AI In Transportation With Chris Torrence And Warren Powell
People are always afraid of the moment AI will replace them in their jobs. Whether it be in transportation or in sales. People are afraid. The truth is, there’s still a long way to go before AI takes over your jobs. It’s mainly because they lack the decision-making function that an average human has. An AI can do many things but the ability to trust and build relations is something they can’t do yet. Join your host, Chad Burmeister, and his guests, Chris Torrence and Warren Powell. Chris is the Head of Growth and Strategic Partnerships at Optimal Dynamics. Warren is a Professor Emeritus from Princeton University as well as the Chief Analytics Officer at Optimal Dynamics. Listen to today’s episode as they talk about AI in the transportation field and how AI decision-making is advancing towards the future.
Listen to the podcast here:
AI In Transportation With Chris Torrence And Warren Powell
I've got a couple of special guests with me. Usually, we do this mano y mano. We're going to go the three of us. Chris Torrence is responsible for Sales and Business Development at Optimal Dynamics. Professor Emeritus from Princeton University and also, the Chief Analytics Officer from Optimal Dynamics is Warren Powell.
Thanks for having us, Chad.
We had a call and we dug into some of the content. I'm very excited to share with our readers how AI is being leveraged in the world. Every call that I have every week with different companies, I see it moving up and up and it's coming to the forefront. I'm excited to dig in here. Without further ado, why don't we start out with Chris? Do you want to share a little bit about Optimal Dynamics? What's the concept of the organization? What do you guys do?
One of our challenges is how do we simplify or distill some complex technology in a larger part to what Warren spent a lifetime creating. The simplest way I put in a tagline that seems to resonate with a lot of industry folks is we were this decision layer of logistics. We have aspirations to ultimately become the decision layer of the broader supply chain ecosystem. When you think about a lot of the complexities that are involved, whether you're managing a truckload, freight network, as we're always seeing here as a supply chain disruption taking place and these black swan events, there are a lot of important decisions strategic and tactical. In the real-time world that we feel relative to make those optimal decisions that our human brains are not necessarily equipped to handle.
We are a SaaS-based software company. We are an AI play, which Warren will tell you. There's a lot of hand waving around AI and machine learning which we can get into. At the end of the day, we're not here to replace existing transportation management systems. We're not a brokerage or digital freight match. Think of us as a layer that sits on top of all that is strictly focused on making the best possible decisions by leveraging advanced technologies.
It's exciting for me because the first two jobs I had at a school, Airborne Express and FedEx, FedEx traveled to China. We talked LTL, FTL, and then break-bulk distribution. The Dell model before it was the Dell model. I’ve met the guy who co-invented it with Michael Dell. He was part of FedEx. Now AI, being part of those decisions, it's not just an Excel spreadsheet anymore that says, “Here's your cost to a warehouse, freight, or transportation whether you’re bringing it in a boat or an airplane.” A lot more can be done. Professor, if you're looking at AI and transportation decisions that are this complex, how does AI fit into the process?
I spent a lot of time on this thing called Truckload Trucking. The biggest complication is you'd had to make decisions under uncertainty. I started to realize the academic community wasn't very good at making decisions under uncertainty. Let me fast forward through many years of research. We got to the point where we’re like, “I think we understand it.” Optimal Dynamics is very focused right now on optimizing truckload trucking. There's a lot of decision problems like which drivers should move a load and which loads we would say yes to. In the background, I also got involved in an activity which is relevant to Optimal Dynamics. I call it Optimal Learning. Let me talk about AI in a sales context. I want to make sure that I'm talking to your readers because a lot of Optimal Dynamics is mainly, “Here's AI for a complicated industry called truckload trucking and we'll lead it into supply chain management.” Let me talk specifically about AI for sales because a lot of people don't know what it is. For most people, it's what we call Machine Learning.
Predicting the probability, that's machine learning. Deciding who to call, that's a decision.
In a sales context, you might be saying, “Who should I call?” On the internet like Google, Facebooks, or Amazon want to put up an ad. They want to put the ad that has the highest likelihood that somebody will click on it or Chris needs to know of all the people that he can call, which one should he call that he has the highest likelihood that somebody will say, “That sounds interesting.” We could take a large data set and say, “Here are all the customers we reached out to who said yes and here are all the ones who said no.” You can take that kind of data, throw it through a machine learning package, and come up with something that will predict, “Here's a customer I've never talked to before, but given all of his attributes, I can predict whether he'll say yes or no.” That is a classic use of machine learning in sales, predicting who will say yes or which ad or image is going to be the most effective. This is classical AI. That's not a decision problem. That's a machine learning problem.
A decision problem could be, “Here are all the people I can call. Who should I call?” That's a decision. Chris has to make that decision. He's got this long list of people that he knows from his experience, but who should he call? If he had a machine learning tool that will say, “Here's customer's highest probability versus lowest probability.” That would be helpful, but it doesn't entirely determine who he should call because there's a lot of uncertainty in that. Google and Facebook will take powerful tools or say, “Here's the highest probability and the probability isn't so high, but I'm not very sure.” Because of the uncertainty, “I'm going to decide to call anyway.” Predicting the probability, that's machine learning. Deciding who to call, that's a decision. Optimal Dynamics does the decision problem.
There's a colleague of mine, Ryan Reisert, who's doing this around the channel of communication. The highest level order, who do I call and when do I call them? The second decision, do I call them in the first place, do I email them, or do I connect with them on social? He's created this thing called a phone-ready lead that his people that are lower-cost labor calls through a list five times each. If you don't pick up after five times, then your propensity to pick up is quite low. Let me give the list to the reps of people who we've known picked up the phone. It might only be 18% of a list of 100 people, but if I call that list, they're going to pick up 1 in 5. I could see how the decision analytics could leverage that information to not only tell you who to call but then what channel to reach out to.
Sometimes you have to do what's called Active Learning. In sales, sometimes, Chris will have somebody who he could call, his best estimate may be a lower probability, but then he made say, “My best estimate may not be right.” Sometimes, he's going to have to reach out and make that phone call and alert. That's a decision. We call it active learning. In classical machine learning, you're given a data set and you fit a model. In sales, I have got to assume that sometimes Chris is going, “That's a low probability, but what the heck?” Chris helped me out. Tell me how your thought process goes through when you're choosing who to call.
Anyone in sales will attest there is certainly some science and the analytics component that wasn't outlined in the idea of automating the SDR function. There is merit to that to an extent. I'm a firm believer in this harmonization of somehow leveraging human capital and machines where it makes sense and I don’t think sales is an exception. I love to leverage my vast network of connections through LinkedIn and elsewhere and folks I've come up the ranks within the logistics and supply chain space. With that said, into Warren's point, when it comes to that decision, I'm trying to process known information and leveraging that with what I feel would be a profile or attributes that I think would best fit into the solution that we provide.
There are other things, and Warren will also agree with this. Sometimes, it's the information that's in my head that's never been downloaded into a machine. When people ask you or big competitors are, sometimes that's the easy answer, whether it's sales or in other aspects of our business because there's so much tribal knowledge and experience that are in all of our heads. Chad, yours as well, that have never been put into a formatted data form. It's a multitude of variables that come in. I look forward to one day having more of the recommendation engine within sales to say. There's already been an extensive analysis done, whether it's at an individual level or a company profile that is saying we could put it on a scale. They “don't waste your time” part of the spectrum and then wide hot. That would be super helpful for anyone in sales.
The challenge now is a lot of it is open to interpretation. When we have the SDR, the traditional sense of humans reaching out trying to extract information to see if it's a worthwhile venture for the account executive, there is a lot of loss in translation. There are so many other variables involved that the human brain often can't process and compute all those variables to make an objective output to say, “This is worth your time and this isn't.” Still Art and Science, but I like to think that this decision advancement that was not outlined is certainly going to be very advantageous for anybody in a sales capacity.
The tension between the computer's ability to handle many variables, that's great, then there's going to be things that the computer simply doesn't know that’s in Chris's head. Chad, what is the tension? I know one of the topics you were interested in is salespeople are saying, “Is this going to replace me?” The answer is honestly, it should be able to help salespeople. Remember, we're talking about the type of product where you do have to make a phone call. We're not selling a million of these things. If we could a sale a week, we're doing pretty well. Two a week and we're probably overwhelming the company. This isn't high volume. This isn't Amazon.
Let's say a rep that does $500,000 in revenue on an average deal size of 10,000 or 20,000. Using AI and being served up with when to call what, what channel to use, and advanced levels of AI, now my calendar is full every day versus being 1/2 full or 1/3 full, my close rate holds up. Where it gets is productivity per head, you can now do more with fewer people. What it could mean is, instead of us adding 30% to 50% as a 30% to 50% growth company, I only need to add 10% to get my 30% growth. It has upper promise that I see. I remember Bill Clinton speaking at a Salesforce.com event. I remember he was bragging saying, “When I was president, we raised productivity per head 5% in four years.” He said how massive an impact 5% had on the economy and the GDP growth. I'm curious, from a professor's point of view, deploying this kind of technology with AI and the ability to make these much more powerful decisions, do you see a bigger than a 5% jump at some point?
One of the technologies that we're most proud of, and it seems to have the best impact, is the one that can say, “Here's a trucking company.” We've got all of our customers, loads, and the process of moving. We had this broker's division and a lot of freight comes from the brokerage division. The way more freight than the carrier could have and you want to strike a balance. Our technology can reach into the broker’s division and handpick loads. That impact is huge. We're increasing the profitability of a truck by $0.15, $0.20 a mile which is a massive number for a truckload carrier. I would say it's dramatically bigger than the numbers that you're quoting, but this isn't something where a salesman could do. This isn't something replacing a salesman. This is reaching into this giant pool of freight and going, “Here are 30,000 loads and here are the 3,000 that matter. This is the stuff in.” Right now, if a computer can't do that, nobody can.
It would be too cost-prohibitive to hire a team of 500 analysts to go through and figure that out. They wouldn’t have time to do it.
They still couldn't do it because a lot of these loads, we're not going to move it for five days. We have to be able to say, “Five days from now, that'll be a good load.” No human can do that. Even a massive team can't do it because you have to think five days into the future, and I don't even know where my truck drivers are. This is a case of selling a technology where our technology is not replacing a salesman. It's helping the productivity of the company. Chris has the job of convincing a company that they should buy this technology.
In that example, you shared, 3,000 loads out of 30,000 loads, these are the ones. If traditionally, you couldn't go after that business or you could only go after a small percent, then your growth rate is X. By adding this kind of capability, your growth rate could be Y which is 10X.
This is a new technology, but now Chris has the job of selling that technology. We're a very high-tech company. We are nowhere near replacing Chris with our AI technology. It's not on the radar screen. I'll be honest, that conversation hasn’t even come up. It's not even something that we chat about, not even behind Chris's back.
As long as people are still buying from people, I feel pretty safe in the role. I don't envision that human interaction is going away anytime in the near future. I would agree more.
One of the hacks that we came up with towards the tail end of our last conversation was around if I don't have access to all this AI, how could I leverage some of the lessons learned that as a professor, you picked up over a decade or more than that at Princeton. You gave me an idea. Honestly, I've been thinking about it a lot. Share with me what was that takeaway with the audience of, “How do I pick my prospects just a little bit better to help me increase my sales by putting the right people in the funnel in the first place?”
If a computer can't solve it, nobody can.
That's why we hired Chris. The thing is, I know about inventing the technology, but going to market and getting people to buy into it, sorry. I'm the guy who builds the engine, I don't sell the car.
There’s something that you said that had to do with choosing a customer who has the ability to spend more money, but not necessarily the opportunity to make a decision. If I remember the dialogue was, “Prospect A, I think of our world. I could sell it to an entrepreneur who's just started his or her business and has $0 in sales.” I go to them and I say, “Do you have $3,000 a month that you could spend on this product?” They'd be like, “No.” I'd say, “What about $500? Tell me more. I'm listening.” They don't have the ability to ever spend $3,000 until two years from now or if they raise a fund of money, but a seed round, Series A, Series B, they've got $3 million that they got in their bank. If I spend my time with five companies that started with no money and five that have unlimited amounts of capital or near-unlimited, my odds are going to be better by going after those people. That was one of the hacks that we came up with in the last conversation.
If you had access to a list of all the people who landed a, let's pick, Series A round. Let's say you can do that. That would be a feature. Let's imagine that you now call people with that Series A and you say, “Wow.” They landed their Series A, the frequency of customers who say yes jumps from 3% to 15%. You've discovered that's a predictable asset and computers love that. We love jumping on that. The thing is, maybe you've got some other attributes. There's something about, “I have money to spend but I need some other attributes. I need a type of business that they're in and what my product does for them.” This is where sometimes, we can quantify that. Once we can, the computer's great, but often, not everything is quantifiable. This is where Chris has got a lot of background. He's got a personal connection. We're not about to get rid of all human beings replacing with computers.
One stat I remembered from the Salesforce report several years ago is it aged, but I suspect it still holds up. If you get an introduction to somebody, they claimed your 181 times more likely to sign that customer than if you cold call them.
Chris, you said to me, when I reach out to people in my network and they hear about me, that's the same thing. If I look at $5 million or so that we brought in as a business in the last couple of years, about $3 million of it is a personal connection with me because of personal relationships over time.
What that boils down to is trust. I trust Chad, I trust Warren and that holds true in life. In sales, it's imperative that you have a reputation of integrity and you're not playing the short game of burning bridges. I can tell you in my software sales career, there are certainly junior reps that are looking to make the buck, and they have no intention of staying in a particular industry. They go fast and furious and we'll sell everything under the sun. I'd let them know if you're going to take that approach, that's your prerogative, but if you plan on having a career in any length of time for a particular vertical or industry, that isn’t going to cut it. You cannot risk burning the bridges.
It's always been about whether I was at companies like Uber that has been proven. Companies like FourKites is another visibility provider industry. The recurring theme for me personally was I'm here to evangelize things that I believe in and that I think are going to work. There is no perfect technology. Let's be honest, there isn't and I'm not here to say that there is. It goes back to Chad, something that AI machines generally have this mindset of this automation, “Are you going to replace me? Am I signing up for something that's ultimately going to make me redundant?” My argument is, “We've all been made redundant, whether we know it or not over the course of technology evolvement for many years.”
My theme is technology is critical if you're going to remain competitive or whatever space you play in. The trick is, how do you harmonize all this advanced technology? The understanding that you still cannot replicate certain human interactions and you'll always need some level of human capital to fill those voids. If you can harmonize and balance that out, that's who wins the race. Unfortunately, in our industry, there's a bit of a learning curve, more so than a lot of other industries. One of the challenges is I have to be able to articulate, “Chad, this is here to augment you, not replace you.” In an executive industry and account executives that are looking to cutting costs and redundancy are much more receptive to having those automation conversations, yet they introduce you to the end-users that will be using the technology and there's always friction, which isn’t. That's another dynamic we struggled to tell the narrative.
In sales, that makes me think as a manager or a leader, you can focus on moving an A player to an A-plus. I remember how this was put to me. Let's say that the A player does $1 million in sales. If you move them to an A-plus, they gain 20%. That's $200,000. You've given them the tools and the decision-making context, you grew $200,000 as a business by focusing to move that A to A-plus. Your middle players, which might be 50% of the curve, maybe you move them from a B to a B-plus, you can't quite get them to an A-plus. As a B player, you're at $600,000 and you get a 20% increase. That's $120,000.
Focusing on the A to A-plus by giving them the tools, then the next lot in the middle, 50%, there is a group of folks that might be bottom 10% or 20% in sales where, “First of all, I don't know if I may even get that 20% gain. Second of all, moving from a $300,000 even if you got 20%, that's only a $60,000 gain.” This one is a non-planned question. How does AI disrupt or improve a team if you've got a curve of 20% A players, 50% middle, and then 20% bottom performers? Does it change that curve or does the bell curve stay the same, and there's always 20%, 20%, 50% or something?
I'm going to possibly suggest that there's a sort of “it depends” type of thing here. It depends if your B players are simply not as knowledgeable. Maybe they don't have as much experience or they don't know as much. If the computer comes in and says, “Here's a bunch of people you can call. By the way, we've done some analysis on the likelihood that this is going to be a good prospect and sort through those numbers so that their calls are more focused.” What the computer can't replace is what Chris keeps calling that trust issue because sometimes, it is the guy on the phone, “I know him from this.” The simple reality is a computer isn't going to help with the trust dimension. We can make people more knowledgeable, give them some guidance as far as being a little bit more experimental, and pulling data together to have better predictions like, “Here's the long list of customers,” That helps in that dimension. Getting the people to return the phone call, all of us must get people reaching out.
The number of people simply don't understand that I'm the analytics guy at Optimal Dynamics. They reach out to me with questions. I understand that but I'm the wrong person to call. There's going to be a certain amount of I don't know that person. Chris, I think you run into this all the time to the extent that people are dealing with, “This may not cost a lot, but it's going to take my time, so it's still costing me something.” I'm not going to do that unless it's somebody I know. That looks promising and I trust who it's coming from because let's face it, there's a lot of not reliable information out there. Sometimes, coming from people with some pretty good titles. I don't know how much the computer is going to help with that. This is where you're going to have to find people who have the experience, contacts, and good old-fashioned sales principles still apply.
I suspect it's true in every industry that the phrase “relationships matter” and there's so much truth to that. What I find ironic is the same people that are promoting relationships matter when you ask them to quantify what their teams are working on any given day, a lot of it is non-revenue generating. It's managing a CRM. It's prospecting. It's doing everything that I would argue machines can do better. What does that yield? It enables those people to do exactly what those machines can't and that's built, harness, grow and strengthen relationships. If I can call Chad like we had, we talked for a while. That was a great conversation that no machine can replicate but it's too bad and unfortunate that more sales folks don't have time to make those types of calls that are meaningful because they're too busy trying to do things that the machines could have done much faster and more accurately.
You nailed it. That's exactly why the curve can shift and I think revenue per head's going to go up. We're already seeing it in multiple industries. That's why I go to the Bill Clinton conversation because if 5% had that big of an impact, then imagine what 10% or 20% impact on GDP. We thought the internet was big. Wait until you overlay AI on top of everything that's going on out there. Professor, you wrote a book that's available on Amazon. How would people find that if they want to take the red pill and go deeper on some of this?
I'm working on a book. Yes, I have a book that's on Amazon, but that book is years old. I've got a couple of new books, but they're both available on the internet right now. One will always stay on the internet and the other will be sent off to Wiley and you can buy it in 2021, but you can download it now. If you go to my main website, Castlelab.Princeton.edu, you'll see close to the top “new book” but it's technical. This is Math. This is not for your average readers. As much as I'd like to say I've got something for a broad audience, I don't. This is for the people writing the code.
It’s for the data scientists of the group.
As long as people are still buying from people, AI will never replace you in your job.
If you go to Castlelab.Princeton.edu, there's a link off to the left called Sequential Decision Analytics. This is my pitch for a new field because to be quite honest, machine learning and data scientists, that's an established field. There are hundreds of academic programs, there are books that people read, you can hire people, and they have software, that's an established field. Sequential decision problem is not even an established field. It's a complete mess. There is no one book that you can go after, so I'm hoping that I'm writing the first books for somebody like a data scientist, but the field is sequential where you make a decision such as who to call then you get information such as they returned your call where they said yes or no. You make another decision like, who do I call next? There's a lot of decision information.
We don't have a field that solves those problems but we help humans make decisions. We're pretty good with that. That's what machine learning and most of AI is helping a human make a decision. Optimal Dynamics were in the decision analytics space. We want to be the company that helps the computer make the decisions like, who to dispatch on a loader? How much inventory to order? When you have those decisions where you make the decision, you learn something, and then you make another decision. That's not even a field. You can't go to any school and say, “I want to hire one of your decision scientists who's an expert in decision analytics.”
I'm sitting here going we need a field. We need to have every school build this program. I would say it stands on the shoulders of the data analytics people. It uses data analytics but it's the next step up. The funny thing is I’m writing a book on the choices. A kid's book on making choices. Depending on who you ask, we all have a fundamentally different way of how we make choices. What I'm trying to do is pull together 6 or 8 stories of funny choices that a kid makes that lead to one outcome or another.
I would love to see a draft of that book. I may end up using it in my sales process for adults because I think sometimes that's exactly where we need to start.
That’s the plan. We can take that book and read the kids' book chapters to adults that are like, “You're making decisions in an interesting way. Think of it this way.” We're going to do some gross ones, funny ones, and some that are obviously right versus wrong. There are all those decisions that are in the middle and choices that we make that what's our knee-jerk philosophy on making it. For example, Rich, our CRO, he's thinking about it. He used to be a pastor. He puts that in the filter and his outcome was, “What's the impact on my family, other people, and then me,” because he thinks about others first in his decision process. Other people might say, “What's that impact on my pocketbook, the ordering, and how people make choices?” The goal is to give them a guideline of step 1, 2, then 3. I may need to buy the book and become a data scientist here.
It's never too late. That's a great example of your CFO that’s a pastor. We were talking a little bit and I don't remember exactly the site you sent me. It's a Myers-Briggs type of approach. Again, I look at how sales folks can use AI. Think about personality traits and propensities to make decisions just like you outlined with your CFO. If salespeople understood that going into the sales process, the personas, and the personality types, not based on some unconscious or conscious bias in their own mind, but objective data that says, “This is how Chad thinks.” When you go into that, I know what motivates and incentivizes you in the decision-making process. Instead, CRMs like Salesforce have 87,000 fields for data points, but is it actionable? You're still leading it up to that sales rep to go, “How do I interpret all this?” I'd rather have AI that can instantly create a craft to narrative and a profile that enables me to go right into it and not waste time trying to figure out who Chad is.
Think about the outreach that we do for customers. Email, phone, social, and paid ads, there are all these different channels and influencers, etc. The way I write is the way I write. It was built in me innately. It’s how I am as a person. Great salespeople can say, “Let me look at the profile of the person then I'm going to use bullets or a paragraph for that person.” We tend to go back to our knee-jerk philosophy, “I'm a bullet CEO, so I'm going to use bullets in my paragraphs.” That may not fit with the person who's an audio learner or kinesthetic. Where I think the AI needs to get in this space is, if I send an email to professor Warren, that's very different than the message that's going to go to Chris. I should be able to write it in my own format, then the AI should be able to flip it and say, “We're going to make this one bulleted and this one's going to be more of a formatted paragraph.”
That’s if the computer has that information. How would that even be stored on the computer? How would you even collect that data? That's a great example of something that a knowledgeable human might know and a computer simply wouldn't know.
When we talked with Chris, Codebreaker Technologies can read your LinkedIn profile and it can tell you what your primary characteristic is. It's called the B.A.N.K. Code. Blueprint means you're very methodical in blueprint. Action means shoot first and ask questions later. N is for nurture. My CRO, Rich, is a nurturing type. He ran an org of 500 salespeople before. Knowledge is last. Knowledge in blueprint are similar traits, but different. Imagine tagging a dataset of all of the people that are action first, and then we're selling a certain product to people that we know are the action takers. The levels of standard deviations of grouping a group together have a propensity to make a decision faster versus, as you said, when we picked earlier in this conversation. People cold calling you and you're like, “I don't make that decision.”
This is a case where you could use that classification system to get an initial estimate. Over time, you're going to realize that the initial estimate isn't perfect. There's some uncertainty about that initial estimate. We could bring in decision analytics and that initial estimate says, “I think here's the probability that guy is going to say yes but my uncertainty could be this high.” I'm going to teach your readers something that they can walk away with right now. Anytime you have an estimate and then you have, “It could be this high and it could be this low.” Here's a little piece of advice. Go with the optimistic estimate instead of your point estimate. Take your level of uncertainty, go to the upper tail, and judge who you're going to call based on that. There's a lot of numerical work and theory supporting that policy.
I think that's what you were referring to earlier. It’s what stuck with you the last time we spoke. That is more just elaborated a bit.
In other words, be optimistic. It turns out a lot of academic theories came out with it. In fact, the simple rule that says, “Be optimistic when you try it.” If you're saying, “No, that guy doesn't look very good.” You've got to ask yourself, “How sure am I?” By the way, a very bad trait is people who are wrong but they’re sure. Those are the most dangerous people. I tend to find is often a gender bias there.
The one that comes to mind is I remember there was a rep at one of my prior companies. I won't even name the company. He said, “Ford doesn't need our technology.” I've already met with their Director of IT in one division of Ford. Our CEO of that company at the time got a call in from someone at Ford. He's like, “You shut the door on that because of your incorrect assumption that they aren't buying.” That's because the one guy made a blanket statement that Ford wouldn't need this stuff. He's like, “I guarantee you, they're not going to buy. It all goes back to him.” The CEO was like, “There's value here. Let's talk.” They ended up getting a multimillion-dollar deal there.
This is an issue of any time you do an estimate, you never going to estimate perfectly. It's never going to happen. It tends out, the Googles and the Facebooks are masters of the theory behind this. It's called a Multi-Armed Bandit Problem. It has to do if you have slot machines. Let's pretend that not all slot machines are the same. I know it's a funny story, but that's what the literature is called. This little slot machine has a higher winning probability, but I don't know that. I've pulled the arm a few times and my best estimate is that this looks like a bad machine.
The theory says, “Take your estimate, do confidence intervals, and make your decision based on the upper confidence interval.” If I have a machine over here and I've played it over and over again. The fact is I end sore it. This is how good it is and that's what I'm going to get, but over here, I haven't tried this very often. Being honest about your uncertainty is a big deal and that's a walk away. Your audience can take that one piece of advice without any computer and apply it. You've got to understand how sure you were. If you've got a marketplace, you have a lot of experience with, and over here is a marketplace that you're new to, acknowledge your uncertainty and then look at how good it might be and make your decision based on that.
I'm playing in a game right now that has a multimillion-dollar upside. It's uncertain for me. It's a game I haven't played before. I'll know within a year if the game plays out properly. The downside risk of playing in that game is minimal so it's going to be a fun game to play in 2021. I'll tell you more when we regroup in early 2022. This has been a fun conversation. If there is one piece of advice that you guys would leave with our audience around AI, leave us with a taste of where do you think the future of all this is headed. What's going to happen in the world in the following years?
There is no such thing as perfect technology.
AI now is machine learning. It's taking data and predicting things. The next wave is decision analytics. Predicting something is one technology. Recommending something and saying, “Here's what you should do.” That's a big step up. A lot of people confuse the two because sometimes, if I predict that this is a good product, I'm done. The decision is made, but there's a lot of problems where that's not done. Logistics is a great one, but even sales. Understanding that you do have to try things. Doing the decision analytics is the highest step or at least the highest steps where I think computers will be valuable. There are some things out there that I don't think the computer's ever going to be useful for. Computers aren't creative. The decision analytics from predictive analytics machine learning to decision analytics, I think that's the big step up. We're having a lot of fun at Optimal Dynamics and I think that's where the world's going.
It's interesting on that point. I talked to the CEO of a company who monitors SEO, websites, and content. They've taken it to this tool called First Draft where they can creatively write. The argument around what is creative or not because it ranks high in SEO words and things, his point maps to what you said, which is I still need a human in the loop to be able to look at the first draft, not the final draft. By the way, the computer didn't come up with the topic to write on necessarily. There's still a human in the loop even in pretty advanced, sophisticated SEO-related algorithms. Chris, what are your final thought on the future of AI?
I would echo both comments. We’re not obviously from a bias or somewhat biased because that entails exactly the solution that we're providing. I don't see that changing. It’s a sworn stated this idea of moving beyond all this data collection aggregation into what are we doing with it? In our view, having humans still trying to make very complicated decisions may not be the most optimal means to go about it, but I'm with you. I don't ever see a world, at least not what I want to live in, where we've allowed the machines to go on complete autopilot. That's scary. It scares me. I would also leave the audience with this idea of never stop learning.
Keep that curiosity going. We can all agree that technology is not going anywhere. If you ask Elon Musk of the world, his thought is AI is here to stay so we better learn to get along with machines versus trying to go against the flow of trying to incorporate them into our lives. There's been a lot of empty promises around AI. The industry we play in is certainly no exception. It's all about this idea of how do we best harmonize machines, AI, and machine learning, with understanding and human capital will always remain imperative no matter what you're working on.
When you guys get outside of the logistic space or if there's interest there based on phone-ready leads, email-ready leads, and social-ready leads, it'd be interesting to partner with the company. My company does millions of transactions, but there are companies that do hundreds of millions. It'd be interesting to overlay the decision-making analytics on top of a system that says, “Do I email? Do I call?” Keep it simple for the salesperson. “Who am I going to reach out to and which channel?” That would be an interesting problem to solve. I think 1 or 2 other companies tried to solve it haven't solved it yet.
It's a tough problem Chad. Keep us posted with the children's literature you're putting out there. I'll be honest, we’re not a test because we were dealing with a very complex topic. We're not selling widgets. We had actual customers or prospects at the time that gave us feedback saying, “Could you distill this down a bit?” The furthest form of distillation would probably be, how do you explain this to a four-year-old? I know that sounds comical, but there's so much truth when you're talking about advanced concepts and topics like AI. I hope we can leverage you.
That sounds like a plan to me. I'll get you the book where I've hired a person to help me write it. He wrote a book when he was 12 for his 5-year-old brother and he became an expert at mapping the right emotion that a four-year-old would want. A lot of times, we, as adults, tend to want to preach. It can't be like that. The finished product here is going to be interesting.
I'd love to see that book. Something that I do every time I'm looking for opportunities for decision analytics is I ask how easy it can be quantified. We all know that there's a lot of decisions humans make that can't be quantified. Leave those up to humans. Like sales, it’s something that can be quantified. You can tell whether the lead pays it off. Logistics is full of nice quantifiable decisions, so as health and finance. You want to have those decision problems where you have a clear, “Yes, it worked and here's how well it worked.”
If I'm going to send an email, call, or connect on LinkedIn, it wouldn't be hard. I'd call you 5 to 10 times. I'd email you 5 to 10 times. I already know that one is going to be very low-performing, so the person who loves to get emails and replies. That will be 4% of the whole audience. LinkedIn is probably going to be pretty high, but to know which people on a dataset, it's highly quantifiable. I can get you a 50% reply rate on your emails versus a normal 1%. That'd be magic. You guys were onto something. I've really enjoyed the conversation now and the last time we spoke. Let's stay in touch, I'll definitely get you the outline of the draft when I do it. If people want to get ahold of you guys, what's the best way to reach you out?
For me, I'm Chris Torrence. You can ping me on LinkedIn. CTorrence@OptimalDynamics.com is my work email address. I’m happy to chat with anybody in the audience that wants to learn more about me or the company.
Thanks for tuning in. We'll catch you on the next AI for Sales show. For now, Chris, Professor Warren, good to see you.
Thanks so much, Chad.
Likewise, Chad. Thank you.
- Optimal Dynamics
- Ryan Reisert - LinkedIn
- Sequential Decision Analytics
- B.A.N.K. Code
- Chris Torrence - LinkedIn
- LinkedIn - Warren Powell
About Chris Torrence
Integrity first, service before self and excellence in all we do. These are the core values instilled in me while proudly serving in the U.S. Air Force. These are life principles I relentlessly strive to achieve both in and out of the office, each & every day.
I've spent my entire career within the dynamic logistics, freight transportation and supply chain solutions arena, which I enjoy immensely. I am a forward-thinking, tech-embracing strategist at heart, yet I still firmly believe people are the most valuable resource of any organization, period. No matter how profound the technology, there is no algorithm more powerful than the human spirit or more creative than the human mind.
I strive to bring out the very best in people and I feel there exists no greater satisfaction than helping others succeed. I practice empathy and prefer to focus on individuals holistically, not just one's quantitative outputs. I consider constructive feedback a gift. I believe in always doing the right thing - no exceptions, no excuses. In my world, the C-suite and custodial crew both warrant the same level of respect. I'm also an unwavering believer in servant leadership and always delineate between true leaders and authoritarians. We truly are stronger together and every voice on the team deserves to be heard.
I'm looking to connect with others of like-mindedness as well as the most persuasive devil's advocates, to everyone else in between. I wholeheartedly welcome thought-provoking conversations and classy debates.
My unique skill set within operations, people and process improvement and cross-vertical business development has equipped me to objectively evaluate, collaboratively diagnose and ultimately overcome even the most complex of challenges. I excel in providing creative solutions while driving sustained results. And I always aim to have a little fun along the way!
About Warren Powell
Warren Powell is a faculty member in the Department of Operations Research and Financial Engineering at Princeton University where he has taught since 1981. In 1990, he founded CASTLE Laboratory which spans research in computational stochastic optimization with applications initially in transportation and logistics. In 2011, he founded the Princeton laboratory for ENergy Systems Analysis (PENSA) to tackle the rich array of problems in energy systems analysis. In 2013, this morphed into “CASTLE Labs,” focusing on computational stochastic optimization and learning.
In the 1980’s, he designed and wrote SYSNET, an interactive optimization model for load planning at Yellow Freight System, where it is still in use after 25 years. In 1988, he founded the Princeton Transportation Consulting Group which marketed the model as SuperSPIN, which was adopted by the entire less-than-truckload industry, stabilizing an industry where 80 percent of the companies went bankupt in the first post-deregulation decade. SuperSPIN was used in the planning of American Freightways (which became FedEx Freight), Roadway Package System (which became FedEx Ground), and Overnight Transportation (which became UPS Freight). SuperSPIN stabilized the LTL trucking industry in the 1990’s, following its deregulation in 1980.
Also in the 1980’s he developed a series of models for truckload trucking, starting with LoadMAP (written by Ken Nickerson ’84), which then evolved to an integrated stochastic model for driver assignment called MicroMAP (the senior thesis of David Cape ’87). As of 2011, MicroMAP was being used to dispatch over 66,000 drivers for 20 of the largest truckload carriers in the U.S.
He has started three consulting firms: Princeton Transportation Consulting Group (1988), Transport Dynamics (1995), and Optimal Dynamics (2016) (CEO is his son Daniel Powell), but he has continued to do his developmental work through CASTLE Laboratory at Princeton University, where he has worked with the leading companies in less-than-truckload trucking (Yellow Freight System/YRC), parcel shipping (United Parcel Service), truckload trucking (Schneider National), rail (primarily Norfolk Southern Railway), air (Netjets and Embraer), as well as the Air Mobility Command. As he moved into energy, he has worked with PJM Interconnections (the grid operator for the mid-Atlantic states), and PSE&G (the utility that serves 75 percent of New Jersey).