Unknown:
0:37
Welcome
Blythe Brumleve:
0:43
into another episode of everything is logistics, a podcast for the thinkers and freight. I'm your host. Blythe Milligan. We are proudly presented by SPI logistics, and we've got our first live interview at manifest, the future of supply chain and logistics. We've got Roger Boza with clone ops.ai. Brand new partner to the everything is logistics podcast, so you'll be hearing a lot more from them in the future. But Roger, welcome
Unknown:
1:05
to the show. Thank you. Blythe, very nice to be here now. What is your role within clone ops? I'm
Roger Boza:
1:10
the CTA Chief Technology Advisor. Advisor. Okay, yes. So I'm the bridge between the CEO David Bell and our CTO Sasha and so how did you get into logistics? It's actually a funny story. One of my cousins is married to David Bell's nephew. So when David decided to start clone ops and do this whole agentic AI for the logistical transportation industry, you know, they were having a lunch. I get together. He was talking to his nephews, like, hey, look, I'm excited about this new company, this new type of technology. And, you know, it's going to be full on AI conversations. And Jacob goes like, hey, we have somebody in the extended family, which he has. He's getting a PhD on artificial intelligence. You got to meet him. He's really, you know, intelligent, and let's see if you guys pan out. So we had a couple of meetings, a couple of conversations. I love David, his mentality, his outlook on the technology's face. And he's like, Roger, if you're the one that I need to help us bridge the technology AI and the transportation industry.
Unknown:
2:13
That's awesome. I didn't know that they made PhDs for AI yet, but it's actually for computer science. But when you write your dissertation, you pick a topic, and I focus on my whole dissertation on AI, on two parts, computer vision and natural language processing. Okay,
Blythe Brumleve:
2:27
so let's get into to AI. What is sort of, I guess, the status quo of AI and logistics, because there's a lot of, I think, misunderstanding of, is it automation? Is it artificial intelligence? Give us sort of the landscape, the eagle eye view,
Unknown:
2:39
all right? So I'm going to start out with, what is intelligence, right? Because this whole word, AI, artificial intelligence, a big buzz. But intelligence, and there's a couple of famous quotes, is, you know, the ability to adapt, to change, right? A little bit more formally, for artificial intelligence, is when you're using computer science skills, algorithms and techniques to mimic human behavior. So with that being said, there's a big difference between automation, which is when you got a set of processes and you're trying to execute them, one after the other using some type of software, but there's no intelligence. There's no predictive power. There is no forecasting involved. It's just a set of rules, and you're advancing forward. Now AI, when you start implementing that, is when you have that predictive capability to make better decisions.
Blythe Brumleve:
3:29
So how does, how does that fit into what clone Ops is, is doing now, because it, you know, there are a couple of No, just AI focused logistics companies that have entered the market here recently. And I think it's, for me, it's kind of tough to tell, like, Okay, how what is the landscape of what AI looks like? How can I apply this to my business? I think there was a quote that David said in an interview. He said, I wish I could clone myself, and that's how the company name came about, and that's how the operations came about. So tell us a little bit about so
Unknown:
4:03
clone OP is focused on phone conversations, to automate them, streamline them. And what is trying to do is any phone operation that deals with a repetitive task, something that is very monotonous, boring, but it's low risk and high value. We get clones for clone ops, right, to try to duplicate and mimic what a rep would actually do in that conversation.
Blythe Brumleve:
4:25
And so it's a lot of like the brokers. Is it just broker specific? No,
Unknown:
4:29
it doesn't have to be just for brokers. We're targeting brokers right now because they're the first ones that have want this, right? That they love the idea. And I have personally seen a broker put a load on that board, and within a minute, he had five calls. Now he can only pick up one. He tries to negotiate a rate, and it might be successful, but it might not come down to nothing. So he wasted the time. He looks at his caller ID, he's calling back the next one, but he doesn't pick up or he already got a load from somebody else. So this is where the agentic AI, the conversational AI, comes in. Nobody goes on hold. Those five calls come in. It picks up for all five of them, and it begins the negotiation. When one of them actually is successful, it disengages the other ones in real time. It tells really listen that load is already booked. You know, better luck next time. And for the one that actually had a rate negotiation successful, a confirmation is sent to their email, and then it proceeds.
Blythe Brumleve:
5:25
So what does it look like, I guess, from a user perspective, if I'm going to work, I'm a freight broker, I have clone ops like, what does it look like during my work day? Is it a dashboard? Yes, interface that I'm using,
Unknown:
5:39
it is a very interactive dashboard, so we wanted to make it very user friendly, easy to use. In fact, one of the things that we're doing is self onboarding. You go, you create your account, you log in, you're gonna see a marketplace where we have some of the AI agents already configured. That way you can hit the floor running. You see one that you like, let's say that you wanted to carry yourselves doing renegotiation and load booking. You will hire that one, and it starts working immediately. Now, if you have a TMS that you want to integrate to pull in the load information and the carrier verification, we integrate with a couple of them, including highway for carrier verification, but you put your credentials in there to get the API access, and the agent is live. So next time somebody calls a phone number provision by the AI, it picks up and starts the conversation and negotiating the rate. And if you got 1015, 20 calls, I'm not saying that you want to look at 20 calls at the same time, but the system tracks all the calls, and you can pick a subset of them, you put them in your on your live dashboard, and you're just looking at them going on. Now. The idea here is to be able to intervene, to have a human in the loop with a takeover functionality. We are the first ones doing that now. And in fact, we actually have a we went after a patent for for it, because what you want to do is one of those conversations is going either extremely bad, right? You know, the customer is screaming at the AI agent. It doesn't understand that he's having a hard time. It just doesn't like the technology. You don't want to wait for that to be transferred to a human rep. If you're looking at the dashboard, right, and you're monitoring that, you just take over the call, you step in and you mend the situation. Oh, so
Blythe Brumleve:
7:15
you almost have to, this is like a second screen, so you're working on the same dashboard. You just switch tabs, yeah. Oh, good. Because interesting, because I was thinking, like, if I'm a broker, I usually have at least two different screens. A lot of them have four. And so maybe they're working on, you know, the working on their customer relationship, and, you know, building those, you know, cold calling, or anything like that. And then they can watch these agents in another screen to make sure that they're doing their jobs too. If they
Unknown:
7:40
have multiple monitors, they can be from our dashboard. They can detach that tab and put it on a separate screen and just monitor that live. That's crazy. And since you mentioned cold calls, you can have an agent right that is doing cold calls. You can have another agent to negotiate the rates for you. You can have another agent that can work as a secretary for anybody that calls in. It's like, hey, I need to book a load. It can transfer you to the agent, the AI agent that does that. If it's not available, he's not busy. Doesn't matter, though, the information goes to the screen whenever he needs. He can sit down, he can take over, and he sees that live transcript with keywords and sentiment analysis for it.
Blythe Brumleve:
8:14
Oh, wow. So is it? Is it one voice for a lot of different customers. Or how does that work? Is it trained on your own voice? We can, we
Unknown:
8:23
can clone your voice. So there's two ways of doing that. Yeah, we can clone your voice. There's a fast way and a professional way. There's always a trade off between accuracy and quality and speed. So if you just want to speak for about two three minutes, and want us to clone your voice. We can get that done. In fact, we can get it done with only about 30 seconds of your voice, but it's not going to contain the full range of vocal sounds, emotions, and then it starts sounding robotic. With about 15 minutes to 30 minutes, we can get something that, in general, most people will say, All right, I think you might be an AI agent, the robot. But the voice sounds nice. It has a good quality. So
Blythe Brumleve:
9:02
do you have maybe several voices for each company, or is it maybe one voice? They can
Unknown:
9:06
they can pick their own voices. We have a selection of over about 2030, voices. Now we can detect spoken language in 16 different languages, and we can speak back in 32 Now, ideally we want to have the same amount of detection voices as spoken voices, but that's technology that still needs to be improved. Well
Blythe Brumleve:
9:27
that should I mean light bulbs are sort of going off because for a lot of drivers, they speak different languages, and not all brokers speak all different languages. And so I would imagine that that helps facilitate those conversations too,
Unknown:
9:38
correct? And once we go live, when you call to, let's say, negotiate the rate or book a load, if you start speaking in Spanish, the agent will greet you and talk to you in Spanish. If you're talking in Russian, you will have a Russian accent, and it will talk to you in Russian. It's crazy, yeah? What a time to be alive. And it can be, yeah, it can be either female or male. Wow. Okay,
Blythe Brumleve:
9:58
so what does that I guess, a typical onboarding process look like? How long does it take to get set up? It takes
Unknown:
10:03
a couple of minutes. Well, you create the account. It's a couple of steps. Typical account creation, right? Username, password, the name of the company, et cetera. Once the account is created, you log in into the dashboard, and you see our marketplace if you see an agent that you want for your use case or your scenario, you hire that one, you give it the credentials to connect to the TMS or to any other API endpoints that it needs to pull in data you might have, Excel sheets, CSV, it really doesn't matter. We support all of that, but you'll connect it to that with our connectors and the agents. Starts working right there, then, and
Blythe Brumleve:
10:38
so they integrate with all of your different sort of software tech stack, correct?
Unknown:
10:43
Yes. Wow, yeah. So what
Blythe Brumleve:
10:45
does, I guess, sort of the, what do you envision as sort of the role of a modern day brokerage look like, like if I were to start a brokerage tomorrow, it sounds like I wouldn't need a lot of that overhead. I could have a lot of the help. Right off the bat, it sounds like kind of
Unknown:
10:59
a game. So this is a perfect opportunity if you want to scale right your operations and your business, because human resources, they're very expensive, they're hard to come by. You got to train them. Instead of going that route, which is very difficult and sometimes very costly, you can use the AI agent. So what would it look like for him? He goes in, he logs in, he sets up the agents that he wants for his scenarios to automate those processes, and he sits down and looks at the screen. Now, if he gets bored, he can go and do something more productive. And that's actually what we want do, something more productive, more efficient with your time, strategic thinking, strategic planning, and that type of
Blythe Brumleve:
11:35
things. And so what there's just so much to because I feel like I need this, and I'm not even a broker.
Unknown:
11:43
And that's why David Bell said he wishes he could clone himself, because it can represent, at least with the modern technology that we have right now. Even though it's not 100% perfect, and AI will never be 100% perfect, it can mimic a lot of the things that you can normally do. We're actually working for something that is going to come out later in the year, which is a virtual avatar that one will clone your voice, will clone your appearance. And you can even have a join a Zoom meeting and have a conversation in the meeting for you, they can take the notes for you, and then it will give you the details and the summary. What is the biggest sort of the reaction
Blythe Brumleve:
12:16
of other people who may be on the call is there is
Unknown:
12:19
a big spectrum, and it's all over the place. Some people are very receptive to it. Other people really hate it. Usually you don't want somebody on the other side of the phone that sounds robotic, right? And we have identified three things that make it sound robotic, the voice itself, the content that is saying and the delay, if it speaks too fast, or it speaks too slow. As soon as you tell from any of those, like, Hey, this is a robot, you disengage from the conversation. You get angry, and they because they get frustrated, right? They're like, it might not understand me, previous interactions with AI quote, unquote is usually very, you know, frustrating, like you call your bank and they have an automated response service like press one for just yell on the phone customer support. So there's this mentality that as soon as you think robot, you have to ask for a rep, right? Representative human, give me some, and they start cursing at it. So there is a big spectrum. We see all of it, but we're not trying to mask the voice and the service as a human right? It's very friendly. It's there to help you right that way, you don't go on hold, you don't get frustrated. You can ask you questions. You can interact with it, right? So it's not your typical answering service that it's a predefined answer to a question. This one will engage in the questions for you. That's where the modern technology is at with the large English models like GPT Gemini and the famous ones out there.
Blythe Brumleve:
13:42
So what does, sort of, I guess, the next step look like? So like the final frontier technology, and
Unknown:
13:50
that's part of the AI roadmap, right? So right now we're targeting brokers. Are the first ones to, you know, to get this type of technology implemented, not necessarily for the customer forward facing, but you know, for their regular operations. Now eventually, the carriers are going to get annoyed, upset and frustrated because they have AI agents coming from all over the place, and since this is all technology and automated, they're going to get blasted by it. So we're going to transition and provide them with AI agents for carriers. So now you got a broker with an AI agent calling a carrier, but the carrier doesn't pick up. It's another AI agent that picks up, and they have the conversation and they negotiate the rate and do whatever they got to do. And at that point, we're trying to coin the term API AI, where it says AI talking to AI, doing everything on the back end, and you get your time to do something more productive. The truckers just drive, and everybody does what they're good at.
Blythe Brumleve:
14:45
Because I'll be honest, that was one of my first initial thinkings with drivers, is that they're going to hate this, like they are just so sensitive about technology and things like that, but if it's to their advantage to start using it, then I think that they would be much more willing to, you know, use this technology, because then they don't have to do their own grunt work that they don't like doing, either they don't
Unknown:
15:06
like talking on the phone, they don't like being on hold. So imagine you're a trucker. You set up your AI agent. You're interested in four or five loads. Your agent calls all of them at the same time. The first one to negotiate the rate that is to your advantage. That's the first one that closes the deal, and now you just go make your money. What does that,
Blythe Brumleve:
15:23
I guess, sort of the price points look like for these are you paying per agent, or it's,
Unknown:
15:27
at the moment, is per agent per minute? Okay? Yeah, there's a lot of services that go behind it, but it's usually permitted. Now we're going to have a pricing tier. So. Right? That, depending where you either pre buy the minutes, right? And then it goes a little bit above that, right per minute, or, you know, depending on the tier, is just, you know, 2025, cents per minute. Now, technology advances very fast, right? It actually follows something called Moore's law, where the amount of, and this gets a little technical, the amount of transistors in CPUs double every two years. That being said, we get faster models that make better predictions. They're more engaged and more accurate and less expensive to train and have that lowers the price. So eventually this is going to come down into the order of cents, below 10 cents,
Blythe Brumleve:
16:16
yeah, because I would imagine that that's a large problem with the cost structure right now, just paying for all of the different tokens that you get from all of these different large language models. Correct now, for companies, are you building sort of their own large language model for them?
Unknown:
16:33
No, it's not just for them. We do use chat GPT, but we don't retrain that model. Now, I'll be honest, since I come from that background training chat, GPT itself requires a company like Google, Microsoft or Amazon with that amount of resources and hundreds of millions of dollars, so we couldn't retrain those models. What we do is something called knowledge transfer, right where it has a so by the way, do you know what GBT stands for? Is a is a generative, free trained transformer. Oh, what was that last word transformer? Oh, so whole bunch of fancy words, but they actually each mean something, right? Generative, because the AI model can generate content outside of its distribution. And this is where we like it for AI conversations, because it can generate the conversation flow, and it follows like a human conversation, pre trained, meaning the model is ready to use and it has general knowledge, right? The way to train them is you grab a massive corpus of text. I'm talking about Wikipedia pages, I'm talking about library books. I'm talking about websites. You grab all that data and you feed it to the model, you let it train with it. Now, the last word transformer, that's just the structure of the neural network in there, which is very good to handle temporal information as well as conversation in this case. So all of that given together, they're very expensive to train. Now, once we got that model, what we can do is like, Hey, you already know some general knowledge about logistics, transportation conversations. You've seen it. You train on data like that. But here's my specific domain right now, logistics and stress, information and transportation. This is how the data and how you should behave within this sub domain. So we transfer that knowledge into the model, and is now fine tuned for these scenarios
Blythe Brumleve:
19:55
I'm just blown away at, like, how fast we got here. Does it surprise you that some of this technology, like exists, it does,
Unknown:
20:03
it does, and actually it changes everything when I started. So my background is in computer science, with a track on software engineering up to my doctor right now, which is specializing in AI. Sorry, but back in 2016 when one of my professors convinced me to go for Masters and PhD, he told me, like, Roger, you need to focus on AI, because in 10 years, and this is nine years ago, in 10 years, AI is going to be mainstream. It's going to be all over the place. And if you want to make money, study it, I don't study the rest. And true enough, it had progressed so much, and it's mainly due to technology. Right? Hardware advancements have been going through the roof. They get faster, they get cheaper to use. You know, we have more industries getting involved into it, so we have more resources, and everybody you know it's pushing to make the technology better. How
Blythe Brumleve:
20:55
do you deal with some of, I guess, maybe the pushbacks that you might get from from potential customers. And then how do you answer to those
Unknown:
21:01
pushbacks? What kind of pushback You mean, like, a, is this robotic? We don't like it, or it's like a nothing along
Blythe Brumleve:
21:06
those lines, like, I know, no, I want the personal touch. I want to keep a personal touch with my customers, or, you know, drivers, or something like that. So
Unknown:
21:14
one thing that I recommend to them, and that has actually happened to us, right? They want that personal touch. It sounds like give me your conversations, right? You have recorded conversations of you talking to your customers. Let me fine tune the model with that type of information that we when the agent is talking to it, it has your same personality or very close to it. So that's one way to go about that. Another way is for people that are rejecting the AI side of it, don't try to hide it, right? This is an AI agent trying to help you, right? So that, yeah, so that you know, you don't go on hold. You can get somebody on the phone really quick. You can get what you want other conversation. And in our case with clone ops, right? There's always a human in the loop. Somebody's always looking at a dashboard ready to take over if something doesn't go accordingly. How
Blythe Brumleve:
22:00
do you know if something gets screwed up with the agent?
Unknown:
22:08
Those are the phone questions,
Blythe Brumleve:
22:10
much like a regular employee, but. Yes, you would want to have corrective action, right? Correct? Yes.
Unknown:
22:14
So one thing that we do, and I think a lot of the competition actually does this, is called sentiment analysis, right? So a conversation can be either neutral, negative or positive, right? We take it as a little step further by also analyzing the mood of the conversation itself. Is the mood happy? Is that aggressive? Is this frustrated? You know, are they angry? And an additional step, we also analyze the voice. Are they whispering? Are they screaming? Are they talking with a regular tone? So we got those three metrics, and using some clever functions, right mathematical functions, we get a conversation score. So that's one of the first metrics that you can track. And normally, if something is going down south, right, they're screaming right. They're toning the voice has a very high pitch. The wording that they use, the context is very negative, and the mood of it is going to be angry or upset, and that drives that conversation score very low. Now in our dashboard, when that conversation score drops that conversation comes up on the dashboard closer to your eye where you're normally looking. That the first couple of conversations maybe five or six, and it turns red, right? We have visualizations to let you know. Hey, something is going on here. We also track and highlight, you know, keywords, right? Hey, you're not understanding me, right? That combination will highlight certain words that are indicative of something going wrong.
Blythe Brumleve:
23:37
I'm just stunned at this is, like, where we're at right now, because I feel like I want to hire like, 10 of these for myself, and I'm not a broker, but I do. I have really thought about how I would introduce that, how I would disclose that. Am I going to be secretive or not? So I love that you brought that up, that you should be upfront with it, and you should let other people know. Because I think once people know, then they feel much more comfortable interacting with the system. Now we had a couple more minutes left. Anything that you feel is important to mention that we haven't already talked
Unknown:
24:08
about. Yeah, so for companies that are trying to adopt this and they're a little bit scared. There's a couple of things that they can do, right? We need data for this conversations AI agents to work. So preparing the data whether you have, you know, voice recordings available, you know whether you have scripts that you use to train your reps, we can use that data to custom tailor the AI agents for them so that it gets their personal touch. Also, they're sometimes worried about implementation cost, integration cost. The way that we're doing this is we're reaching out to the major TMS so that it's one click button to hook up to their TMS, right? They all have API endpoints, and all we need is the credentials to access the data, so they shouldn't be too concerned about that. And as far as security goes, we are in the process of getting our SOC two type two compliance, which means our services are monitored for 90 days to make sure that, you know, everything is secured in there, there's penetration testing. And you know, overall, in general, you get an easy to use dashboard. You got conversational AI to help you expand scale and, you know, operationalize a lot of your scenarios and automate them. But you also get the security on the back end with us.
Blythe Brumleve:
25:27
I love that you brought up security because I meant to ask about that just in case. You know, I think everybody's worried about their own sort of data security, so thank you for bringing that up.
Unknown:
25:36
Yeah. And one more thing I want to say about that is the data at this point is not shared with any of the other customers. So when you create the accounts with our services, it gets deployed in a separate space that even if somehow it manages to get hacked or penetrated, it doesn't interfere with anybody else's data.
Blythe Brumleve:
25:54
That's this was a really insightful conversation. Thank you, Roger for sharing this, because I it's one of those things where I've tested around with like Zapier and, you know, these other things, and it just didn't quite work too well for me then. But this was a couple years ago, and from everything that we've talked about today, feels like all of this is
Unknown:
26:14
just and that's where I've seen a lot of people get discouraged. They have tried, you know, years ago, you know, some form of AI, and it doesn't pan out. It was not good, it was not ready, and you just have this general feeling that we're back to that now. It's just high people are excited, but no, that's not the case. Technology has made a lot of progress. Well, that's
Blythe Brumleve:
26:32
awesome. Hopefully, maybe I can get it like a demo account, and I can try. We have those available. Yes, awesome. Well, perfect. Where can folks, if they want to sign up for a demo they want to follow clone ops, connect with you? Where can they reach out
Unknown:
26:44
to you? Yes, so I'm in LinkedIn. We have our website, clone ops that AI we're planning to launch for the public at the beginning of March, maybe first week or second week, but any of those you know, media venues are available to amazing Roger, it's great to have you. Thank you very much. Awesome, perfect.
Blythe Brumleve:
27:06
I hope you enjoyed this episode of everything is logistics, a podcast for the thinkers. In freight, telling the stories behind how your favorite stuff and people get from point A to B. Subscribe to the show. Sign up for our newsletter and follow our socials over at everything is logistics.com and in addition to the podcast, I also wanted to let you all know about another company I operate, and that's digital dispatch, where we help you build a better website. Now, a lot of the times we hand this task of building a new website or refreshing a current one off to a co worker's child, a neighbor down the street, or a stranger around the world, where you probably spend more time explaining the freight industry than it takes to actually build the dang website. Well, that doesn't happen at Digital dispatch. We've been You
Unknown:
28:39
tags, oh,
