Mike Vizard: Hello, and welcome to the latest video interview for Techstrong.ai. I’m your host, Mike Vizard. Today, we’re with Dan LeBlanc, who’s CEO for Daasity. And we’re talking about generative AI and where we are in the world. Dan, welcome to the show.
Dan LeBlanc: Thanks, Michael, really excited to be here.
Mike Vizard: Everybody’s running around talking about how general AI is going to turn our world upside down. But maybe you think perhaps the use cases are somewhat limited. And maybe it’s not quite all it’s cracked up to be? What’s going on in your mind?
Dan LeBlanc: Yeah, I’m definitely cautiously optimistic about generative AI, within kind of this digital commerce space where we help and work with a lot of consumer brands, there’s use cases that we have where generative AI is already making a huge impact on things like how do I write copy for my products? How do I go test product copy, product images? How do I do things from the marketing side? But there are areas where we’re just not ready as both brands and, and at Daasity to go implement them. So things like customer service, or even decision making. So it’s still very much a human interaction with the generative AI to help us go in and tweak and accelerate sort of the ability for us as individuals to be more productive. But we haven’t turned around and sort of unleashed it and let it kind of run by itself and take over our jobs or decisions for it to really execute by itself.
Mike Vizard: Is that a function of the data that we’re using to train the models? I mean, if I look at something like ChatGPT, it is a general purpose model. And so maybe it doesn’t lend itself to everything. And of course, general purpose also means that it’s subject to error. Do we need more domain specific models that are going to be driving a particular tasks such as an E commerce transaction, or whatever it might be?
Dan LeBlanc: Absolutely, I think it’s both a function of having a more specific model to a certain use case. But then I also think it’s a little bit of just the control that we’re going to want to have as individuals. So if I kind of think about, you know, the advent of, say, QuickBooks – great platform that’s allowed a lot of people to do a lot more sort of bookkeeping; the need for bookkeepers has decreased, the need for financial analysts has actually increased. And so if we think about that, that’s a very specific software. And so we can think about generative AI in that same example. So if I want a tool that’s going to do a really good job for automating, say, generation of copy for my email, I may be think of it as I need a model that’s going to be specifically for how do I run that model around that specific piece, maybe even specifically within a certain vertical. So beauty is probably very different from automotive. So I might want a model that’s very specific around those kinds of industries. And I would need that model to be very different from say something that was going to go and interact with customers around product problems, or helping them direct them into, you know – to say I’m at IKEA and I brought it home. And wouldn’t it be great to have generative AI that kind of guides me through and helps me – actually interacts with me – that is able to help me build my new bookshelf or whatever it may be? And those are very different use cases where the models have to be trained around that specific data to help, and having something that’s very generic is going to prove to be very challenging.
Mike Vizard: Do you think this will force us to maybe clean up our data? Because historically, at least as far as I know, most organizations aren’t very good at managing their data. And so if I don’t have good management practices for the data, how am I going to build an AI model that works?
Dan LeBlanc: That is actually probably the biggest challenge that this generative AI is going to have. Because you’re exactly right, how many of us go and sort of catalog everything that we have, you know, if I sort of thought about the simple analogy of it – I’m gonna go apply generative AI to choose whether I should have to use a fork, knife or spoon. And if I opened up my, you know, the drawer where I keep my knives, forks and spoons, I probably haven’t labeled any of them. And I might have a teaspoon and a soup spoon, and I probably haven’t labeled those. So it doesn’t understand that data. And that’s definitely a challenge as you start thinking about larger and larger organizations. And really, that AI needs to have an output that we want it to drive a result; how does it understand what the value is, or what is the real sort of driver, what that result that we want it to generate is? And that comes back to, as you said, labeling data. So we don’t do a very good job within most of our organizations within the databases, within the data structures of providing sort of that instruction around these data elements, feeding these other data elements. And this is the data element that really matters – like revenue.
Mike Vizard: Speaking of things that really matter, it seems like a lot of the data science projects themselves – once they get in the real world – they kind of fail; somebody will create a model that will determines that, you know, revenue drops every seventh day, and they’ll hand that off to the business and the business will scratch his head and say, “Well, of course, the seventh thing is Sunday.” So we kind of get models that are actually meaningful to the business.
Dan LeBlanc: Yes. And there’s two parts about the models. I think that’s a great question. So the first part is, a lot of times when we’re when we’re kind of thinking about those models, more time is spent actually building the pipelines – how do I get the data in? And that goes back to what you were saying about – well we don’t catalog it, we don’t sort of standardize it, we don’t do a lot of things around making it very clean. And then the second piece is really, you know, often, what is the business problem we’re trying to solve? And how does that person interact with that data scientist? So you think about kind of the goal of a data scientist is I want to apply these very fancy statistical methods to have a certain outcome. And the business is not necessarily asking the question that they want or not being detailed enough. So it’s tell me how to predict revenue. Without the outlier in your example, tell me how to predict revenue without the outlier, with the information around, I know, I don’t want you to do it by day of week because I know Sunday’s are going to be worse; I need something else. And I need you to tie it to this type of information. And that’s usually the challenge that we have is – the business comes back with a very generic problem. I don’t understand how to improve my traffic. That’s kind of vague and opaque. And what we need to do as data scientists is really transform that into something that’s a little bit more tangible, that’s based on the data. We can say, I’m going to do X based upon why will this work? Let’s go build that.
Mike Vizard: Is there a fundamental cultural divide between the data science teams and the business folks? And how do we kind of address that? Because it seems like a lot of times these people are hired and shoved in a room somewhere, but they don’t really know much about the business.
Dan LeBlanc: Exactly. And it’s that cultural, I hired somebody who’s really good at math, science, statistics. And we don’t involve them in the business problems. We don’t involve them in the business. We just kind of think of them as, oh, you’re the person in the corner, whose job is to go just build models for me, and we don’t provide enough of that information. So if we can really bring them into where we’re discussing strategy problems – how is the business performing and include them in those conversations, they’re going to get more insight into what we as a business – think driving; what are those drivers that actually result in the outcome that we want, which is generally either higher revenue or higher profit?
Mike Vizard: What is the confidence level that they have about AI because, as one wise guy said, it’s one thing to be wrong, it’s quite another thing to be wrong at scale. So if you’re a business executive, are you a little bit worried that, you know, there’s going to be 50,000 gallons of yogurt spilling on a forest somewhere because some AI model said we shouldn’t produce it?
Dan LeBlanc: Yeah, that’s the giant fear. And so it’s a bit of a challenge right now, because AI is interesting. Not – valued is probably the wrong word – maybe maybe certified is a better word at enterprise. And yet, it’s also already unleashed. And so as a business executive, it’s actually kind of a scary technology, because the likelihood is, if you were to go into your organization and say, “Hey, how many of you are using ChatGPT for something?” I bet, you’re going to have a lot of people raise their hand, because your marketers are already probably using it without you knowing to kind of help them accelerate writing product copy, or maybe playing with images or with video, because those use cases have already come and been found to actually be very useful in terms of accelerating the ability for these employees to go generate outcomes, which is more versions of copy, better copy. The downside is you have no control. It’s, you know, like a million people, it’s like a million people using – they’re bringing their own laptop to work with all their own software and everything. And that poses risk for you as an executive, just from an IT security, etc. I heard from a colleague around a lawyer that was using ChatGPT to help write legal documents. Well, that’s kind of scary, because now, some company’s legal documents may actually be part of that library of knowledge. And so that becomes really, really scary from that perspective. And so, as executives, we have to do two things. One is start thinking about how we put this into our organization in a thoughtful way, and how we control it and maybe start to look at sort of enterprise technology, for use services like dialing? Would you be more comfortable with your Salesforce or your Microsoft Outlook or whatever, or your HubSpot or whatever platform you have had built in? Or would you rather work with, you know, just some, let all your users have some app that’s unrestricted by themselves. And so that’s going to be a big focus for a lot of executives. And as that starts to take frame, and it starts to show positive results, then they’ll be more willing to go into these areas of allowing it to make more decisions by itself.
Mike Vizard: So what’s your best advice to folks? I mean, should they just be experimenting right now? Or do I hire a bunch of developers and data scientists and turn them loose and tell them to go build some models? And we’ll see what happens, or that’s an expensive way to do experimentation I’m sure. So, you know, how should people proceed?
Dan LeBlanc: Yeah. So I think you should go find tools that you can leverage that already have it built in. And so for most companies, unless you’re a tech company, or less, this is going to be core to your business going and developing your own generative AI is going to be an incredibly expensive endeavor, and it just doesn’t make sense for you to go invest that kind of capital, because what’s already going to happen is, companies are going to go do this where they’re going to sell it as a service. And so if you think about, you know, within the e-commerce world, I may be using platforms like Adobe to do my photo. Well do I really want to go and build editing that uses AI to help improve my photos? No, I should just let Adobe go and then just go use the better version of Adobe to go do that. Similarly, if I’m a sales organization, you know, if I use Salesforce and Salesforce owns Tableau, and Tableau has already started to put generative AI into their analytics to basically recommend other reports, then that is what you should go look at if you’re analyzing your sales funnel. And so we should not invest ourselves. But look at how these enterprise applications are starting to think about AI there and see whether that’s the fit for us, because it’s not about how AI is gonna get rid us of all of our jobs. But it’s about how is it going to make us more productive and allow us to have a greater sort of output from what our main task is. And that’s really what it’s going to do for business.
Mike Vizard: Alright, folks, you heard it here. No matter how advanced the technology is, it still comes down to a bill versus buying kind of conversation. Hey, Dan, thanks for being on the show.
Dan LeBlanc: Great, Michael. Thanks very much. Appreciate it.
Mike Vizard: And thank you all for watching the latest episode of techstrong.ai. We have this episode and all our others on the website. We invite you to check them out. Until then, we’ll see you all next time.