AI Revolutionizing Freight | Episode 304

Speaker 1: 0:19

Welcome back everybody. It's another episode of the Freight360 podcast. We got a good special one here today for all of you tech junkies, or AI junkies, like myself and Ben. We're going to talk some AI and some technological advances. On today's episode, we're joined with Tilo from Levity. We've had him on the show in the past a few times, so, tilo, welcome back. Man. How's, uh, how are things? You're in europe right now, right, yeah, yeah, I'm in berlin, germany, at the moment very nice, very nice, nice, uh, summer in germany.

Speaker 1: 0:52

I've actually never been to germany's uh, but we're gonna get into what's new with ai, what's new with levity. You know how the landscape of um, the application of ai and transportation has evolved and some of the challenges. But first, for everyone out there, if you're brand new, you've got hundreds literally over 300 other episodes to check out, as well as over 100 Q&A sessions where we answer your guys' questions in a shorter-form podcast. We've got educational videos, downloadable content, blogs, the Freight Broker Basics course if you want a full educational option for you or your team, all on our website, freight360.net. A lot of content on YouTube. Share us, leave comments. You know, reviews all that good stuff Ben real quick in the sports sector. Did you watch?

Speaker 1: 1:41

Tony Scheffler, I was going to say did you watch?

Speaker 3: 1:43

Absolutely.

Speaker 1: 1:43

I didn't watch it, but was there like what's the 20 second highlight?

Speaker 3: 1:48

He kind of ran away with it. At some point he was up by like eight strokes between number one and number two, first and second place. I think the closest it got on Sunday was maybe six strokes. He's just super solid, really steady plays, really steady plays. Every hole, exactly the same. I mean not really a lot of ups or downs. There are a couple miss hits that I saw on Saturday where he looked like he was getting a little frustrated, but literally he had it pulled back together by the next shot, like he's just very consistent, very solid. Other highlight is I think it is 1197 days between his first major and his fourth, and that is the exact same amount of days between Tiger's first major and his fourth. So literally to the day they have the exact same amount of majors. I think Tiger might've started a little earlier career wise. I think he might've been a little younger when he won his first.

Speaker 1: 2:38

But still super impressive. I think Shepard's in his twenties still, though, isn't he like 28? Oh yeah, he's still young, yeah.

Speaker 3: 2:43

Love, I think, schaefer's in his 20s still, though, isn't he like 28? Oh yeah, he's still young. Yeah, love, watching him play. I was also. You know, I'm a huge fan of.

Speaker 1: 2:54

Brooks Koepka Loved watching him miss the cut. He wasn't even there anywhere to be seen.

Speaker 3: 2:59

Doesn't seem to be able to show up to a major anymore.

Speaker 1: 3:03

Aka the guy it was. Was it Ireland or?

Speaker 3: 3:06

Yeah, it was in Ireland so everyone was pulling for Rory. Obviously Rural Portrush, I think I can't, I don't, I've never been there. So like conceptually, the course has kind of seemed the same to me. It wasn't the old course. I think I'd have to look it up. Exact, and I kind of should know that I watched it all weekend but my mind's blanking, oh good.

Speaker 1: 3:23

Anyway, let's, we'll shift to news, so, and we'll just get right into the AI topic here, because, ben, I know you've been big on some of the recent releases and we're going to segue this right into our conversation with Tilo.

Speaker 3: 3:35

We can segue right in and I was sending this to you yesterday, nate, and I use this on a project that Tilo and I were discussing yesterday, and I use this on a project that Tilo and I were discussing yesterday, so it's like a really good segue.

Speaker 1: 3:50

Chatgpt just released the agent model in ChatGPT, not to be confused with the agent model of freight brokering.

Speaker 3: 3:53

Correct. So they've had deep research, which you know, to my understanding, it definitely takes longer and I believe the model like goes and checks its work against itself and will like, step by step, to try to find the hallucinations, come up with better answers. And I've listened to interviews with people in like the like finance world that are doing like case studies with like MBA students for companies and research and they're like these are. Some of them are as good as what we're paying to have done with consulting companies. So, like anecdotally, I've heard really good things about deep research. Then the agent model just got released like two days ago, and then it got released to all of the accounts, like pro, premium and plus. So I got it yesterday, watched a YouTube video on somebody using it and like to explain what it does is basically is a little window that'll pop up, and like to explain what it does is basically is a little window that'll pop up. You give it a prompt and it will go out into the web and just search different websites to pull information back, check it against itself, to come back and give you not just what's based on their training data, which is theoretically everything on the internet, but like real time going to places that you can prompt it to, and, like I was able to do some cool stuff. Like I saw someone do a demo where they're like, hey, find me, all of the orthodontists in Austin, that's who I want to market to. Then I'd like you to compare all of their publicly available email addresses to a metaphor, maybe to cross-reference them. Then it said right after that write in a very effective cold email to each of these people, make it eerily stockish, as if I know what is going on in every one of their social media platforms. It pulled up all the dentists, all their email addresses, created an Excel sheet and then wrote an email for everyone. And it was like hey, I noticed that you went to school at Harvard and then you did your dental training at University of Penn. Noticed, you're into kayaking. Hey, you're an overachiever like me. And then it correlated everything it knew about the guy using that model to write a very descriptive, very good hook to each of the people it found and their email addresses of which he just plugged in and sent emails to them. So those are some really cool things I saw like literally two days ago. I used it yesterday and it still hallucinates. It still makes errors, because the stuff that I did I could find errors in both deep research and the agent model and the thing that Tilo and I were working on and this is what I'll segue with because I want Tilo's thoughts is we were trying to figure out an integration we're doing for AI between our TMS, our tracking system, and our teams, and we wanted to be able to pull information from our TMS for updates.

Speaker 3: 6:36

But our TMS doesn't get all of the information from the tracking system. From the tracking system, their user interface, like when you go to their, their login, you can see more things like this driver's behind. This driver can't make it. This driver is on schedule. Our TMS just shows where he's at and when he checks in. So we wanted to pull all that and then we want to feed that into like we use G suite to communicate with our all of our you know, carrier, sales reps, track and trace and basically we wanted to have 100% visibility into if this driver is not going to make the pickup or delivery. We want to be able to send a message to our team and just say hey, on macro point, for instance, shows this driver's behind. You might want to take a look at it, for example or hey, this guy's tracking has turned off, maybe you want to give him a call. Then we want to either layer in some AI phone calls to maybe just touch base with the driver, see if it can resolve it, or kick it over to a chat. So it was like how can we get 100% visibility into the problems before they occur?

Speaker 3: 7:36

And then Tilo and I had to look into the integrations and basically I think for anyone out there it's like some information gets pushed from one place to another, some gets pulled I'll let Tilo explain how and why that's the case, whether it's an API or a webhook and then some of them basically the API documentation is just a big sheet that says hey, if you connect your computer and your software to our software, we can send this to you from these fields, and this is how the information comes back and forth. So after we outlined how we wanted to do all this, we had a bunch of questions after we read it and went okay, what information gets pushed, what gets pulled, how often, how frequently and how can we get it to where we need it and what needs connected? So I went to start doing what I normally do was read all of these things to see if I could figure it out, send emails and ask questions, and I was like, oh, this tool just came out, let's see how good it is with this. So I put in every single question we had everything. I wrote all my notes, all my questions into deep research and said see what you can do to find a better understanding of how we should do this, what is the workflow, and also, please write me a prompt for your AI agent.

Speaker 3: 8:42

I let deep research run, then put it into the agent, because I knew the agent could go directly to their websites live, find any information that wasn't in the training data that might have been changed, and then I asked it to do the same thing again.

Speaker 3: 8:54

Then it gave me a breakdown and it answered like I think, almost every question we had. And when I read it and checked it back, I'm pretty sure I could see some of the things that weren't correct that I've manually gone and changed for like our document for you and I to work. But I'm like it looked like it got like 95% of some of the harder questions that we couldn't get. And the last thing I wanna say is I emailed some of these questions to both MacroPoint and our TMS. The answers I don't think they gave me from my TMS are actually correct. I think the AI agent got us better information than the tech support team at the TMS. We'll see as we go through the process, but I'm pretty sure it gave me more accurate information than the human being that works at that company.

Speaker 3: 9:34

Using AI to enhance AI hey write better prompts for yourself than I can write this outline kind of blows me away as far as like. I did that on 45 minutes. That would have taken me like a whole day.

Speaker 1: 9:47

I want to throw Tilo. Can I get your perspective, like your side of all this, so from you know, from levity, or even, if you want to go broader, like from the from the technology side of it? How does like? I guess, where are we at in the landscape of like the implementation, implementation of AI, and all of this for the logistics industry?

Speaker 2: 10:59

Yeah, I mean, changebt is obviously a general purpose tool, right for any kind of request that you might have. And the question really is, what tools do you use for which you know task, right? So for these kinds of research things where it's like, hey, find me all of the dentists in you know Minnesota and you know, do research on them and put that in a spreadsheet, or, you know, create a presentation for this topic, or these kinds of you know research-heavy things where you have to visit a lot of pages, accumulate a lot of information. But it's also there's not that requirement, right, that you have to be a hundred percent accurate or you have to have certain information in there for this to be successful, right, because you can add the missing information and you can remove the incorrect or the unnecessary information. So there is not that requirement to be super, super accurate.

Speaker 2: 12:01

But when it comes to the kinds of tasks that we're automating for our customers, we have to be almost perfect, Because if we automatically create an order in a TMS and then we mess up something there, right, then worst case, the truck is showing up at the wrong location and these kinds of things are way more. You know need a way higher. You know, quality need to meet a higher quality bar than something like that, right? So it's more of these. You know I don't want to do this right. I don't want to click on 40 different pages and read them all, you know, combine them into a document or something like that and just automating that perfect, right.

Speaker 2: 12:47

For other cases there are other tools and you just have to get accustomed to that. I think the best advice that I can give anyone is just to keep working with that stuff. See what works, what doesn't. You know. I mean you could ask the agent hey, find me, you can make it higher level, right, because find me new customers. Or, you know, sell freight, see what happens, right, we're not there yet, I'm sure, but to kind of see where the limits are, what it does and these kinds of things.

Speaker 3: 13:17

So to that point I want to say something that happened yesterday that I thought is kind of funny but not, is it like? I talked to an executive that you both know of a TMS and he was talking about how they're having so many issues. To what you're saying, the bar in our industry is 100% accurate. If you're wrong, it costs money or something really bad happens. Send a truck somewhere, paid attention, whatever, right, there's lots of repercussions. So getting it to that bar, he's like, is incredibly difficult. And he was telling me some of the things that he read about uses in the medical field of where like, yeah, it will catch lots of things, but it also catches lots of things that aren't wrong, meaning like there is also a risk at saying somebody has cancer that doesn't and you give them chemo because it's just wrong on that side too, right.

Speaker 3: 14:02

And the funny thing that happened yesterday I guess it's not really funny there was a little kid that I think had pink eye at the place that my daughter, my wife, went and my wife was like, oh, like you know, is your kid okay, like what a normal parent would say.

Speaker 3: 14:15

Right, I'm gonna say like, hey, like we have a doctor we use locally if you need somebody to be happy to give you their number. And she says, oh no, he's had it for about four weeks and chat gpt said it should go away in a month or two. And I put some like oh my god, remedy on it and she came back like she was so visibly upset like it upsets me still even thinking about it of like I'm like I just it was literally when I got done doing what we were just talking about that she came in and told me this and I'm like Jesus, I'm like honey, like I just found all these errors and things that like I know because I'm doing what Tilo said, of like going back and checking to see where it's wrong and where it's right, but yet you have people just going. Well, the computer told me, I guess I'll just put my child's welfare in what ChachiBT said, and I'm like this is terrifying to some degree. Wow.

Speaker 2: 15:04

Oh my gosh.

Speaker 2: 15:04

Yeah, yeah, I mean just a thought on that right, like it really depends on the task and whether you are able to loop in human oversight or not. Right, because in some cases, getting the human oversight and checking every single case, I mean you're not winning anything there, right, like the AI is doing something and then you know you're still doing everything manually, so you're not winning any time there. There are other cases, though, where you know. What we've seen, for example, is we have lots of customers that get a ton of emails from their customers about damages and you know saying packages, right, and then they get those to a central inbox, but they need to forward it to the right person, and that data needs to be sort of prepared for that person to then work through the case. Right, so they're still doing the same work, but the prep work of taking all of the available information and making it very easily digestible is done by AI, and you're saving I don't know 15 minutes for every case.

Speaker 2: 16:09

And then the other case that I was thinking about right, like the cancer case. It's not bad if you, I mean it's a little bit bad, right, if you tell about, right, like the cancer case, it's not bad if you. I mean, it's a little bit bad, right, if you tell someone, oh, you might have cancer, the AI detected something. And then you double check and you're like, oh, it's actually not cancer, right, so okay. But the reverse is very bad Telling someone oh the AI said there's no cancer, but there is actually cancer.

Speaker 2: 16:35

That's way, way, way worse. So that's why these kinds of companies that are also doing cancer screenings with AI, they're optimizing for that right and they're optimizing for false positives to be higher than the false negatives. And you can do the same in, for if you, if you use ai to to process invoices, you want to, you want to check every single invoice is it correct? Does it match what was agreed right? And then in the past you might have to check every single one for correctness.

Speaker 2: 17:07

And then you have a thousand invoices and nowadays what you would do is you pre-scan it with AI and AI sorts it by the likelihood that it has some errors and highlights those errors. And then you just go one, two, three, four, five, and at some point you're like, okay, there are no errors anymore and I can just, you know, approve the rest. And then maybe you only need to check a hundred right, because maybe there are in 1,000 invoices you have 10 that are, you know, faulty right and have something in there. But in the previous case you would have to check all 1,000 to get the 10, right, and now we only have to check 100 to get the 10, right.

Speaker 3: 17:46

In that use case. We've talked about this. Like I'm doing lead research for the salespeople right, when I can do that using the tool you showed me, which was clay that I started using, or it's like basically step-by-step AI that goes in and then checks the last column so I can look for errors and say, like 5,000 leads of like people to a company, to a lane, that all match up with who we want to talk to, but then I can find those errors Like I'm literally going to do it this afternoon Like I have a list of 5,000 people, their email addresses, their company and what lanes we think would be a fit with that they ship, and 240 of them came back incorrect. So now, instead of me manually checking 5,000 leads, I'm going to cross-reference that 240 to my original set and it's probably going to take me an hour where that would have taken a human being. A week before we email, reach out and pay human beings to do the actual work of building relationships with those people.

Speaker 1: 18:41

Let me ask you this um, how are you? And like because I really want for the audience to to be able to understand this on a on the most basic level and I'm getting at here is that this whole conversation, like you know, the what AI can do and some of the risks, like I always think of it as like, give me time, as the expert, to be able to make decisions that help me increase revenue and profit and take the tasks off my plate that stopped me from doing that Right.

Speaker 3: 19:25

So take me through how you're implementing it.

Speaker 3: 19:28

We can go through track and trace, but I think we can talk through one that we haven't done yet that I want to do. That I think is really valuable. So the other person that I work with, owner of TLX, that we manage the company together and call it 25, $30 million company with like 25, 30 people give or take right Now the questions we are trying to figure out that we were talking about and spending a lot of time on looking at our TMSs which of our customers have increased in business, which have decreased. Some TMSs will tell you that when you kind of look at reports. But then we wanted some more better data, meaning like okay, well, like how many orders, offers for loads are coming in versus how many we're actually being awarded. Because one question is how many loads do we move with this company? And the next question is is that company shipping more or less right now, like are they sending more quotes that we're missing your batting average, right? And it's exactly the analogy. We wanted to know what our batting average is how many pitches were thrown, not just how many we hit, right? That gives you a percentage which will help you gauge those.

Speaker 3: 20:30

Then we wanted to go a little deeper, because we know that days of the week pricing changes a lot on some lanes. Like we'll do well on a Monday on this lane and we'll lose money Tuesday for the same rate on that lane and then do well Wednesday and Thursday and maybe not Friday. And we know that information is in the TMS but we can't really get it out and sort it and like I'll spend hours in Excel trying to figure out are our highest margins on a Tuesday or a Monday? Are they always on a Tuesday? Are there certain weeks in a month where that change? And then we wanted to see okay, out of all of that information, what we're really trying to figure out is our sales reps. Who do we need them to reach out to more? Who are they not talking to enough and are they quoting too aggressively, like too low, on certain days of the week? Do they need to quote a little higher on certain days of the week? Is that based on a region of the country or is that just there's a lot of trucks in the Atlanta market, for example, on a Monday and none on Tuesday? So we either need to price higher, but if we can see that our sales reps can talk to that shipper and say listen, hey, it looks like the market is changing a lot on Tuesdays. Would it be really difficult for you guys to load a little more on Monday and Wednesday? We can keep your costs down. There's more trucks, you'll get better rates. The trucks need the loads anyway. It works better for the carriers, better for the customer.

Speaker 3: 21:48

Everybody saves money by finding efficiencies that are there that you can't really see this information at TMS and in my head I'm like OK, I really want to be able to load all our TMS data into, let's say, deep research in a project in GPT and ask it these questions which customers are going up or down, when, why, what days of the week? Who is doing more, who is doing less? Now here's where I ran into what AI couldn't do is. It's not really that great at checking its own math, so it would constantly give me different answers and I spent an afternoon trying to get it to work that way. What Tilo and I talked about is maybe the data needs structured differently, like it needs to either go into a database in a way that GPT can go and do this, but the other answer and we didn't really talk about this is there are workflow steps where basically you can have GPT do one step, then it goes to another place, maybe a database or a CSV file like an Excel sheet, then does the next step, then does the next step. So it's like a string of basically a calculator, a computer, a calculator, a spreadsheet, then a computer, because ideally you don't want to spend four hours to get the answer to six questions.

Speaker 3: 22:58

You want to just be able to ask the question and say, hey, where do my reps need to spend more time on what customers and why? And if it can give you that easier, you spend way less time talking to people that don't need help, more of your time on the people that do need help. And then you can do the same thing with your carrier base which of our carriers are most likely to need this lane, on which days of the week, which lanes do they need? So when our carrier reps call the carriers, we have a better understanding of what they've been doing with us, what they need, and spend less time calling carriers like, yeah, I don't run that lane anymore. Yeah, I don't need that, we don't even run that anymore. That's been three months. We're over on this side of the country, but yet we know there are carriers that you want to spend time with.

Speaker 3: 23:37

The biggest waste to what Tilo said is doing this manually, which all I mean, nate and I have done is like you're calling 120 trucking companies a day, you find two or three where you have a good conversation. The rest was just pitches and misses, just throwing the ball across the plate and you're just watching it go past and you have nothing actually of value being created. So that's like one of those, I think, really valuable use cases for any brokerage, and I don't know any TMS that really does that well and I'm like that is incredibly valuable to give insight to the executives, the managers or the brokers and carrier reps where to spend your time and who you should be speaking with to create more value for everyone in the chain carrier brokerage all the way up to the shipper.

Speaker 1: 24:20

My initial thought on a lot of that is like AI will like the, when implemented in its most effective sense, will like help leadership identify their bad eggs sooner and recognize their studs sooner. To be able to like, because if I know it's, you know, it's kind of like, if you're raising, I'll just, you know, make up racehorses Right, like, and you figure out like, hey, we're going to stop wasting our time and money on this one. It's not going to win us any races down the road, but this one, hey, in two years from now, is going to be, you know, a rock star.

Speaker 1: 24:54

And I think you could probably start to identify your reps that way and who's who's best to serve in certain roles.

Speaker 3: 24:59

There was an interview with I can't remember who it was which one of the founding members of Google that is back at Google working on these things he's not the CEO, but ah, my mind's blanking. He ran all of their communication data through an LLM and he started playing with it and said which team members in this group of our company are underappreciated but probably need promoted. It identified like this one woman, so he then went by, like reading their emails and stuff or what.

Speaker 3: 25:27

Yeah, like I guess all their internal, their Slack messages, probably like their communication of employees talking to each other and it identified somebody and he then did the manual work that we're talking about of went and talked to her manager and said, hey, how's this person performing? What are they doing? What are your thoughts on like upward mobility? And he said it's interesting that you pointed her out because I've noticed like she's really effective, she probably has the personalities to be promoted and like I've been thinking about possibly doing that and I just haven't really brought it up. And he's like you know it's anecdotal, it was one instance he's like but it accurately did help me identify one of our employees across tens of thousands, that we should maybe spend some time and do a little more whatever you know HR work to see if they're fit to be able to be promoted.

Speaker 3: 26:16

And it was like exactly what you said, nate, like it identified who had the potential and who to go spend some more time with. I mean, there's always unintended consequences, I think, to even things like that, but like that was, I think, a really good recent use case. I heard of exactly that thing.

Speaker 2: 26:32

Interesting, probably a little bit off the rails from the initial discussion, but if I now think about it, right, especially in large, large corporations, it's super difficult for managers to promote the right people right, because they're busy and often these decisions are made based on you know who? I know right, who I like, and a lot of people that are optimizing for that. Right, they're just being very visible, right. They're, you know, talking to the right people, they're saying the right things, but, you know, sometimes that's representative of whether they're actually doing good work or not, and sometimes, often it's not. And you know, I think you can probably find some more diamonds in the rough in that way.

Speaker 2: 27:23

But it reminds me of what we're doing with our control tower product, which we launched a couple months back, where a lot of the customers that we now have initially approached us and said, yeah, we want to do something, but there are so many things we could do and they're all.

Speaker 2: 27:40

They all sound cool, but we don't know where to start. Right, we don't know how to assess, we don't know how to assess the potential and the feasibility of these use cases. We realized that we were doing the same thing with all of them over and over again where we said okay, please get me an export of your last, you know, three months of email data or communications data, and we analyze that and we find the use cases for you and we also calculate how much potential there is right. Are you getting more track and trace related emails or more quoting related emails from which customers? How much are you winning? All these kinds of things? Nobody knows what's going on because it's a black box, and I remember that we did something along those lines also for you, right, where you gave us an export of all of your emails.

Speaker 2: 28:32

I think it was like I don't know, a hundred thousand or something, right.

Speaker 3: 28:35

And then we got you exactly the data out that you needed to work with, right and shifting through all this stuff, it's very similar to you know what the agent and deep research stuff is doing on websites, but with your internal data and getting you that stuff out, so and I want to go a little further, because that is a thing like, as they're more remote employees, it is incredibly hard to see the things that I think I was used to see in the first 20 years of my professional life Right, which is like you're in an office, you see who's vocal, you see who vibes well with other people, who, at the water cooler, seems to be having the personality that is the fit for their role, and you see the people that are kind of heads down, that are really fits for their role. Like I was an analyst and a salesperson, so like I like we used to joke we're like, hey, we're the squirrels in the closet when you're the analyst, but like there's a fit and there's a reason that some folks like move towards that role, and when you're not with people like you don't have so much information you see being present with them and like we've talked about doing similar things of like we just want to know, like who's doing well in the role, who maybe needs some help, where do we need to focus our training? Because like what we were doing is like we're just spending the same amount of time with everybody and like that's not efficient and like some people don't need as much time and don't appreciate it. Some you're not spending more with that need it and there's no way to identify that unless you can see and analyze all of the communication. Who's sending the most emails? Who needs work to your point, nate, like on formatting grammar and where, and who they're sending them to right? Who needs help on communication skills with their customer service? Who is doing well by building good relationships with carriers? Who's just treating them transactionally and needs coached up a little on that?

Speaker 3: 30:19

And it is a giant opaque black box. And if you ask somebody where do you need help, the irony is that's an unknown, unknown to that that person. You know that person doesn't see their blind spot because by definition it is their blind spot, so they can't tell you where they need help In most cases. You have to identify it. That's the job of a good manager, or I would say leader is to know where to focus people's, your energy to help increase people's. I don't want to say deficiencies, I would say blind spots or areas where they would benefit the most from coaching.

Speaker 3: 30:51

Areas for improvement, right so.

Speaker 1: 30:52

Teal, I got a question for you. One of the things at my brokerage that we're doing right now is we're evaluating a new TMS to move to, and there's a lot of AI powered stuff and, regardless of AI, like people seem to be resistant to change when it comes to like anything new. So I'm curious when it, when it's kind of implementing levity with some of your customers or prospective customers what are like how is the adoption, ben? Are people embracing it? Are you getting a mix of like oh, this is overwhelming, what does that look like? Because I mean, mean you're, you're. The end goal is like is great, right, but getting there is there's probably, like you know, some hurdles or, like you know, a resistance to adoption. I'm kind of curious what? What does that look like? And then you know, once you get over, how is it? How's it all pan out?

Speaker 2: 31:45

Yeah, I mean, in our case, you always have, I would say, three groups of people that you need to get on board, right? One is the managers of a team, right, that just want their team to excel and become more efficient. Then you have one level above. You need to convince obviously, c-level that this makes sense. You know, economically right, there is strong ROI. And then, even more importantly, you have the people that need to adopt it, because if they don't adopt it, the whole thing falls apart, right? So the team of these managers need to be, you know, first of all, educated, right, because a lot of the people have they probably know ChatGPT and they've used it for some stuff, but there's so much more to AI and how it's deployed in different kinds of use cases. So maybe you know, just giving an example from one of our use cases in quoting right, where emails are coming in people asking for FTL rates from one zip to another, and you know, know, we're plugged into tools like green screens or sonar or you know that, to to kind of get, uh, these rates and then assist them with rating faster, right, especially when it's these mini rfps of, like you know, 20 different lanes and they would have to go to green screens for every single one right to kind of understand where the rates are and then apply additional markup logic and all these kinds of things. So they know that it's painful and they know that it's like this task that they don't want to do, but they also don't want to lose control over what's going on. So we're usually coming in where we say, okay, we're not gonna take this away from you, but we're gonna automate these steps. And what's happening for you is that you only see the email when we've done something with it. So there's already a draft reply that you can review and you can check the rates and you can see. Okay, you know I might want to adjust this a little bit, right, but I know it's taken away a couple minutes of really dreading work.

Speaker 2: 33:52

And then over time people start to trust it more and more and more. And you know, we've seen it with a large client recently where in week two the users already started sending 85% of those drafts that we created for them without any adjustments. They just said, okay, looks good, send, send, send. Right, and at some point we might send certain emails right away, but we're not doing it yet, right?

Speaker 2: 34:17

The ROI on this is also pretty strong, and the other thing that's helping them is that they don't have to learn another software, another tool, another tab that they have to use in certain you know context. It's just their email, right, it's happening right. There we're more like a background tool and that makes adoption much easier. The other challenge is more in the how do you set everything up so that you get to this point right, and that's more working with the manager and the people that have the operational you know, process knowledge, and that's what takes really a long time right. So sometimes you spend several weeks to set everything up, account for every edge case, and then you present it to users so they're not you know like oh, this is a wrong rate, this is bad. They don't know how it gets to that right. So you need to be also careful to when do you expose it to a larger audience?

Speaker 3: 35:09

in that sense, here's one on that use case, right, that is close to one that we use, that we want to do too, right? Say you have a customer right that you've worked with often, right, and they're sending over a request for a quote, like on the same shipment types every week and that person is manually doing what Like exactly what we're talking about. In the final mile they go and look at their company's TMS history. What have we been paying for that as a company? Then they're going to go to DAT and go what is the average, what's the high, what's the low? And also like what is the average posted rate on the load board right now? Okay, so you got four pieces of information. Say they got five lanes, they've got to manually go in each of those, write that email and then make that decision.

Speaker 3: 35:55

What we found is like we found some plugins to try this Tilo of like, basically email plugins that will auto-populate the response and it'll say like so I don't know, whatever Acme shipment says hey, these five lanes. And says, hey, can you quote these, ben, these go out tomorrow and that happens every Tuesday or Wednesday, right, the email auto populates and goes hey, Frank, based on the way I responded, it literally writes that email, very similar to every other email I sent him every week just pops up and then it just drops in. It says right next to that lane here's your rate, here's DAT high, low, average and average posted. So now all the person has to do is look and go okay, dat, average is right, what we paid last week and the highs around where it was boom. Delete this, leave that one. Then go to the next lane and go oh, I want to add a hundred dollars to the rate because it's gone up a little bit from last week. This one went down. Subtract 100. Now all they've got to do is read it to make sure it looks good, pick the numbers that are already right there in their email, instead of going to three other websites and then hit send right. Like that saves that person like 15 minutes worth of time to do the exact same thing with the same information, just by presenting it in a more efficient and effective manner.

Speaker 3: 37:12

Right, and to me, like it's we tried that with a similar way. Like I kind of manually hacked it together and like the rep that used it for like a little bit, he's like dude, this saved me so much time, like, and he was quoting so many lanes a day. He's like dude. If we could do this at scale like this would make everything so much more effective. And if you could layer that on the thing I said earlier of like what days of the week do you pay more and what you don't, and now it just goes hey, it's Tuesday. You've been paying more every Tuesday all month. You should be around here.

Speaker 3: 38:58

You're helping that person get better at their job too. You're not actually dumbing them down. You're helping them with more feedback, right where they see it. Because the big drawback I see with some of these is like, literally, people that are using this, they're doing studies and they're like you're using GPT every day. Like your writing's getting worse, your critical thinking is getting worse, just like when you use GPS to drive everywhere, you forget how to go to your kid's school because, like you're just used to the car telling you and you forget which direction you're driving, and that for sure happens Like your brain atrophies the less you stress it and the less you do hard things. I think I want to find ways to use the tool, like you said, to help them learn better, faster and give them more feedback in a way that's more usable.

Speaker 1: 39:40

Yeah, and so this is just kind of generally speaking here. This goes back to what I said before is like the, the good use of automation and AI in general is like remove the the time that or, if you can, the time you take, spending the tasks that don't produce revenue and profits. If you can take that off your plate and give you time to spend more, um, more of your day, you know, building relationships with customers, developing business, producing more profits, that's the goal, right. So, like I think about when I'm as I'm evaluating TMS platforms right now and I see how quickly, with some of the automation now a load or an order can be created in a TMS on some of these new platforms and I compare it to our existing. Like we, my company uses McLeod right now and if anyone listening uses McLeod, they know it's McLeod. It can take you 10 times as long to build a new order in McLeod than some of these new ones.

Speaker 1: 40:39

So I've got to come in here On one screen, I've got my rate counter, my tender for my customer. Now I have to go and I have to add a new order. I've got to enter the pickup information, the address, and I've got to create this new location and I've got to add this. No, I've got to add this com and I've got to add this equipment type, blah, blah, add the customer, the rate, but all, and I'm on 10 different screens. Right, where some of these platforms now it's like it just drop your email in here and it's going to read it, and right, and it's like the larger use case with with levity, is it goes beyond just creating an order, right, you guys are talking about, like sending emails back to a customer to respond to a request for a quote.

Speaker 1: 41:17

Um, so I I think about like, if the average, like the, the frontline user right, you talked about the three different tiers, tila, like that frontline user. Like, if, if you, if you think about like you don't, you're going to save four hours a day, maybe, like, with all the little things, you're basically doubling your productivity in a day, with being able to go get more business and penetrate accounts. Um, because you're not doing things like drafting an email. You ever see people that, like it takes them 20 minutes to write an email because they're like it's got to be perfect, and I go back and I'm like, just send the freaking email and you have a lot of these mundane tasks done for you. Now you can spend more time actually doing the things that produce revenue. Here's the relationships Talking to your customers about what's coming up next quarter, things like that.

Speaker 3: 42:02

So to what you said, nate. I wanted to. I wanted to connect those Right, because the thing that levity and we're doing just on like the track and trace piece and like we have that scoped out to do some things all the way beyond that and go down to keep doing this Right. But like he made a really good point and this is the thing that I've really enjoyed about working with Tilo, cause, like I talked to, at least a half a dozen companies that are trying to compete in the space do similar things I don't know a month and like Tilo and his team are very upfront with the time it takes to do this and the work it takes to get it right, where lots of the other companies will just be like, oh, it'll just do all this. And then I ask it the questions I've learned just from working with him, and then it's like, oh well, maybe not yet or we'll see. And I'm like, okay, well, if it's only doing 90% of it, like we don't really want to use it on that thing because like that's going to create a lot more work than it saves. And like they're upfront.

Speaker 3: 42:54

And just to what you just said, to tie this in, just that track and trace thing right For a company of our size.

Speaker 3: 42:59

We figured we have four human beings that are doing a task that I can tell you they don't enjoy, which is staring at macro point updates, looking for whether or not the GPS has slowed down, did it fall off, and then going back and forth and just staring at these things all day, like a human doesn't want to do that. They want to be able to be notified when they need to pay attention to it so that when they're not, they can do things that add more value and are more fun for them, like carrier sales, building relationships, talking with dispatchers, like doing things that create real value. Not staring at little numbers on different user interfaces to see when they changed enough to do something like that is to me like a mind numbing work for a human that nobody should be doing if they don't have to and like that's the thing we want to take off their plate to do things like. But relation, like the whole industry runs on relationships between people. If you're just staring at numbers changing all day like that is incredibly life sucking.

Speaker 1: 44:02

To take your point a step further, like not, is it mundane and mind-numbing? It doesn't develop people. If you're just doing data entry, or if you're just looking at one screen and comparing it to another, you're not developing or learning or growing as an individual within your, a team member within your organization. You're literally doing what automation and AI should do. Yeah, you're not growing at all.

Speaker 3: 44:26

The first six months of my job at a bank as an analyst and this is 30 years ago they would print out a ream of paper like this thick of like legal paper of every bank account for every business in that department and we had to manually go through that by paper and type those numbers off paper into a computer for six months and like at the end of it, I'm like I learned nothing, like I didn't develop, I didn't grow as an employee, I didn't get a better understanding of the job. I will be doing Like I'm literally just taking information from one place and putting it over. It's like, oh, I'm going to pick up a book and put it over here, pick up a box, put it over here. Like you're not, you're not even developing physically doing that. Your brain just hurts by the end of the day.

Speaker 3: 45:07

Yeah, I would ask to you Go ahead.

Speaker 2: 45:10

Yeah, just another thought, and I'm curious about your opinion on this as well. I met a guy recently that is in a completely different field, but he's also building AI tooling and in this case, for nurses. You know AI tooling and in this case, in this case for nurses, where you know there are a lot of nurses out there, but they're even more, even more patients, right, they are chronically overworked because you know they have like 12 hour days and you know what. So they have these tasks where they call patients to remind them to take their medicine. Right, and now you could think of, you know it, through the lens of oh yeah, we're, you know, taking away this task and we're saving them one minute.

Speaker 2: 45:58

Now, the other way to think about all this is what is it that you can do additionally that you are not doing right now because it wouldn't be economically viable or people don't have the time to do it, now that you basically have unlimited resources in certain ways? So he said like, hey, if I'm calling with an AI these old people to remind them of taking their medicine, then I have all the time in the world, right, I can talk to them for 30 minutes. I can listen to them. Yeah, you know, in Vietnam back then you know you can really do the work that you usually cannot do, because the usual call is like hey, have you taken your medicine?

Speaker 3: 46:42

No, please do it.

Speaker 2: 46:43

Okay, goodbye, right, and now you're not constrained anymore in that way, and maybe we should also think more about not only things that you're currently doing, which you can automate, but also things that you're not yet doing or have never done because it wasn't economically viable.

Speaker 2: 47:02

Example that I have from, actually, one of our customers is they had a team for a while, like three or four people, and they were calling customers or former customers all day long, and basically what they did they went through their TMS or CRM, I remember and then they checked okay, who has moved this lane with us in the past, but stopped doing that for whatever reason, right. And then calling them and just like, hey, are you still moving this lane with us? In the past, but stopped doing that for whatever reason? Right. And then calling them and just like, hey, are you still moving this lane? And then they say, oh yeah, we do. Then why haven't you called us? Right? And they're like, oh yeah, sorry, I forgot, right. And then they're creating this incremental revenue on top of that. Right, but they stopped it because it was so expensive.

Speaker 3: 47:44

Right, to have people on the phone all day it, but they stopped it because it was so expensive to have people on the phone all day. It's funny I'm doing that this week. We're literally manually doing this. I scraped our CRM, matched it back out, compared it to the data I was talking about earlier to see which lanes we aren't for whatever reason, and it's always just out of sight, out of mind. And then I build an automation in a TO to basically create tasks for all our reps to touch base with these companies. And then I got an email sequence built in that will follow up if they forget and then set the next task.

Speaker 3: 48:12

Because, to your point, it's super tedious to figure that out and I'm like if we can just get the information at least to the people, then now it's just like oh, it's Tuesday. I completely didn't realize I haven't spoke to Frank over at you know whatever produce company. He hadn't sent me an email and I just forgot. Just a quick little phone call. Build some rapport like making that more efficient is like a huge, I think, value add and I think that's what we want humans to be doing more of doing things with other humans, right, like at the end of the day, if they're doing more of that, your business is probably creating value, whether it's carriers, shippers or even internally with each other, like we said, identifying who needs on your team to spend more time with more FaceTime.

Speaker 1: 48:56

I got a question. So I think about you know, think about emails, and oftentimes, like an email comes in, it gets deleted, like the. The information in that email never gets connected to anything else. Right, the amount of emails I get every day? And if I were to look, if I were to aggregate the entire company that I work for, right, the amount of emails that I get every day or that we get every day from carriers telling us they're available trucks, and the amount of emails we get every day from customers requesting quotes? Right, does levity or could it in the future, right Connect those to talk to the TMS? A lot of the TMS is now have like, oh, it'll look at historical trucks that were available, or hey, if a carrier calls me or an email comes in, I can log that truck as an available capacity. Is there the ability for levity to take those two and be like hey, quote requests came in from customer Also, this carrier mentioned capacity available at the same time. Does that capability exist or is that possible in the future?

Speaker 2: 49:57

Yeah, I mean, both cases work very similarly, right? The only thing that would be missing here is where do you store that information? Right, because I mean we're not a capacity platform, you know, like Parade, for example. Right, I think they're also building something there. But if you have an ability to kind of track that in your tms, that would be best, right? Uh, just to say, okay, we're watching all of the incoming emails from carriers, we're getting the truck lists, we're organizing them, you know. So it's like a real-time feed of you know what's available and matching. That it's its own thing, of course. Right, what you can do is just see okay, I have this lane, I have this lane here, equipment that matches came in, you know, five hours ago and now someone's requesting. You know, you can do these kinds of things, but we're not a system of record, right, like a TMS. So the question is always like where do you move that data and what's the logic for matching and how do you present that information to your users?

Speaker 3: 51:03

So that use case and what Tilo pointed out is we're working with Garrett at Load Partner because they're building an open source TMS and that is a use case, nate, that I've run into a lot and like let's go one step further. One is the emails about trucks and where they are and your customers' needs. So you got, hey, the supply and the demand that is just there and you're not matching. But also, like we have gen logs. We've got other sourcing tools that are telling us where carriers are and it's like when I tried to integrate this with the TMS, I'm like it just you can't feed it in and there's no good way to do anything with it, even if you could.

Speaker 3: 51:38

So one of the things we're trying to outline is like how do you get that all into one place so that all of your emails are connected, so it can pull all the carriers that are sending over where their trucks are? And theoretically you have just a screen where it goes hey, customer has been asking for a truck on this lane the past three weeks. Gives you some identifier. These carriers have been emailing you about this lane. Oh, and, by the way, genlog shows that carrier has been running it for the past three weeks and I don't know what you do after. Maybe it sends an email to the carrier, maybe it just tells the rep to call the shipper, maybe you have that information. Go to the rep and the carrier. Rep. Like we haven't really figured it out, but like your point, nate, is I'm like there's so many inefficiencies just because you're not matching this and feeding it to a person in a way that they can do something with it. That, like the opportunities to just do more with what is already there, I think is a huge.

Speaker 1: 52:31

yeah, I see, like the the two big things. I see long term in the future is like this these kinds of developments and enhancements will allow us to be um, to provide like the best service possible to our customers, right? So if so, on one side you've got like develop new business and on the other side it's like, well, let's service our existing business. And by service I mean if I know ahead of time that this issue is coming and I need to address it, if I can address it immediately and I'm prompted like, hey, this service failure is likely going to happen. I can call my customer right away. Or I can call the carrier, the dispatcher, whoever, right away and take care of that. Or an automated email gets you know, gets out there and gets ahead of the issue. Right, we always talk about like bad news. Bad news gets worse with time, right, if I if two days down the road.

Speaker 1: 53:22

I don't know like, hey, by the way, two days ago, you know, this happened, so your delivery is going to be two days late, right, they're gonna be pretty pissed off about that. But if I know right away, I'm like hey, this happened just now. What can we do to get ahead of it and prepare for a change in the delivery? Any of those use cases? Right, the level of service will be top notch, and that'll really set apart a good broker from someone who's just being lazy, not adopting technology.

Speaker 1: 54:02

It's almost like if you go back 20 years and people didn't want to use a new type of you know, they didn't want to use a load board that was online, right, and eventually, if you don't adopt that new technology and those new ways of doing business, you probably won't have a business, you know, at some point. So that's my because I don't. The technology and the tech stack that I'm currently using is what I would call, you know, somewhat archaic, and that's why we're, you know, we're exploring new products and things, and I'm like there are for sure companies out there that will not exist in five years or will not exist in 10 years if they don't start using what's out there and implementing it in the best way possible for their customers.

Speaker 3: 54:36

So Well, I and I think this is a good thing to kind of wrap with, but I want both of your thoughts on this, because Tilo mentioned something with Control Tower that I think highlights how to look at all of this technology and what you just said, nate, meaning like there are lots of things you can do, but the real question isn't like what can you do?

Speaker 3: 54:57

It's more maybe what you should do and what is going to create the most value and how much time you're willing to spend to do those, whether it's time or money, right. So when there's so many options, it's the paralysis by analysis issue of like I can do everything and there's so many things that choose from, like I ended up doing nothing. Right. I think what is really helpful for anybody looking at this is like working with a company, like levity to first just get these first questions answered. You can start to think about these things of like in there and then trying to determine right, like where should we be spending our time? Where are we going to get our the most return for that effort, and then go what is possible. And then the next question is what can the tools I'm using actually do what? What are they not able to do and where do we need to move from there? Because it is not just buy this, plug it in, and it's going to work the way we expect it to, and I want both your thoughts on that piece.

Speaker 1: 55:50

Yeah, I mean I'm with you on the you know what can you do versus what should you do. It we talk about it a lot. It a lot like ben, you and I uh were we help the tia with coaching their new um, their new brokers that join their uh, the new broker program. And when we do the technology uh discussion, which I think we actually did recently one of the things I always tell them is like, hey, there's all these things out there we're going to talk about a lot of them, um, just because they exist doesn't mean you need to have them right now, on day one, for your, for your bank account number one. But just the functionality, like if just because I can do X, y and Z doesn't mean I need to do X, y and Z, depends on where you're at in your journey as a freight broker.

Speaker 1: 56:33

So that's my take on it is, you know, I think as general as I can say it, like understand what's out there and its capabilities and then figure out through, you know, consult, consultation and whatnot, like what is best for my company at this point and think long term, like as we grow and as we scale and as we change, maybe we pivot the type of business that we're or that you know the markets were targeting, and things like that. Where can I use these things in the future and what is being developed that in a year or two years from now Can I use and can I leverage. That's my biggest take on. It is just understand you know what's there, how you can best use it and what could you do in the future. That's my best advice to anybody when it comes to any kind of tech or automation or AI is, you know, just the understanding of it.

Speaker 2: 57:15

Yeah, what I would add to that? Right, you have a lot of, you know, kind of FOMO driven activities as well, right, so people doing things for the sake of doing them, because supposedly everybody is doing them, and that's not really a good approach in general, right, like I would always recommend to check the math and not only check the math on a unit level of if I automate this, I save two minutes every day.

Speaker 2: 57:40

Also, check how how much effort is it to get there? Right, because there is, you know, this fallacy, especially among software engineers, who look at something and are like, oh, this takes a long time. What if I now spend 200 hours building something that automates? That right, but then, if you think about it, you only spend two hours a yearates that, right, but then, if you think about it, you only spend two hours a year on it, right. So you need to check that math, right. And there are certain things that only really make sense at a certain scale, right.

Speaker 2: 58:10

So we realize that a lot of the smaller brokerages are very interested.

Speaker 2: 58:15

But then you realize, okay, you're not at that scale yet where it's a real pain, it's more like an inconvenience for you and it would be nice to automate this.

Speaker 2: 58:23

But then you talk to a brokerage that's 10 times the size and they have the exact same issue, right, like everything's the same, and then it's a big issue. So, you know, we talked to a guy that you know recently, you know, started a brokerage they're like three people and then he told me, like, oh yeah, that would be nice, but we're not at that scale yet. So he realized that himself. Then, a couple months later, we talked again and now there were 15 people and he told me oh yeah, now I see how this becomes a problem down the line, right, and I guess when we catch up again in two or three months and you know he keeps growing then at some point it does make sense. But you have to do the math on every single case, you have to look at it on a case-by-case basis and not everything is, you know, equally easy to do, right, and some things are not possible really to the level of accuracy that you need to do at all, right, so, yeah, be very cautious about that?

Speaker 1: 59:18

Yeah, that's a great point. Well, I guess, before we wrap it up, Tilo, folks that are you know, want to learn more about Levity, or just have a conversation about what could it look like for their brokerage. How do they find you? How do they get in touch?

Speaker 2: 59:34

Very simply, you go to our website, levityai and yeah, send me the form or email us.

Speaker 1: 59:41

Awesome Ben do levityai and um, yeah, send me the form or email us. Awesome, ben, you have any.

Speaker 3: 59:44

Any other ai related uh thoughts or wrap up the conversation yeah, I mean we should probably talk the rest of the afternoon. I have lots of other thoughts, questions. I'm like once a week I want to send you like a 45 minute like video on like what I think, but what? What his thoughts are. Because it's exactly this Every time.

Speaker 3: 1:00:02

I see something that I'm like this would be a really big time saver and add a lot of value. But then I'm like I don't have the technical expertise to answer the second question of like, well, if it saves six hours a week, is this a 200 hour project, a 12 hour project? And also, even if it's 12 hours, am I getting 90% accuracy, 60% accuracy or 99? Because if it's not really doing what I think it can, that changes the whole math right. It's like, hey, this could be a good bargain maybe not a great bargain if you got to invest 10 times the amount of time to get back out what you think you put in and it doesn't do or isn't capable to do it to the degree you need it to yet.

Speaker 3: 1:00:41

So, like it's interesting times, man, I think, for anyone out there like, definitely dig into these things, play with some of these tools, and I encourage everybody to use things that you can go and check the answers to to really find where it is working and where it isn't right. When you do really big projects, it's very easy to go oh wrote me a large report, it must all be right. But you actually have to do the work of reading it line by line and then going back and going wait, where was this right, where wasn't it right? What is true, what isn't true? Because it's definitely not perfect and it's definitely not 100 percent in a lot of these cases.

Speaker 1: 1:01:17

Trust but verify Right. Yeah, exactly, well, cool percent in a lot of these cases. Trust but verify, right, that's it. Yeah, exactly, well, cool, t-low. Thanks for joining us again. We'll definitely get you on again later this year and we'll see you know what things have progressed and how things have changed. Um, it's super exciting. This is, this is literally the future of our industry, and I'm super excited to see everything that it does. So thanks again for being with us. Anything you want to to wrap up with, no, thank you for having me.

Speaker 3: 1:01:42

Awesome Ben final thoughts whether you believe you can or believe you can't, You're right. And until next time go bills.