Listen to most recent episode of Road Signs

SEASON TWO

SEASON TWO: EPISODE THREE

HOW DOES A.I. CHANGE TRUCKING'S BACK OFFICE?

Did you ever wonder what might happen if your back-office decision makers had time to strategize and think, rather than be weighed down by menial tasks? Could the digital brain help support your real human brains? And could they do so in cost-effective ways? In this episode of RoadSigns, host Seth Clevenger takes a look at the way an entirely new technology changes trucking’s and freight transportation’s back office. As artificial intelligence and machine learning begin to appear in trucking and logistics software, he wonders, “How is this trend changing transportation management? And… what is the future?”

FEATURED GUESTS

Parker is passionate about building teams and supply chain systems. He currently serves as the Founder & CEO at Fraight AI, his third logistics business.

Ram is a Principal Data Scientist at Omnitracs, where he develops analytics, predictive models and cognitive AI solutions that help leading fleets make critical business decisions. Ram has over a decade of experience in software engineering, and for the last five years has focused on building data solutions

Parker Holcomb

Ramprasad Renganathan

EP. 3

Brought to you by:

Episode Transcript

Roadsigns S2E3.mp3

Dan Ronan: From Transport Topics in Washington, D.C. This is RoadSigns. Here is your host, Seth Clevenger.

Seth Clevenger: Thank you for listening to RoadSigns, the podcast series from Transport Topics that examines the trends and technologies that are shaping the future of trucking. In this episode we're going to take a look at how artificial intelligence and machine learning could change the way transportation companies do business. While artificial intelligence isn't new, it has become much more accessible in recent years as more data has become available along with greater computing power at lower costs. As a result we've seen a strong push by many technology companies to take advantage of AI and machine learning and we're seeing that trend emerge in the transportation industry as well. The term AI tends to conjure up all kinds of images from movies and pop culture. But we're not talking about killer robots or Skynet from Terminator. What we're really dealing with here is software designed to help people work more efficiently and make better business decisions. But how will this move toward AI change the jobs of dispatchers, fleet managers, freight brokers and other workers in this industry? We'll try to answer that question in this episode. On one hand AI can allow businesses to automate simple but time-consuming tasks like wreaking information from freight documents so workers can focus instead on more productive tasks such as interacting with customers or drivers. AI models also can be designed to analyze data and maybe come to the same conclusions as a human worker, but do so faster and more efficiently while reducing errors. Beyond that, AI programs may be able to identify patterns and trends that a person never would've been able to spot on their own. And this automation will be fueled by more and more data from onboard sensors, telematics, electronic logging devices and onboard video. To learn more about what the future may hold, we've lined up a couple of guests who are applying AI and machine learning to freight operations today. Later in the program I'll speak with Ram Renganathan, a principal data scientist at Omnitracs and the trucking industry's largest suppliers of in-cab communications, telematics and data analytics. But first I'm excited to welcome Parker Holcomb, co-founder and CEO of Fraight AI, a technology enabled freight broker. Thanks for joining the program.

Parker: Thanks for having me.

Seth: So artificial intelligence and machine learning may still sound somewhat futuristic to some people but as you know many businesses are already using it today and the transportation industry is no exception. So Parker I'd like to start by getting your overall assessment of the current status and future potential for AI in transportation. Why is A.I. a big deal for our industry?

Parker: So AI is definitely in its most recent resurgence and that's been due to a lot of convergence of cheap computing power, the ability for new data to be processed, but there's not much new about the techniques and algorithms that we're using. The term AI was coined back in the 1950s and now it's gone through a different kind of resurgence as winters and springs as they talk about it but today it's so accessible and it started to actually drive people's bottom line.

So the reason that you may not see AI everywhere or the reason that you might not say, “Oh, hey, well, what is this new technology because it's familiar.” So you know two different areas of artificial intelligence in the toolkit that we'll talk about. Number one is recommendation engines. So recommendation engines are how Netflix finds your next show to watch or how Amazon finds your next frying pan to sell to you. And so just like 20 years ago you would ask your friend, “Hey what movie should I watch?” He goes, “Well I know you really liked the Star Wars you know flicks … and so, you know, I think based off of my past experience and observations of you, Seth, you would really like The Matrix. And so that was familiar to us.

And so it's taking that personalized past data and that experience and delivering something to us that you know feels familiar. Now we're going to zoom out our industry, which is one of the most massive industries in the country, in the world. There's so many different points of data, there's so many relationships that it's hard to keep them all straight. And so, you know, historically an intermediary or a freight broker they might know 20 or 30 different trucking companies and they might know those preferences and behaviors. And so they in turn can recommend, “Hey I think you would be good for the Pennsylvania to Miami route. I know you like to be at the Jags games in the fall.” And, you know, AI based on our previous experience can make this recommendation to you. And so as we start to scale up and think about how many different options or how many different people there is no way that any one freight broker, any one human could keep this all straight. To put it in perspective if there were 10 trucks and 10 cities there would be 3,600,000 different ways to organize those resources. If there was a hundred cities and 100 trucks it would be 132-digit number of ways that you could organize. Now fast-forward to reality which is 3 million trucks and millions of pickup and dropoff locations. No human could ever optimally make all of those different matches or check against all those preferences or see where people have been in the past. And so the recommendation engine recommendation algorithms you can take a lot of those past experiences and whether it's hundreds or millions or billions of data points and cross-reference it with current options and opportunities which again might be millions, billions, or hundreds of billions of options and say hey, based off your preferences and what you like to do. I think that this could be a good match for you. So AI is particularly well-suited for this industry because it's actually the most human-centered technology. The past three generations of tech which have been desktop, web and mobile. Those have been generations of technology where it's asking people to speak computer and AI as a generation of technology whether it's computers learning to speak human. And so it's not going to be this like, oh hey, we just got taken over by this new set of technology. It's here today it's slowly approaching but it feels familiar and it's actually human-centered, which makes it feel like a more natural transition.

Seth: Let's go ahead and talk about it a little more about some of the core benefits of AI.

You gave your example of all the different permutations of ways that you might want to, you know, route trucks or determine loads and shipment and just figure all this out all the all the many possibilities. But one of the common threads for using AI is to automate some of the mundane repetitive tasks that can be time-consuming but really aren't very maybe mentally stimulating. This theory can help workers focus instead on more complex issues and get more involved in customer service. I want to get your take on that as well and how much of the value of AI comes from improving worker productivity by kind of relieving workers of some of the grunt work and in repetitive tasks that are out there.

Parker: I think it falls into two buckets. Yes it's about, you know, relieving repetitive tasks and kind of those are those low-hanging fruit that's kind of the repetitive bucket but then there's also the superpowers bucket, right. The … you know, some of the algorithms today can make a better recommendation on what trucking company to use than 15 years ago five Ph.D.s sitting in a room for five straight days. And by the time they're done with their analysis the information is over. So it's definitely about enhancing the productivity of those workers but it's also not just removing the mundane. It's also giving not giving them superpowers.

So, you know, an example of how we do it here at Fraight. So let's talk about dispatching a truck. So any legacy freight broker, traditional freight broker -- the carrier rep should find when and where that driver is supposed to be empty if it's 45 minutes away and an hour before pickup. That's a lot different than three hours away and a day before pickup but you want to be able to check out and make that verification. So, you know, maybe first step is you reach out to the dispatcher, you find out when and where they can be empty. Then you set a reminder to reach out at that time to confirm that what's happened, then the dispatcher says, you know, let me check with him I'll get back to you in five minutes and you have set another timer five minutes. Then they give you is tracking info or maybe a phone number. You try to call once, you try to text. You've been spending 45 minutes thinking about this and you don't know if it's a problem you don't know it's OK. Forty-five minutes later we finally get the driver on the phone he says, “Hey sorry I was taking a shower,” I will be at pickup in 15 minutes. So everything was OK but you still spent 45 minutes trying to check and to figure out setting timer one to follow up the appropriate time. And so what Archie does is like, you know, our internal assistant, our co-pilot, he will know that he needs to reach out at 3:02 p.m. and reach out to the dispatcher if they say “hey I'll get back to you in five minutes.” Well then Archie can usually follow up in six minutes and say “hey just following up here.” Archie can get the driver tracking information, can send a text, and then you know if it's still an issue or if it's outside the normal parameters of response time Archie can pop up to one of our agents and say hey I've done X, Y and Z. But looks like we need to call the driver. Can I connect to you … connect. Hey was just taking a shower. I'll be there in 15 minutes. So that just took 45 minutes of meaningless, nonsense work and shrunk it down to 45 seconds of a verification. And so, you know, the ultimate dream of an assistant is it brings you decisions to verify, rather than, you know, a problem to solve. So that's an example of it shrinking downtime and just taking what a human could do and doing it automated. But at the same time there's you know other parts of a flow that a human could never do. So, you know, when we're we are reaching out or we're sourcing a new lane a traditional broker might have a hundred, a thousand people on their floor starting to make phone calls and they can make 100 phone calls a day.

They're only capturing as much data as they want to before they move on to the next call. Well Archie with one person overseeing it can have 100 simultaneous conversations per minute. And so, you know, a human can only focus on two, three things at once. Archie can focus on a hundred things at once, a thousand things at once, a million things at once, and then that can start to pull patterns out of those conversations. So let’s say we’re having 100 simultaneous conversations or text messages, e-mails, whatever it might be. All these different communication channels and you start to pick out these different patterns or behaviors or hear the response rate is usually you know, a minute and 30 seconds. The response rate right now is 20 seconds on average. What does that mean? Do we find you know is capacity loosening up? Do we find a great opportunity lane? And so those types of relationships and associations, you know, the best freight broker ever couldn't figure out. So it fits into two buckets.

So it fits into automation of low-hanging fruit it fits in two superpowers and you know Ph.D-level expertise.

Seth: And let's also take a quick moment to talk a little bit more about how AI could change sort of the daily jobs in the industry and just the day-to-day tasks that a typical worker might perform, say some of your employees vs. an employee additional freight broker. As we've been discussing, AI is really here to provide support for employees rather than simply replacing them altogether. But it's just this tool that can handle all this information and make decisions and then the workers now essentially overseeing it and verifying the conclusions and just how different will the jobs of a typical employee at a freight broker change?

Parker: Our target is to automate 80% of the process. Yes there is the opportunity is for advanced insights AI is not going to just make your life easier. It's actually going to make your life harder. You're not going to show up to work in two years and just be sipping a coffee while Archie does all your work for you.

Archie is going to be doing the low-hanging fruit, the easy stuff and you're going to be stuck solving hard complex problems working on relationships making tough decisions, you know, weighing the tradeoffs of what the customer wants versus what the carrier wants versus your insurance company who recently got involved versus a new company or pre-existing so you know it's somewhat of a paradox here that it's actually going to make all of our lives a lot harder as we try to keep up and still outcompete the algorithms.

Seth: So really we'll be doing more sort of strategic thinking and handling relationships rather than some of the more mundane.

Parker: And what we're great at is balancing tradeoffs and interacting with humans. And one of my favorite things to think about is I still believe that there will be radiologists in 10 years, they'll be using the best AI image-recognition software. But you know once that AI presents it to a radiologist he'll have to weigh the needs of the patient the insurance company, the consent of the patient, the family and the hospital. And, you know, an algorithm is not going to appear in court either.

You know it's going to be out there, you're going to be interacting with people, it's gonna be a long, long, long, long time before we have general intelligence that can hold a random conversation but there is a whole lot of opportunities to enhance our process and make our systems more efficient.

Seth: Before I let you go, I do want to ask you the typical crystal ball question. Where do you see AI in the transportation business say 10 years from now?

Parker: I think I have a primary opportunity for it for AI will be to listen to all the needs across the supply chain, there will be billions of sensors throwing out different pieces of information from electronic devices to sensors on manufacturing equipment talking about inventory levels. And I think AI will be the primary tool kit that listens to all of this internet of conversations and is able to pull out suggestions and insights on how we can efficiently run our supply chains and allocate our resources while humans work on a hard or soft -- the relationships and moving things forward.

Seth: Well, I think that's a great place to leave it. Thanks again for joining the program and sharing your insights.

Parker: Yeah, thanks so much. It was a pleasure.

Seth: Absolutely.

Next on RoadSigns, we're excited to welcome Ram Renganathan, principal data scientist at Omnitracs or the trucking industry's largest technology suppliers. Thanks for joining us.

Ram: Thank you sir. Pleasure.

Seth: So I attended Omnitracs User Conference in Dallas earlier this year and one of the big topics of discussion was how data and AI could allow fleets to begin making better business decisions based on cost for hour rather than cost per mile. I thought that was a pretty interesting take. So we'd like to dive into that here with this interview. First off why does it make sense for the industry to shift its focus toward time as the most important metric rather than distance?

Ram: So I mean let's talk about this huge problem that the industry is facing today. They're really struggling to retain drivers. Let's take, for example, a key driver drives about an average 2,005 miles per week. And with all the stops let's say he drives 70 hours a week. And if the rookie is making 30 cents per mile. Then he's just making $750 a week which comes to about $10 a week. And then on the other side of it we wonder why you're not able to do both. So the miles might not be a good measure of how do you pay this. So the other way if you see if you paid these drivers by hours or ... Then I think that is a much better way of paying them. It will also be a better way of retaining your drivers.

Seth: Sure. Certainly from the driver’s standpoint time is money as well. Not just miles you think of your driver detention time at shippers and receivers facilities, traffic delays, weather conditions, and of course just hours of service constraints all add up in simply the miles that a driver travels doesn't capture all the work that goes into it and how profitable that can be.

Ram: Yeah, I mean, you know, we did an analysis a couple of years ago for one of the large trucking company in the Midwest. And the analysis that we did for basically detention time across the different customers. There was just one specific retailer. We were looking at the data from January 2016 to September 2016. So it's like a nine-month period. And ... at this one retailer there was, you know, approximately 12,000 hours of detention time or just about. And if you just put any dollar-per-hour cost then you can see the amount of money that's being wasted. Just at that one location. So it is a big problem, right. And you know you see the industry is picking up on it. And Omnitracs has all of this data. About you know, where the truck picks up the load from and why it's dropping off the load too. And we have enormous amount of historical data across the U.S., Mexico and Canada. And at each point or each geolocation we can actually say, you know, this was the amount of time that we spent at that location. Which is a proxy to detention time it could be, you know, the context also matters because if it is if he's detained on a road. Or if he's detained on a highway. Then that could be traffic. But if we actually know it is where he's loading or unloading. Then we can say it is detention. So the Omnitracs has a lot of data to do those kinds of predictions.

Seth: And your CEO Ray Greer at the user conference earlier this year did make the argument that your company with its, of course, a very large base of fleet customers collects enough data all these different locations to pick up delivery locations really across the continent to calculate detention time really for each individual site.

Do you feel like you're at that level now where you can now just kind of look at this not just as a, you know, kind of a broad issue but let's pinpoint the real wait times at each facility.

Ram: Yeah, I can personally say because I work on the project and, you know, there's lots of people he's different because he's being built internally right now. And the results actually look pretty promising. We cannot exactly say that for all of the locations but for a large amount of the locations that Omnitacs has. I think we can confidently say we can predict that retention time.

Seth: And I also want to talk a little bit about truck parking you know this is another big issue that drivers are encountering these days and we're hearing that drivers may have to spend some extra time at the end of the day just searching for parking. Or maybe they stop their day a little bit early because they know they can actually park at their current location and maybe they're not so sure. A little bit further down the road. So that's also eating into time a little bit. So how can technology companies like Omnitracs start to get a handle on the time associated with just finding available truck parking as well.

Ram: Yeah, I mean right now you see different companies providing these parking-related apps and saying, you know, in the next 10 miles you have a parking spot.

But, you know, I think Omnitracs has a unique perspective because it has the geolocation of the truck. From point A to point B. It also knows the hours of service involved and how many hours the driver has. On the day to how many more miles can you drive how many hours can be left in this day that he can drive. And if you can … if we can add, you know, locations of parking along the way you feel you know. If you're able to enrich our data we have a medium level two locations so the entire trip. And one of the other projects that's going on is to interest just geolocation highway. It could be a parking spot. It could be a gas station. It could be a truck stop. Could be loading, unloading and detail that could be a distribution center. So internally Omnitracs is trying to enrich all of these locations. And once these enlistees geolocations meaning … this geolocation is at a particular address. These are the businesses in this address. Then I think these detention times to dwell time to mean, somethings. For example, if you know that this geolocation is a parking spot or a truck stop. And if we say, you know, this is how long trucks wait. Then we can, I think, along the way say OK, you have this many hours of I have left in a day to drive and you are entering Atlanta or a big city and you're not going to make it to the other side of the city and you might want to park here. So I think Omintracs, with the amount of data that it has and with you know, the different data sources that it has. Geolocation in terms of all these. Enriched geolocation and also hours of service I think we are in a good spot to say, you know, predict parking spot and also say this is a good spot for you to park.

Seth: OK. And as we discussed was relatively easy, of course, to measure miles you know how far a truck will need to travel to make a delivery. But some of these examples that we've gone over we see that time is much more complex it's harder to predict. But now with the data that's increasingly available we can start to understand things like detention time and traffic delays and hours of service availability to make better assessments of the actual time involved in making a delivery.

So once you do have this reliable measurement for cost per hour what do you do with that and how do you see fleets choosing to accept or decline loads based on that time or even start pricing their services that way.

Ram: Yeah exactly. So that the example that I gave for the trucking company in the Midwest. When they accept a load from the dealer again they actually say know hey ... give them specific numbers on this is, how long we spend in detention when it comes to one of your retailer or one of the offices. And we this is this is not something that this is either you reduce the detention time or this is how much money you need to pay us. To pick up this load and drop it off. So these are conversations that a trucking company can have with their customer. And negotiate better dates. Or even think about dropping that customer altogether to make, you know, the operations much more profitable. So when you export this kind of data I think that the customers our customers in a much better position to negotiate.

Seth: Absolutely. Especially in an industry like trucking that operates on pretty thin margins, you know, having that intelligence and knowledge of exactly what is profitable and what isn't profitable is really something that seems perfectly suited for this industry. And I also want to ask you just how AI will factor into these cost for our calculations. I mean do you see this making it easier and faster to make all these kinds of determinations and calculations looking at all these different factors at once and the use of AI makes it so much easier than trying to have a data scientist like yourself trying to crunch all the numbers and do this manually.

Ram: Yeah I think to some extent, you know, with the amount of cloud computing resources that we have and the powerful algorithms that we have. And also the most important part being you know the amount of good data that Omnitracs has, the historical data that it has. This is something I think AI and machine learning can be, will be, able to crunch much more faster. And much more efficiently.

Seth: And how else is Omnitracs using AI today and how do you see that expanding in the future?

Ram: We have been building predictive markets machine learning models for the last decade or so. So currently we use it in terms of predicting accidents. So we can do we can predict accidents using several different data sources. You can go to a customer and get all of the data they have. And build custom machine learning models for them. Or we can just use the hours of service data and build predictive models. So some of the most popular products for us are, you know, predicting driver quits. Drivers who quit in the next week. Or in the next month. Or drivers would have an accident in the next week or next month. The way I see AI and machine learning being used in the future is you know you want to see AI being used in pretty much every area of trucking in safety in operation efficiency. But you're trying to recruit drivers. You know, when you have so many applications coming in. You want to make sure that you get the right kind of drivers into your company who will fit your company culture. If not then that's going to lead to more attrition. So recruitment is another way you can use AI machine learning. Retaining drivers which are already on the trucks desk is another way. And with all of these dashcams coming into trucks. Right. There is also going to be a lot of real-time predictions on traffic sign violation. Traffic light violations, stop sign violation, following too closely. There's going to be a whole lot of additional critical events that can be generated using a computer vision, using computer vision and mission learning.

Seth: There's certainly a lot to look forward to in the years ahead. So much is happening and there's so much more we can understand when we would use the data the right way. You know, this has been a lot of great insight but I do think that's a good stopping point. So thanks again for joining us room and sharing your thoughts on this.

Ram: Thank you.

Seth: Let's take a moment to review what we've learned and revisit our original question of how AI will change jobs within the transportation industry in the years ahead. It does appear likely that the day-to-day tasks performed by workers in our industry will shift to some degree at certain functions become more automated through the use of AI but will not automate everything over time, computers will become more adept at performing repetitive tasks and analyzing data to recognize hidden trends. But they won't be able to build relationships with customers or talk to a driver who's having a rough day. The industry will still need to rely on people to handle the human interactions that are at the heart of our industry. And, of course, people will still need to make strategic decisions and solve problems as they arise. But with AI and machine learning transportation workers will have more powerful software tools to support them in their missions and hopefully help them take their businesses to the next level.

That's all for season 2 of RoadSigns but season 3 is right around the corner. In the coming weeks we'll be taking a close look at how fleets are making progress on the endless march toward greater efficiency and fuel savings. Until then I'm Seth Clevenger. Thank you for listening.