This Week in Applied AI with @ElectrifAi

Hey everyone, welcome in. We are here this week in AI episode 5, lucky number 5. Okay, I mean, you make— we make our own luck. Um, but thanks so much for joining us everyone. I am extremely excited to see you here. It is Thursday once again. Wow, what have we done this week? We've had some really fun, fun stuff going on here at Electrify. I've also had the pleasure of doing a couple of really great engagements in the community. I talked to these incredible women who launched a new podcast called the AI Literacy Podcast. I highly encourage you go check it out. I also today, for those of you, thanks for joining. Who joined me. I was live with the US Consulate in Kraków, Poland, and it was amazing. We had a bunch of people joining and learning all about the future of AI from my humble experience. But it was a really fun conversation about what's possible, and that's what I'm here to talk to you guys about. So if you are here catching us live, please go ahead and say hello. The chat is open. You can ask questions. You can say hi, you can do, you know, you could just give me warm, you know, support as I do this every week. But what we're gonna actually talk about something pretty fun today. This Week in Applied AI is all about things that I uncover as I'm on this journey to helping more customers implement practical machine learning, right? So one of the things that I focus on is how do we listen to a customer, right? Listen to you and the problems you have and marry them to an algorithm that can help you solve that problem faster, stronger, better than you ever thought before. And so I'll tell you a little bit about how we do that here. And then I'm gonna talk to you about this use case that I uncovered that I was like, we have to talk about this. So, so great to see you. Thanks for showing up. If you are here live, you could go ahead and type in #live so I know that you're here or just say hi. And if you're on the replay, feel free to #replay so I can come back and say hello to you as well. Thanks for all of you joining. Us on LinkedIn, and for those of you catching the replay on YouTube, you're all awesome. So thank you so much for spending a little bit of time with me. My goal every week is not to go much over 20 minutes. Sometimes I get excited and we do, but that's my goal, is to do short, sweet snippets about my journey in applied AI and what that might mean to you as far as, you know, use cases in the real world of how machine learning is being used by companies just like yours. So All right, here we go. Let's do this. So I'm going to flip over and go into sharing my screen and just walk you through kind of the standard journey, not spend more than a couple of minutes here doing that. Hopefully I'm in the right place. There we are. Now we're in Inception. You can see the world. But this is what I want to talk to you about. Every week we are talking about how you can travel along the machine learning journey. Most companies around the world know that artificial intelligence is coming. They know that it's happening. They know it's even possible, but not everyone knows how to do it or how to use it or how to even get started in a meaningful way. So what we do at Electrify is we have a bunch of machine learning models, an inventory of them that we have built for years, decades, right? Something you'll hear me say often is there's no compression algorithm for experience. We have built these models train these models, they work already. And now what we get to do is create an instance of that model and all the lessons learned and all the IP that we've built and share it with you as a service, machine learning as a service. It's something that will literally change the face of businesses. It's very similar to that moment, you know, you can imagine when Kodak and Blockbuster and those, you know, those companies that may not be with us anymore, where they had this moment of, Do we do this thing that kind of seems new and maybe I'm not really sure? And the decision they made, right, set the, the pace for their company forevermore. And so I wanna help you get on the track and the trajectory of a company that will be here 10 years from now. And those implementing AI are part of that journey. I also though say with all this great power and great capability, with all this power comes You know what's coming, great responsibility. So we wanna make sure you understand not just the cool, you know, goodness and capability that you get, but also the responsibility you have in collecting data for these models, in using data for these models. What questions to ask, right? How do you build an AI manifesto for your company to make sure you set a good, positive, and consistent intention for how AI will help your company grow? So I'm here to help you with all of those things and just so excited to see you all again. But let's get to the fun stuff. So I always go through kind of a, a use case or two in this, and I'm gonna demonstrate one for you today that I was just floored by. Me and my team of solutions architects, we were just talking through it and we're like, wow, that's an interesting use case. But what we often do is we wanna come to you. Let's say you tell us, hey, we've got this use case, churn. I wanna figure out why are my customers churning? We happen to have a model for that. How does that work? Right? We sit down, you tell us the data you have, we map that to a common data model, which is the data we need in order to execute our model. We give you a model report that shows you transparently, right? Interpretive. We give you the ability to interpret that model by understanding how it made its decision. And then you can decide how valuable is it to know which customers are churning when so that you can take action against that. Right, we wanna literally marry the business problem right up to, right, the algorithm that can give you the insights that you need to make all this data you're collecting worthwhile. We do these first initial product projects, we call them machine learning accelerators. We do them in about a month. So within a month you can get a really good idea about how these models will work and how they'll work for you. It does of course require some data, so we have to work together on that, but it's a really quick way to see How much AI, you know, what do we need to do to get AI started in this company? And this is one way to do it. So we'll talk more about that. But I really want to make sure we get to our fun example. So for those of you who are not familiar with machine learning as a service or leveraging an existing machine learning library, there are— we call it prebuilt AI, right? This is AI that companies like Microsoft, Google, Electrify, we've already invested in a huge organization, 250 data scientists, right, come together to actually build our machine learning models and then we create them in a way that we can reuse that IP to solve customer problems. So we have several different ways to do that. One is through what's called a pre-trained model. For those of you who are not familiar with that, these are the models, I'm actually about to demonstrate one for you that are trained commonly on natural language, You know, translation, right? Things that are known datasets, computer vision are good examples of pre-trained models where the data is actually already trained in the model. What we specialize in, we do have a few pre-trained models, but what we specialize in is our ability to actually take an existing model and train it on your data. We call it transfer learning, right? We wanna custom train a model on your data. So that the results are very, very specific to you. It actually takes your data, adds it to our algorithm, and creates a very unique result specific to your business. And this is how we accelerate, right? This is how we grow. So of course there are cool situations. I'm talking to a few customers now that are doing things that no one's ever done before. And I am all over that. I love being part of those projects. So we always entertain the idea of working on something that's new and innovative. And so we never turn down an incredible opportunity to work on something that might change the world. So we often, or we sometimes work on models that are brand new. And then of course we get them added to our inventory so everyone can take advantage. So what I wanna talk to you today about is invoice fraud. I don't know if you knew this or have seen this, But let me share it with you. Go into inception. Have you seen this? Like, it was from last year, or from pre-pandemic, I guess, right as the pandemic was breaking out. But I found this article and I was just shocked to see it, right? A man pleads guilty to stealing $100 million from Google and Facebook by sending fake invoices. So let me just break that down for you all. What that means is that Someone is sending invoices, like fake invoices to a company, and that company is blindly paying them. Now, if it happens to Google and Facebook, likely it's happening to other people too. So what I wanted to do was demonstrate to you what you can do with Applied AI today that can help you handle and more importantly, avoid invoice fraud. Right? And there's so many different use cases for Applied AI. I just saw this one. I was like, oh my gosh, we can solve for this. So what we have to do is first think of as an organization, how are invoices coming into our company? Obviously if we're a very, very small company and we have a handful of invoices, maybe a single person can handle that. But we might want to improve the ingenuity of those people by actually taking the data from the invoice and making it more usable either in applications, in UI dashboards, who knows how we might leverage it. So I wanted to demonstrate what an applied AI model can do for you. And then what we at Electrify do As we, you know, sometimes there is an option for you to do it yourself. Obviously you can go and build the model from scratch. You could even build against a prebuilt model from Microsoft or Google. But in both of those scenarios, there's quite a bit of work and massaging that has to happen to get the result you want. And that's why you might leverage a company like Electrify, who does it as a service, right? Where you say, here's the business problem I want, the KPI I desire, and we do the work to generate the model results.. So there's a lot less of the minutia that often shows up in these AI projects that you have to worry about that then falls on your partner Electrify to do instead. So let's take a look. I just couldn't believe this, this article, right? So what I wanna do is first, Experian, many of you know this company, they, you know, right around this time, interesting timing, right? So a month before that announcement came out, Invoice fraud was one of their top blog posts. And what they end up mentioning is things you want to avoid, right? Here are things you need to be thinking about. Now, if you are a company that doesn't have a whole data science team to actually build a process for solving this, I just wanna give you a glimpse into how AI could be used. And one question I always get is like, you know, is AI going to solve this problem specifically? Entirely holistically? And the answer is no. Its job is to actually help those who are already dealing with this problem every day, right? Those individuals that deal with invoices day in and day out, many times they are looking with their eyes at invoices trying to detect errors, anomalies, mistakes, invoice— invoicers, is that the right word? Right. Trying to figure out like Did the person who sent it, do they exist within our systems? Like there's so much that they have to keep thinking about. So how can we leverage AI to simplify that? So I'm just gonna show you really quick a demonstration from our friends. You know, many of you know I am an MVP from in Microsoft in the area of artificial intelligence. So I always like to show them as our baseline. They have lots of cognitive services that allow us to build on top of existing pre-trained models. And then I'm gonna show you what we can do at Electrify, right? What, what, how can we, what might be your next step if you don't wanna build it yourself, if you don't want to like go in and actually create a web app and you know, figure out how to access an API keys and all that, we can give you another option. So let's take a look here. You'll see, first thing I'm gonna do is load up an invoice, right? So this is just an example and of course this would happen at scale, but I'm gonna load up an invoice. I have a few downloaded from the sample set that Microsoft provides for testing. For testing purposes. I feel like I put it on my desktop. So I will go in and just grab the test data, and there's one in here to test, and I could actually grab any of them, but some were used to train it, and I feel like that's kind of cheating. So I want to use the test data, meaning that this specific picture was not used in training the model. And now what I can do is I'm just going to run the analysis. Some key points here. First, I did have to actually create the service, which is a service in Microsoft. I had to grab the API key and of course add it here so that it knew, kind of visualize in your mind, I am, you know, building a business process and I'm taking a collection of invoices and I'm sending them in as a parameter, right? I'm passing them into this model. The model's then gonna analyze it. I'll go ahead and hit that run analysis. The model's gonna analyze this document and figure out all the little bits and pieces of it so that it now turns into information that I can use as a developer. So if you have a web developer, right, the benefit of this is I no long— this isn't— I don't need a data scientist to do something with this output, right? I need someone who's an engineer, a software engineer, a web developer, an app developer, a full-stack developer who can go in and take this information. So you can see it provides this highlight. Now all of a sudden I can see the field information that was pulled out of that document But what I love about it is it also gives me my percent of confidence. So that percent of confidence is what I use as a developer to decide, do I keep it or do I not? Even as a business owner or a stakeholder in this project, what is the threshold of confidence I need in order to be, um, sure that I'll use that information to help me make a decision, right? So these are all pretty high levels of confidence. Um, you can see the invoice total, not as high. I wonder why. Um, but Good things to know, right? All we're trying to do is improve the ingenuity of those who have to look at these invoices every single day and who are not using OCR to extract out this information. Because guess what? If I'm in a web application, I can now take this information— oh, I'll tell you a story really quick, okay? And then we'll be done. Um, but I'll tell you this story. I have 4 children, as many of you know, and I will go into healthcare providers dentists, doctors, and I will fill out a form. And I won't just fill out one, right? You all know how this is gonna go. I go in and they hand me a stack of 4 of the exact same piles of paper that I have to fill out. And so what I will say to them is, so is there a way I could maybe fill out one and you could just duplicate it, right? There's AI for that. And they will say, no, honey, we need you to fill out one for each, so please continue on. Your 20, 30, 40-minute journey into filling out all these forms. What are the odds that I'm going to get it exactly right on every single one of those forms? Every duplicate time I enter in some information, I have transposed my own address, I've transposed my own phone number, right? But what— wouldn't it be great if instead I could do it once? It could be fed into a system like this. This extracts the information even in handwriting that I've written in and puts it into a document. Like, this is the future. The future is now. If you are in, you know, any place that you can help me make that happen, like, let's do it. Because we could save, you know, moms and dads. We could save them time and energy and change the world. But that's just one anecdote of the many ways that we can use this technology. But really, to go back to the original, you know, story, like, invoice fraud is real. So now I can extract out, right? Bill to, I can extract who the bill is from. I can then take this data and start doing analysis. So if I go over and— oh goodness, I am— okay, good. Just want to make sure I was sharing my screen. That would be terrible. But I could go over and look at, you know, what are the ways to prevent invoice fraud? I could now, as a web service, call out and figure out, have I ever been invoiced by this company before? If not, validation. Make sure we're doing business with them. Make sure we're doing this on purpose and intentional. Compare receipts against the purchase order. Again, there's AI for that, right? Check payable addresses. Validate those addresses. There's even this really great service from Experian. You can actually just sign up for that service and make it real, right? So this, this technology is here to help with a common business problem that could be losing you money or worse, just the the opportunity, right, the potential cost of this type of fraud within your organization. And you can solve it by just adding these simple checks. Because if you— you could add those checks, of course, manually, but having a human actually go and do that, this is where we're like, oh, I think I've seen this before. Yeah, I'm pretty sure I have. It's fine, right? We do that. That's exactly, I'm sure, how this person was successful, is that in these large companies with massive amounts of invoices, we look at it, we're like, That sounds about right. We're tired, we're human, right? But using machines to detect this information, extract the information, and then present it to me in a way that I, as the person who needs that data, can see it and act on it— that's where the magic can happen. So, uh, I've been talking to you, uh, almost 20 minutes now, so I just wanted to go back and, um, I'll stop sharing my screen now. And there we go. And just really reiterate how interesting the world is now when you put on the lens of practical machine learning. When you realize that anything that might be difficult or extremely manual that an individual is doing, you can improve the ingenuity of that individual, let them do the work that they care about and that they're good at without having to actually use their eyes to scan through PDF documents for information. Right? How do we make their lives easier so they can focus on the problems we hired them to solve? So I was on the consulate today. I was asked a question, you know, like, aren't you worried about jobs? And what about, you know, people who have these jobs that you're like using AI to eliminate? And, and I just want to re— I want to mention that in every project that I've done where I've gone into a role of a, what you would maybe call a non-skilled worker, A, that person does have skills. And hey, Jerome, so good to see you all. Thank you for commenting. But yes, it's incredible. But there is these people that I help with this AI, they don't look at me and go, how dare you take 3 hours out of my day so that I have to look at PDF files. There's tears of joy. Like, they're like, you know how many more invoices I can process now? Do you know how many more customers I can talk to now? Right. Those are the things that people want to do. They care, care about their role. They want to build momentum in the company that they're a part of. And when they are bogged down by these types of processes, like, it's just, it's just not right. So let's help them. So a lot of these non-skilled workers, I always say they're extremely skilled. Like they want to do more. They're bogged down by things that they may or may not really wanna do anyway. So let AI help them. It's not about eliminating jobs, it's about improving the ingenuity of those who have them. And I have seen, like I said, sheer joy in those that we have implemented some of these for. So let me know what you think. Pretty cool, right? I, I love these types of projects. I just play around with these things all day long. I am the So excited to be part of Electrify and get a chance to do this every single day. So if you have any questions, drop them in the chat. Otherwise, feel free to head on over to our website electrify.net and send us an email. Uh, we, we are actually going to be starting to do a webinar series where you can join me once a week. We do a webinar and we are going to be doing some incredible projects together. So I'm so excited. It's been amazing. Happy Thursday! Enjoy your weekend. Take some time to relax, uh, this weekend And like I said, if you need anything, if there's anything I can help you with, please don't hesitate to reach out. Thanks for following. Thanks for watching. And we'll see you next week on This Week in Applied AI. Thanks so much, everyone. Bye.