is going to produce any reasonable level of accuracy. For a wide array of approaches. And when I'm when I'm talking about approaches here, I'm talking about things like regression or classification or anomaly detection or clustering. So the different categories of machine learning. So you can very quickly get to that decision of is this something I can move forward with. And then the last one is customer service. So this can be internal facing or external facing. But building these tools for communication that automate some of those customer experiences through chat bots.

So imagine when you go to a website and I'm sure you've seen it, where there's that pop up. "Hi, do you have any questions or Hi, welcome to our site, let me know how I can help or anything like that." A lot of times that's automated and sometimes it's very obvious that's automated. Sometimes there's some things that you can do to customize it to where it's a more realistic experience for, again, either internal employees or for customers who are coming to your website. And then it's also an option you can build through the logic to where, from a fully automated perspective your customers can quickly get the answers to their questions. And then there's also options for, okay, what you're asking me is not part of what this chat bots has been built to do, let's go ahead and transfer you off to someone who can actually help, so an actual person.

But the idea here is, through some of these, again, pre built or customizable solutions you can reduce the amount of time that your critical human resources are spending on tasks that can't be automated. So what should we be thinking about for making these first steps into AI within an organization. The first thing that you have to do is do an initial analysis of how AI can help your organization. So I want to understand what problems I have, what problems I think can be solved with AI and machine learning and prioritize those items. And then the most important thing is getting executive level buy-in. If the people who are in charge of making the decisions within the organization are bought into using AI, let alone spending the money on AI, which we'll talk about in a second. You're not going to get very far.

So producing that business case showing the likely business value and getting executive buy-in is huge for preparing your organization for moving forward. We all want to be on the same page. We all want to appreciate the value that AI can bring. And we all want to have an understanding of where we're wanting to go. And so that brings me to the money item. So we want to be prepared to make that investment in machine learning and AI, both from a ... we want to understand what the cost is going to be. And so having support be it from a partner or a partner organizations such as Microsoft, Google, AWS or Dell to help you cost that out. Get an estimate together. From a time and costing perspective.

We want to take a realistic look at where we are as an organization from a technical maturity perspective. Do we have our data in a central location like a data lake? Do we have data that is currently siloed throughout the organization that it would be difficult to bring together. Are there ownership or control issues with respect to data or logical processes related to the data such as calculations or anything like that. That we need to break down barriers for before we can move forward. So we want to have done some form of an analysis of where we are and what our actual next steps need to be. A lot of organizations in the past, so when I started with Redapt about five years ago. It was when some of our cloud partners were really starting to put their money into building out these AI and machine learning services.

And so a lot of clients are like, "Great, Microsoft, you have this new tool we want to use it." Or with any of the other cloud partners. How do we do that? So they'd come in, they'd have data in Excel spreadsheets or maybe in a SQL database, and we do a PLC. And they'd see varying levels of success with that PLC. And realize, hey, we need to take a step back. And we need to really catalog our data, get it into a data lake or some form of a central repository, or go out and get these third party data sources that we've talked about today. So that way we can build a more robust model because we're not quite seeing the level of accuracy that we would like. Or maybe we are seeing a level of accuracy that we would like but we know that there are these other factors that we need to consider. And we need to bring in this data to actually have a viable solution, not just something that's statistically accurate.

So we really need to take that internal analysis and make sure that we're in the place that we need to be in before we can move forward. So a lot of time, we will do what we call a data estate assessment focused on data science, and produce findings and recommendations document that lays out these are the scenarios that we want to approach. Here's where we are from a maturity model perspective. And these are the ones that we should tackle first in order to really be successful in our implementation. So we want to make sure that first step was organizational alignment and then we need to analyze and prepare for our investment. And then the third thing is we really want to make sure that the entire organization feels empowered to use and consume AI. So we might have built out a model whether it's a PLC or something that we're ready to move into production.

Sometimes we have to coach the rest of the organization into believing in what we've built. AI can be a little bit scary. So we want to show the value to the organization be it through reporting or some other process where we can really show them the value. And then we also want to talk to them about what their pain points are. Often AI happens in a vacuum within the BI or IT organization, or a lot of times it gets started with finance. And we want to reach out to the different departments or different parts of the business and see what problems do you have that we can solve with the data that we have. How can we help you? And then that really gets beyond the executive buy-in that gets the organizational buy-in and fills out that triangle of being able to move your business forward and really use and see value in the investments that you're going to be making into AI.

And so a lot of these decisions that we've talked about here can be daunting. Especially doing that analysis of truly taking that look at where are we as an organization, and with all of these options that are out here for using AI, where do we start? And so I definitely recommend finding the right partner. And as Redapt, we do this with a lot of customers. So we'd like to be that partner for you. If we can go to the next slide. But we want to make sure that you're working with a partner that can work with you to understand your needs, your goals, identify whether you're working in an on-prem infrastructure or in the cloud. What the best implementation platform is for you. And help walk you through and be comfortable with all of the different stages of the AI lifecycle. And yeah, that's what we're here for. That's why we're having this conversation today. So I'm happy to help with whatever questions you have. And then I know that Bryan has a few comments to make in closing before we do that.

Yeah. Thanks for those who are sticking around and listening. Hopefully, these are some of the things that you guys are thinking about. And hopefully, there's a few insights that you're able to take away. The big picture is that having good quality data and having good quality analysis is going to help drive better performance. It seems like magic, hey, we can write all that out on slide but it doesn't make it true. But this is what Amie and I have been seeing with our customers. And this is, it's though the modern paradigm for data usage. And it's very powerful. And we have some great tools now to expand it and to make it address the problems that we're focused on. I think we've got a few minutes left for Q&A, so please stick around and ask questions. Hopefully, this was useful. Thanks.

Awesome. Thanks, Bryan. This is David again. And just to wrap up, I did get a couple of questions sent over to me via chat. Please, if you have any other ones please send them in. But one of them, Amie I think this might be for you. It's with AI improving and getting more ... I guess, like it's continuously improving but also getting more complex. What solutions are you seeing that are enterprise ready that maybe didn't exist or were very experimental a year ago?
I would say that there's been a lot of progress with organizations, particularly in the cloud space building out these pre-built APIs. So if we think about Microsoft and the cognitive services or the chat bots frameworks that they have available. You can literally go in, go into Azure purchase the skews for these services and you have an API that you can just run your data against. So that eliminates so much of the time that we were previously spending building custom models to solve these problems. So I think that the investments there have been, have probably been the biggest change.

Yeah, I'd like to add a little bit to that as well. Those are great resources. And I 100% agree with Amie. There's also been a lot of work on MLOps side. And coming from like a IT organization and trying to support data scientists. Tools like Kubeflow and Kubernetes have helped take that and make it possible for an IT organization to stand up cloud like resources and support multiple different business groups. And I think from the IT organization, that's one of the biggest areas I've seen.

I agree with that. And Dave, I know you're familiar with the MLOps offering that we have on the advanced analytics practice to implement some of that for clients. Because it is new and there's a lot to it.

Yeah. And that leads into my next or the next question that we have. And it's awesome that that Redapt can help with this. But if we outsource all of this expertise, how do we also learn it so we can operate it in ... and adopt internally.
Right. So most of the time when at least on the professional services side, when we are working with our clients. There's always this aspect of informal knowledge transfer. We do offer like formal training as well on the things that we're talking about. But our developers and our architects like to work hand in hand with our clients to build out, whether it's a data warehouse architecture or if we're doing something in the data science space, build it out, and hand it off. So that way our clients can move forward with it. We've helped several organizations build out a framework for rapid PLC development. And a lab format using some of the different Cloud ML services and MLOps. So that's generally our goal, is to empower our clients to use the things that we've built and move forward with it. While we'd love to be involved in perpetuity, it's not normally the case.

Okay. Yeah. Cool. We have two more questions. I think we've got just enough time to get to them. So with work from home being our new normal, what trends are you seeing and how are organizations leveraging AI to help productivity?
That's an interesting question. So internal things that we've been working on are more in the automation space. But things that I've seen in the market in general are a lot of things from just a productivity perspective. So I'm not sure Bryan, if you have anything more to expand on that.

There's a couple more questions so we'll just move on. There's Redapt device to do these type of initiatives and a public cloud or private on-premise legality.

Yeah, I think, I tried to hit on this a little bit. But the important question is what does your business need? And oftentimes, Redapt our goal of our engagement is to help you decide where you need to run these workloads. And we work with a lot of customers doing these on-premises, building out their own private cloud infrastructure. We also work a lot with Azure, AWS, GCP. So it's really about focused on what works best.

You actually answered two questions with one answer. That's amazing.

I used IA to get the answer.

Good job. Okay. That wraps it up then, I think we've got to all the questions that attendees had asked. And that I think we'll wrap it up here. Thank you everyone for attending and there was literally no drop-off from start to finish. So amazing.

Great. Well, thank you guys so much. Feel free to reach out if you have any questions. Happy to help.
All right. Thanks, Amie. Thanks, Bryan.