Subscribe
Share
Share
Embed
Welcome to Creating Next, the groundbreaking podcast by TransImpact that delves into the dynamic realms of parcel spend management and supply chain planning. Join our hosts as they navigate the cutting edge of industry innovation, sharing insights and strategies from top experts. Subscribe now for an immersive journey into the strategies and technologies shaping the future of your business.
Bob - 00:00:16:
All right. Good morning, everybody. And I'm pleased this morning to be joined by Norm Pollock, Joe Vocaire, and John Hayes from TransImpact to discuss a really hot topic today around artificial intelligence and machine learning. It's really taken on a whole new life in the last 12 months, and these three gentlemen have been on the forefront of some of the interesting things that are going on in this space. AI and machine learning mean a lot of different things to a lot of different people. I think the topic is often confused, and I'm sure that Norm, Joe, and John are going to help us understand it a lot better this morning. So maybe just to kick things off, Norm, we could start with a basic sort of definition around what is machine learning and how has it so quickly become top of mind for innovation?
Norm - 00:01:11:
Yeah, good morning, Bob. Thank you. You know, machine learning is really a subset of artificial intelligence. It uses algorithms that are trained on data sets, and it then creates self-learning models that are capable of predicting outcomes and classifying information without human intervention. You know, we really weren't hearing much about AI until the last year or two. And, you know, it really came to the forefront over the past year due to tools such as ChatGPT making it into the news and now becoming widely available. And, you know, ChatGPT itself uses several forms of AI, including natural language processing, deep learning, and machine learning.
Bob - 00:01:55:
That's really interesting, Norm. And Joe, you and the team at TransImpact are involved with supply chain and logistics. Where do you see machine learning fitting in that space? And what are some of the opportunities for using machine learning to improve supply chains?
Joe - 00:02:13:
Well, machine learning is really, really valuable. And having the computer really research years and years of data. To help us look for patterns and anomalies. So what we're going to be doing is we're going to be using our forecasting methodologies, our demand planning, looking for other opportunities for actionable insights on information. So we want machine learning to study the data for us. It's going to do the same thing that human beings do. It's just going to be able to produce and go through a lot more data, a lot faster. And allow us to present that data back to the users and the businesses so they can make better business decisions.
Bob - 00:02:55:
John, when you think about your customers at TransImpact, how does machine learning and AI really become of consequence to them?
John - 00:03:05:
Well, it's a real competitive differentiator. In that, um, You know, if it creates... An extra 4% of accuracy or can account for some context that has been seen before but that might not be seen by a traditional statistical model, then it's an advantage. It may end up being making sure that you have the right quantity in stock or that you're able to assess your risk ahead of time. So it's just an advantage.
Norm - 00:03:41:
You know, I can add a little bit to that. You know, years ago, you know, I had discovered ClickView, which is a business intelligence tool, dating myself a little bit with that. You know, when we first acquired ClickView, we started loading data into it and everyone was amazed. You know, reports that used to be two-dimensional and static. With BI, they became multidimensional. No one needed to ask for a second or third report to answer the questions that they needed more detail on. So, you know, many people in the organization would learn how to use it and spend countless hours in front of it each week, you know, answering their questions. And there was a lot of value in that. Now, though, clients are telling us they don't want to have to dig through a BI tool to search for opportunities. What they want is a system that finds those opportunities and lets them know about them and, if possible, even take the corrective action on those. So, you know, we can use a combination of our knowledge of the supply chain and machine learning within our products to help clients identify opportunities that can improve their processes and their bottom line.
Bob - 00:04:46:
You know, it's interesting because today we encounter so many people going out to the web and getting on the ChatGPT and asking all kinds of questions. And they think that they're doing AI, they think that they're doing something around machine learning and John, These folks really aren't doing Generative AI, are they? What really is Generative AI?
John - 00:05:14:
The answer to that is pretty long, but the intuition is that when a model has been built of the real world, what the model has seen before, you can sort of feed that backwards and say, what could you see that would be consistent with what you've seen? And so you can ask for a painting of an astronaut on Mars or something like that because it has seen an astronaut before and it's probably seen Mars before. And you could probably also add other qualifiers to make that even more interesting, like with a party hat on. I mean, that's the gist of it. There are several underlying technologies such as transformers, techniques within natural language processing and, of course, deep learning. And that's a pretty long discussion on exactly what makes that work. But the gist of it is, and I think that it's interesting because people may think that it is generating a new idea. And that's not necessarily true. Generative AI has the... The ability to show you things that may not exist, that could exist, but they're not necessarily reasoned ideas. So I guess there's... I just wanted to add that caveat to make sure that... It's not that we don't assume that it means that it's actually generating something new intellectually.
Bob - 00:06:49:
Right. And that's, I think, where people get really concerned and worried about what AI might do in terms of making decisions. And Joe, you have a... Deep background in data. And I would think that having the right data to use in conjunction with AI is really where a lot of the rubber hits the road.
Joe - 00:07:12:
Yeah, that's going to be really critical, Bob. I mean, in the chat ChatGPT world that we just talked about a little bit. It's searching all over the internet and in the internet world There's valuable data and there's, not so valuable data, right? People look at different areas. The good news is, is when our logistics world We can control the data. That the AI and machine learning is using, right? To really give the valuable insights. And as Norm said, we want to push the opportunities and the issues into the users so they can take action quickly. But it is about the data and being able to organize the data, scrub the data. Now, AI/ML allows us to work with both organized and disorganized data in just layman's terms. So it's very important to just try to get the right business questions and try to make sure you're looking at a quality data set that is factual and information to allow you to drive the appropriate decisions. To improve your business. And to actually help people be more productive.
Bob - 00:08:14:
And Norm, at TransImpact, one of the ways that you guys have generated the market leadership that you've achieved is by providing actionable insights through intelligence and helping your customers manage their spend more appropriately. So what are some examples of how you're using machine learning and AI specifically with your customers to help achieve those things.
Norm - 00:08:42:
Yeah, so on the actual insights, one of the first things we're looking at is adding a fraud detection actionable insight. So, you know, there is a surprising amount of fraud in the parcel industry, probably in the freight industry as well. I have seen it in a past experience that I've had where we had an employee that actually was stealing product, selling it on eBay, and then actually shipping it out of the company's shipping doors. And using the company's labels. And I've seen some statistics that say that fraud could be as much as even 5% of a shipper's spend. So what we're looking at is using machine learning, is identifying some of these patterns. And actually, even with our tool today, as a BI product, we've had clients that during the demo, one of the clients actually found $200,000 in shipping fraud during the demo of the standard BI tool. And so now what we're doing is going to expand that and make it so that our actual insights point that out, utilizing machine learning.
Bob - 00:10:05:
That's interesting. And so I would gather, though, Joe, that to really... Make AI and machine learning come alive for your customers, you still have a lot of traditional and people and processes that have to be built around it to use it effectively and to make sure that they're getting the value out of it versus it just being a technology deployed for technology's sake.
Joe - 00:10:31:
Yeah, 100%, Bob. I think the real value comes when you can combine, you know, people's talent and industry and company experience together with the machine learning. The insights that the machine points out has to be combined with the person to really take a look at it and understand some of the nuances that the recommendations are coming with, right? When you combine this powerful technology, it just leverages the human capital as well as the machine capital, right? And then those two things combined is where companies will really win and compete. And beat their competition when they learn how to combine the people and the processes to leverage them to be most productive. And most importantly, make the right decision in as timely as a fashion as you possibly can. It is all about speed to market and being able to react to these things as quickly as possible. And what the computer does and AI and ML does is allows us to get to that information. So as Norm said earlier, we don't have the time or capacity with the resources that we have today and most companies have to go out and find out where these problems are. The machine bubbles them up, but it only bubbles up when we can ask the right questions, right? Machine learning helps us analyze the data, but we got to ask the right business questions. We got to look for the first opportunities. Then we can have the machine go out and crank through the data and then validate or invalidate our hypothesis.
Bob - 00:11:56:
John, I think one of your responsibilities is to make this come alive within TransImpact. And you're, I'm sure, working with a lot of colleagues to kind of make this happen. What are you doing to build the skill sets of people in your organization? Where are you finding people that have the skills necessary to do some of the things that Joe was just talking about?
John - 00:12:20:
You know, the engineer and data scientist has to come in with a... A pretty broad background in techniques, but applying those techniques to parcel, to supply chain, inventory optimization, those, really require a depth of experience that we get from our customers, we get from our professional services division, and from just a long history of being in these domains. So it's definitely not a one-way street. We have to ask a lot of questions in terms of What's the objective? What are we trying to optimize? What are we trying to look for? And then we can apply mathematical techniques to you know, to take that up to the next level.
Norm - 00:13:28:
You know, Bob, years ago I had a programmer who came to me from a completely different industry than I was in, we hired him. He made a comment after about a year. I don't know if it's more important to know how to program or to know the business because there's so much to learn about a business and how it operates. And what I have learned over the years, having tried both bringing up people from within the organization to teach them how to program and from bringing programmers from outside the organization and teach them the business, is that it's much more important coming in to start to know how to program. So in this case, to understand data science, to understand math, to understand those things. That's the most important piece. What uh... You know what John was talking about is you know we've uh... You know, the people we've hired, we've brought in from the outside world with that knowledge base so that they can hit the ground running, working with our team of experts to train them, to bring them up to speed on what we need to accomplish with that skill set.
Bob - 00:14:37:
Interesting. And when you think about TransImpact, and you think about bringing machine learning and AI to the forefront, how are you actually making that available within your offerings? Is this. Just a way to differentiate the offerings that you had and to take them to the next level? Or do you actually have products that are, that are specifically related to AI and machine learning. Joe, maybe that would be a good question for you.
Joe - 00:15:09:
Yeah, I mean, I think what we're doing is incorporating them into our products, right? So, for example, our forecasting tool in our Evercast demand planning and forecasting product. Uses over 250 algorithms. They're all methods of linear regression and different variables on top of that. We're able to incorporate machine learning. And there's three or four different machine learning techniques that we can use and experiment with and have it work through. We want to incorporate that in order to generate the most accurate forecast. We can generate the most accurate forecast by using traditional mathematical models, as well as machine learning, and then apply those to determine by doing historical tests in the way that we approach our product. So we believe we have one of the most accurate forecasting products on the market today. And by incorporating machine learning as additional algorithms as one example of how we're incorporating it into our products. The other one that Norm talked about was actionable insights. You know, for 15 years, our BI and business intelligence tools have been trying to push information and opportunities to the user so they wouldn't have to fish as much. We can give them the answers. One of our goals here is to use AI and ML in order to drive forward AI on having that create more impactful, actionable insights, right? So what we want to do is help the customers with our logistics and transportation experience and our manufacturing distribution and other industry verticals. We want to incorporate these things into helping customers solve real business problems by bringing those insights there, asking the right business questions, and using AI and ML technologies to help drive those solutions faster, quicker, and better. And that's ultimately our goal is to incorporate these into our products to make them even better than they are today.
Bob - 00:17:06:
Well, that's interesting. You know, you're incorporating AI and ML under the hood. And I would imagine that is creating a differentiation for you in the market from your competitors when you're talking to your customers. From a non-technology perspective, how do you get them comfortable with that differentiation and that you're bringing ML and AI to the table in a way that your competitors are not?
Joe - 00:17:35:
Yeah, the big thing is, how do we help them drive business value? So what we want to do is put it into the business case, right? So what we do is in our products and our demonstrations is we show them how our product is the most accurate. Right? We can show them through historical analysis and proof of concepts on how to generate the most accurate forecast and how we can prove by taking out their last six months of data out of their components. And then we can show them the different algorithms that actually gave them the most accurate forecast. We can show them the techniques where our machine learning algorithm A or machine learning algorithm B and we give out our secret sauce. Basically generated the most accurate forecast. We can show them when we say, Boy, wouldn't it be great if you came in in the morning and we showed you which products had the lowest margin and why and where your cost leakages are, right, what we do in our business intelligence tools. And AI can actually help them do that. And then when we get into the products itself, right, when you go into the AI and you go into your dashboard, you're going to your BI, I'm sorry. And you go into your dashboards. I mean, wouldn't it be great to just ask the question, Use the natural language processing that Norm mentioned earlier and just ask the question of the BI tool and say, hey, what's the most unprofitable products we have? Do we have a sales region that is not performing of the standard? I don't want to just look at the chart and try to figure out that my southeast region is underperforming and is in the lowest quartile of our business performance. I'd like to go and say, who's my worst performing business rep? Who's my worst performing product? Which customers am I making the least amount of product? Imagine if we could ask that and then our AI. Can help search our business intelligence tools. Again, I think our secret sauce is how we have organized our information and our data, how we've set up industry-leading best practices in going after these important profitability and logistics questions. And then basically incorporate that into our product set. And we will be able to do that by walking our customers through that journey, through our demonstration and our proof of concept processes.
Bob - 00:19:45:
Now, Noam, when you're interacting with customers, which I know you do a lot, so you have a definite direct customer perspective on this, are they asking about AI and ML, or are you having to bring them over? Kicking and screaming.
Norm - 00:19:59:
No, actually, what's interesting is with AI/ML becoming so front of mind right now, Any organization, CEOs and boards, they're all starting to ask the question, do we have AI? How are we taking advantage of it? And they're actually pushing that down in their organizations and we're hearing it directly from some of our clients. Basically, the clients are telling us, look, our board is telling us we need to start incorporating AI wherever we can. So, you know, utilizing that, we have already started working with several customers and started to implement AI within those organizations so that they can start to see the benefit and understand what the real benefit and value is from it.
Bob - 00:20:47:
John, as a data scientist and one of the leaders at TransImpact for making ML and AI come alive, What are some of the pitfalls that you're dealing with to make sure that you're doing it right and that you're not taking it down the wrong path or otherwise making things more difficult than better?
John - 00:21:13:
Well, Bob, I don't want to sound overconfident, but it's all been fun. And we work with our customers to prove what we're doing with their data. So, uh, You know, we obviously have a lot to measure and make sure that we're improving accuracy and Yeah, I mean, so I guess we work with our customers and their data and And, uh, We have a lot to measure, and so it's kind of hard to go wrong there.
Norm - 00:21:49:
You know, one of the approaches that we've taken to identifying where to utilize AI is first, we've gathered a list of problems that our clients wanna solve. And then what we've done is we evaluated that list and determined, is it a problem that requires AI/ML or can it be resolved through traditional methods? Not all problems actually require AI/ML, which is a common misconception people have out on, you're reading about and hearing about. People are thinking you can use AI/ML for everything, and it's not always necessary, and sometimes it's overkill. You know, you gotta use the right tool for the job. Traditional statistical methods sometimes, and often are, are still very good and very useful. Then, you know, once we've done, you know, we've created the list, we've evaluated it, then we have to determine, you know, if we have the data necessary. You know, Joe mentioned earlier, it requires data. AI/ML requires considerably more data than traditional methods. And the more data you have, the more granular it is, the better your outcome can be. So, you know, again, we've gone through a process that we've evaluated. And then what we did was we targeted what I would call low-hanging fruit, some quick wins that we can then incorporate into our products on our client's behalf. And we have some additional ones lined up on deck, so to speak, for future enhancements.
Joe - 00:23:19:
You know, I think Norm makes a really good point. I think a lot of people today are just, they're a hammer in search of a nail with AI and ML. What we need to do is we need to make sure we're asking the right business questions and we're using the right tool for the job. And I think that our years and years of years of experience in the logistics arena have allowed us to focus where we want John and our data science team to focus on those components. And as Norm mentioned, we're doing that very collaboratively with our customers. We're working with them, figuring out where are their biggest opportunities, what are the issues and opportunities they're trying to solve. And then we're trying to look at our tools today. And Bob, you asked the question, how do we incorporate it? We incorporate it by making sure that our tools are helping our customers improve their profitability, reduce their costs, and improve their customer service by basically making sure that we're answering the right questions and using the right technologies. And then we'll point out to them when AI and ML was the right approach and when our just basic structural and already traditional techniques has been able to do the job. And as you mentioned, it is putting the system together. It's putting the technology, the people in process. So we bring our logistics industry experience together with our technology tools, together with our clients, and that allows us to create a winning formula. When you put all three of those together. And the people in the process work with the technology, that's when you make a real difference.
Bob - 00:24:48:
So, John, Joe was just mentioning some of the ways in which TransImpact is using AI and ML within your customer base. But I think another area that you're working on is around new product forecasting. Can you tell us a little bit more about that?
John - 00:25:04:
New product forecasting is one of the hardest problems in supply chain. It gives a lot of people grief because they're worried. Am I picking the right number? Am I going to order too much? Is this product really going to be successful? So there's a lot of opportunity there. And I don't want to give too much away, but there's a lot of opportunity to help the demand planner not only know how well something will sell and what its adoption should be, but work a little more prescriptively to give them guidance on what direction to take a product line. That sort of thing. Can't tell you everything.
Bob - 00:25:54:
Well, clearly TransImpact has some secret sauce that I'm sure many of you listening to this would love to hear more about. And I'm sure they would love to share that with you, too, when the time is right. Well, clearly, TransImpact is effectively using ML and AI in the right places, in the right way to help take their customers to the next level. And this has been an exciting discussion today on this topic. And the three of you are obviously real leaders in this space. And I look forward to hearing more great things about what TransImpact does with ML and AI in the future.