[Automated Transcript]

Richard: Most of these companies cannot afford to invest in sustainability. They don't have the margins. They don't have a bunch of capital sitting there where they can say, oh, we're gonna invest in a bunch of electric trucks.

Now obviously, some companies do that. Ikea, PepsiCo, these are great companies, very sustainability focused for their brand, and so they've chosen to invest a lot of money in emissions reduction, but that is the edge case. That is definitely the exception rather than the rule. Most companies, they don't have money specifically for sustainability, 

Deep: Hello, I'm Deep Dhillon, your host, and today on your AI injection we're joined by Richard Savoie. Richard holds a BS in electrical and computer engineering from Northeastern University and he is the founder and CEO of Adona Technologies where he leads the development of AI powered route planning tools that increase delivery speed and accuracy while helping global brands cut emissions and transition to EVs and other sustainable modes of transportation.

Richard, thank you so much for coming on the show, 

Richard: and thank you for having me. Deep excited to be here. 

Deep: Awesome. So why don't we start off with my favorite question to start off with why don't you tell us what did people do before your solution and what do they do now and what's different? 

it's always helpful to just pick a company, if you can tell us one.

Otherwise, just make up a hypothetical or something. Yeah, 

Richard: sure, sure. Yeah, I'll give you a really quick timeline, right. So it's like BCAD. There's before GPS. Okay. And when GPS became available, and that was a very low technology time in supply chain generally. And it's still, struggling to digitize in some places.

But before GPS kind of the dark ages of mobility tracking and things like that. And then after GPS there was obviously a boom in companies like Garmin and Teletrack that offered GPS trackers for trucks and things like that. And, you know, some of us had them stuck. On with suction cups to the windshield of our car.

But we all understand that that opened up a lot of opportunities to optimize commercial fleets and things like that. And I think maybe the most notable example I can give you from that time period, say like the mid two thousands was UPS. They developed a program called Orion that, had actually a bit of virality at the time because it popularized, at least in our world, this concept in, north America, that sometimes three right turns are better than a left turn.

Right. Have you ever heard that before? That kind of adage that maybe, uh, I have not, no. I'm 

Deep: having a hard time thinking of why that would be the case, but maybe you got a really long light or something. 

Richard: Yeah. I imagine that you're in traffic and you're at a really long light, and you're trying to take a left turn, but the light keeps going green.

A couple cars go and then you're stuck. Right. And then it goes green, a couple cars go, whereas right turns, especially where I grew up in New Hampshire, Massachusetts, you can take a right turn on red as long as it's clear. So a lot of states, I don't know about Washington where you are, but you can actually Yeah, we free take a right turn.

Yeah. You can take it freely, right? Yeah. So being able to go around the block unobstructed and keep flowing can be a lot more efficient than that one left turn. And the other impact of that is the psychology of driving. It's a lot less frustrating oftentimes to do that. So , that's one of the insights that started as an example coming out of those post GPS days.

And then there's the new era that we're in right now or sorry, there's two. One was cloud computing. So then with the introduction of cloud compute, suddenly the algorithms that are used to optimize fleets, were able to be run highly more paralleled in cloud. Right? We had a lot more computing resource to then do more quicker and deploy it at scale to mobile apps and things like that.

So that's a big change. Then now we're in the current era of AI where combining cloud compute efficiencies with AI heuristics and automation and agen AI can really make another amplification a compound in improvement on that efficiency. 

Deep: Got it. So what would you say your core problem is?

Is it really on the nav side, like in the Google mapping kind of space in all the way down to the left right turn selection in the route planning? Or is it down to like, which driver to pick with which car, based on what carbon load or whatever? What are some of the core problems that you guys are facing?

Richard: For people that aren't really super experienced in the optimization space there's a couple of examples that highlight the technical problem in the, shall we say, bigger picture problem, right? The economic and environmental problems. 

Deep: You know, why, why don't we start with maybe we jumped in too fast.

Maybe start with like, who's your typical customer? What do they have? Like, do they run a small fleet or something, or do they run massive fleets And what are they trying to do? They're trying to like optimize, who picks up the pizzas most efficiently next and delivers 'em. Like what? Like give us, give us a more crisp flavoring of it.

Richard: Yeah. And, and thanks, you know, two, two engineers. Of course. I know. Jumping to the technical, I know every time I talk to another 

Deep: engineer, 

Richard: we're such a trope, we're such a stereotype for a reason. Okay, so big picture, let's just start at the big picture. Transport in general is the second largest emitting industry in the world.

So for greenhouse gas emissions, transport is number two behind the grid. But as the grid is electrifying transport is gonna be number one by 2030. And if you imagine the industrialization of economies around the world where e-commerce growth is, kind of mature in markets like the US but in lots of other places in the world, it's booming.

So this transport component of diesel fuel that's being burned and cold, that's being fueled to to do, to support them is just a massive growth emission sector. Okay. Now in. that Massive component of emissions, commercial transport and passenger transport are about half and half. Okay. 

Deep: Okay. 

Richard: Passenger transport is much easier and faster to electrify.

Full stop. 

Deep: Smaller, cars smaller batteries, 

Richard: and they're used much less. Right. So smaller, cars smaller batteries and they're, you know, typical utilization of a passenger car is only 5% of its life. Sure. 95% of the time it's just sitting there in a driveway or a front house, sit in front of your house looking good.

And obviously Tesla's trying to help solve that with things like, self-driving robo taxii type things where you can have your car and then you can send it out to fer people around. Yeah. Or 

like 

Deep: ride sharing apps and that kind of thing too. Yeah. 

Richard: and ride sharing. What a almost ridiculous name it is now because for the 50% of the time, Uber's data shows that they're running around empty.

There's nobody in the Uber, and then usually it's one person that's in the Uber. So there's no actually net benefit to congestion in cities with ride sharing. And it's been statistically proven no benefit. You mean

with 

Deep: a driver? 

Richard: Not you don't mean the with the driver. Yeah, with with a driver.

Not, not the cars. You just grab 

Deep: and go. 

Richard: Well even imagine a robot taxii world, right? Where you've got a, a self-driving car, it's still a sedan, it's still a Tesla model Y or whatever, and still carrying typically one person around. And so for one, yeah, one person needing to go from A to B, if you change those from petrol, gasoline, diesel to electric, that's great because you're gonna make a big difference on emissions, but you're not actually impacting congestion in cities very much.

Right? Now you flip over to the commercial fleets, right? Utilized much more. And for instance here in Australia, only 1% of the commercial vehicles are articulated trucks, like big rigs, semi trailers articulated just means that it's got that hinge. Yeah, right in the front. Only 1% of the vehicle population, but responsible for 15% of the emissions.

Deep: Oh, okay. 

Richard: So commercial fleets actually have a much more outweighed effect on emissions 'cause they're used to so much more, more miles driven and more hours of the day that they're sweated. So now you can kind of understand why commercial fleets have a benefit that we can, attack to reduce emissions by optimizing them, by using less of them to do more work.

And the reason is they run empty deal a lot. Yeah. Keep 

Deep: for yeah. Keeping full, being more efficient with their roots. And also there isn't as obvious, clean alternative right now. At least, there's no battery powered semi option right now. And there's, it sounds like most of them are gambling on hydrogen, but I don't know when that's gonna happen 

Richard: Exactly, and then you've got that mix of long range, but then you've got this massive metropolitan, urban dense delivery. So now let's just break that in half, right? So I talked about the articulated trucks being really high emitters, but 53% of the kilometers driven and the cost of transport of a product, like a commercial product, something you get you order from Amazon or anywhere, 

Deep: yeah.

Richard: Is in the last mile. And the last mile is what we call from like the last distribution point, whether it's a store or a warehouse to that store or your home. 

Deep: These are like the Amazon trucks we see driving around or the UPS trucks. I don't know what you call it, but not where it's like a semi detached or whatever, but.

Richard: Yeah. 

Deep: And those are electric, at least in Seattle. 

Richard: Well, and so that's a great point is because Amazon has been the leader in both, autonomy, efficiency and electrification, but the rest of the world is struggling, right? USPS has done, you know, made some leaps.

They're investing in electric vehicles and stuff, but you've got this massive mid-market, right? The bell curve. You've got this huge mid-market that, uh, just doesn't have the capital in place to do the level of electrification like an Amazon or A-U-S-P-S does. 

Deep: Yeah. 'cause it's not only about buying the trucks, right.

It's also about having the charging infrastructure. 'cause these things are huge and they, yeah, they're taking a lot of power. And when you're running a, like a city like Seattle, for Amazon, they've got the scale to invest in the charging infrastructure. But if you're, I don't know, a Pizza Hut or something, like maybe it's not gonna pencil.

Richard: Yeah. Yeah. And, then you think about like, you see parcel trucks branded all the time, but next time you go outside, look at all of the kind of mid-size box trucks that are serving cafes and grocery stores. Yeah, yeah. Food distribution, fast moving consumer goods and food is an enormous market.

I mean, it's insanely huge. And then there's also this kind of mid-market of parcel couriers and stuff too that subcontract to UPS, Amazon. And interestingly, some companies like Amazon and UPS, they don't actually share their full technology stack with those mid-market distribution partners.

So they're kinda left on their own to figure out their own it. 

Deep: No, I mean, like, you know, they partner with Rivian, Rivian makes the truck, but there's a lot more logistics behind the scenes. 'cause like I order a. You know, a Motorola, Moto phone or whatever. It's in my house in two hours.

I mean, it's crazy a city in Seattle, it's like I'm getting like guitar strings within like two hours now. And they're coming from, you know, Spain originally, and they're like, at my door and it's crazy how fast they're getting. Okay, so let's go back to your problem.

I thought, let me see what I thought your problem was. It sounds like it's a little bit different. Mm-hmm. And there's a bit of a disconnect for me mentally in terms of the environmental thing. But I thought your problem was like, I've got a fleet. I don't have the infrastructure of an Amazon. I've got some cars.

Maybe one or two is electric, maybe some of 'em are gas. And I want you to tell me exactly which driver of which car should take which order and drop it off. At which point, like right now. And then you're gonna give me the. Car that optimize whatever it is that I'm optimizing for, whether it's latency, like I'm gonna go pick up the person at the right time.

Or maybe I can wait three or four or five or 10 minutes on average because I'm gonna send in the most carbon efficient vehicle and I'm gonna wait around for that or something. Is it something like that? 

Richard: Yeah, definitely. I mean, so we, we kind of unpacked the macro problems, yeah, the sustainability problems, but let's talk about the company problems to make themselves efficient and to save money.

If you have one van and literally you go step outside, you hop in your van, you got 10 items to deliver. In Seattle, there are millions of permutations of different routes that you could take. You pop one into Google Maps and it'll give you the most efficient route.

But if you're trying to find the most efficient route to go through 10 points, that's a classic operations mathematics problem called the traveling sales person problem. Oh, yes, I remember really well. Understood. Yeah. Yeah. Okay, cool. Yeah, exactly. Like it's a pretty classic problem. It's been studied for decades, you know, for 70 years.

And, it's funny 'cause research teams still compete to shave like a, a 10th of a percent off the solution time for the kind of standard research problem. But in practicality. You're 

Deep: that it doesn't account for, right? Like it doesn't account for traffic, doesn't account for Exactly. Accidents doesn't account for, I mean, we had some guy, drop 50,000 boxes of crabs on 99 the other day, like a year ago.

Like none of that's in that data. Right. So I'm assuming No, you guys have a more kind of adaptive machine learning like approach that kind of can listen to real time signals, something like that. 

Richard: Well, you jumped way ahead to the realtime piece, which is very common, but let's just talk about the strategic piece, right?

Deep: Yeah. Okay. 

Richard: So you've got your 10 items that you gotta deliver. You gotta figure out a route. Not too hard, okay? You can figure it out, but , there's millions of permutations. But even the human brain can do a reasonable job to look at a map and be like, I kind of know. Yeah. And you're gonna be within say, 10 or 15% of what a proper algorithm would solve.

Like a, heuristic algorithm looking for a really optimal route for you. But then imagine now you've got a thousand items to deliver and you've got 10 vans to do it a hundred each, right? If you just have that as raw data and you look at it on a map, you're gonna be like, I have no idea as a human.

Like, what's the most efficient solution? And choosing wrong in that instance could be an enormous difference. A hundred, 200% difference. Between what's actually optimal and what you choose, but yet a huge portion of the logistics industry still makes defined territories on a map with a marker.

Deep: Yeah. Or 

Richard: nowadays on Google Maps, you know, digitally, but it's manually done. 

Deep: So it's the decision of what products to put in what vans also. 

Richard: Yep, 

Deep: exactly. And then and how to group those products such that, maybe they hit one area of the city or something. And then once you've got the individual truck or van populated, now it's like for the items that are sitting in that thing, like what's the optimal route?

Richard: Yeah. And there's a massive problem in that algorithmic industry, which has been around for 20 years, you know, since GPS and whatnot. The massive problem is that the algorithms, when they solve for a really efficient set of routes, they don't match what the driver wants to do. Lemme just be, I'll give you an example, right?

If, if you take a classical, what we call vehicle routing problem algorithm these classical algorithms create what we call flower pedal routes.

Meaning that say you've got a thousand deliveries, you got 10 vans. All the 10 vans are gonna originate from this one distribution center, and they're gonna go out and they're gonna do a big loop and they're gonna drop off stuff along the way and they're gonna loop back to the depot. And that is mathematically the most efficient way to do it.

Okay. That's proven, no debate there. But it is very disconcerting and psychologically wrong for humans because humans are gonna look at all these routes overlapping each other in this big flower pedal system and be like, why is deep going. Crossing over where Mary goes and where Bob goes and everybody's crossing over each other.

Humans have this massive psychological rejection of that being efficient, what they wanna do. And if you're a parcel car, you're definitely, what you wanna do is you wanna hop on the highway, head to a dense area, park as few times as possible, and smash out as many deliveries as you can on foot when you park the minimum times.

'cause parking is a nightmare for couriers. Right? Right. Just look in any city. 

Deep: Yeah. 

Richard: So the psychology of driving versus the traditional algorithms is all wrong. It doesn't match up. And that's part of what we're solving with AI, is we're taking all of that human experience, the human preferences, the results in the real world, and then combining it with all of these other data points, traffic, weather and real time dynamism and creating something that may not be the most efficient mathematically, but it'll be really close and everybody will wanna do it.

It might be better. 

Deep: It might be better, right? Like sometimes it 

Richard: is. 

Deep: Yeah, so let me read between the lines. You start with a new client. You're gonna track their drivers and let them go where they want to go based on whatever their knowledge, not they already have, not fully, but 

Richard: yeah.

Deep: Okay. So not during the learning phase. You're not like letting them run wild yet even? 

Richard: Correct, because we 

Deep: don't want them, them to 

Richard: train it with bad habits. 

Deep: Okay. So let's walk through that. What's that training look like? What does that new customer onboarding period look like?

And the other thing I just wanna point out is going back to my earlier question, like what did customers do before you, I feel like you just answered that. The baseline were these mathematical model approaches that didn't take into account both the human knowledge and real time signals , and traffic patterns and other signals.

Richard: Yeah, exactly right. It was a very, experience driven heuristic approach. Then the technology's gotten a lot better and our competitors do really well in terms of creating efficiency. But this concept of making it more humanistic to then increase adoption and the speed of adoption is really critical.

'cause, um, we've seen lots of these projects fail over the years because the CFO says, oh, this is great. It's gonna be so much more efficient. We'll save a lot of money, we'll reduce emissions, great. And then they implement all these things and the drivers reject it and just refuse to do it. And a lot of these drivers are unionized and drivers are actually really expensive.

And there's 450,000 manufacturing jobs going unfilled in the US right now, including transport. Like you don't wanna make your drivers angry because they will leave, they'll go to your competitor and when they come back, they'll ask you for 10 to 20% more money. It's actually a really tough market for drivers right now.

So this concept of drivability and safety in. Distribution and transport has actually come to the forefront. 

Deep: So walk us through what happens with a new client then. How is the body of drivers knowledge, like captured and represented in a model and how do you actually, , go about training?

How much of it's, like generalized knowledge that your models have and maybe even just like, how is the model even formulated? Like what's the problem formulation? Is it a reinforcement learning like structure or like what it, what does it look like? 

Richard: Yeah, so the typical journey is this three archetypes that we deal with , of companies of clients.

One is a transport company. That's all they do. People give them stuff. They keep it in their warehouse and then they just deliver it for them. Okay? Mm-hmm. That's a third party logistics. Three pl you've got then a manufacturer, which is kind of the polar opposite. They only manufacture stuff. And then they rely on these transport companies to do all of the deliveries for them.

It's fully outsourced. And then you've got probably 50% of the market, which is so maybe call that 25 and 25 of our client base. And then 50% is a manufacturer who has a mix of their own distribution, maybe for some things in some areas, and then a massive use of third party logistics companies in other places for other products.

These medium and large companies. And these companies could be, you know, a hundred million dollars a year in revenue to $500 million a year in revenue. And there's just a massive cohort of them globally. So you got these three archetypes. All of them are interested in being more efficient, saving money, reducing emissions.

And so you have to triage it and do a lot of discovery to figure out, okay, well what it stack are they on? Is it API capable? Can you integrate in these types of algorithms via our cloud? Or if not. Then what's their plan? Are they planning on moving to a more advanced tech stack replacing their ERP system, their transportation management system and, and whatnot?

So depending on all those options, we're a very API forward company, right? We have some competitors that are very platform centric. You have to move on to their entire delivery management platform. But what we find is that that alienates a big part of the market that just doesn't want to go onto somebody's end-to-end platform.

And that's fine. So we integrate in at the right point to solve the right problem, which depending on those three archetypes, is typically around how do I do the same amount of work or more work with less vehicles. That's always gonna be the top problem that they wanna solve. And if we can come in and show them how they can do the same amount of deliveries with 10 to 20% less vehicles and less costs associated with those vehicles, and as I said, drivers are super expensive.

It's kind of a, a shrinking field. Yeah. You know, automation is a perfect solution for that. Does that make sense from a, you know, like a first pass? 

Deep: Yeah, it does. Yeah. Well, let's walk through some of the details. So you start a new client, , at a minimum, you gotta get access to the GPS signals coming off of the vehicles. 

So you gotta plug them in. You've now got that, and you 

Richard: gotta Not always, not always. So most of the time, most of the time, to be really clear, and this is probably counterintuitive most of the time, we don't need GPS, we don't need tel what we call telematics in those, from those vehicles historically or not.

And I'll tell you why, is because we don't want it because it's very gappy. They lose signal a lot of times it's got massive errors in it. GPS jumps around, so it can be really difficult to clean. And it's compensated for by their other records, their proof of delivery timestamps typically. So we, well, let's, 

Deep: let's talk about the core minimum.

I mean, , yeah, sure. I take that back. My guess is to the very minimum. Who's going, where are they going? When did they leave? When did they get there? What did they have? And yeah, basically. 

Richard: Yeah, exactly. That's exactly right. The type of data that we actually consume the most is this combination of historical data.

So what did you do previously? 

Deep: Yeah. 

Richard: And then what do you wanna do and how do we simulate and create a digital twin to look at options for how you could go forward in a more efficient way? 

Deep: And do you know whether they got somewhere on time or were later? Early 

Richard: typically? Yeah. Typically you have, uh, again, in this industry, you would have a time window and if you're a parcel courier and you're getting a parcel from Amazon or something, usually it's a pretty wide time window.

It's like 8:00 AM to 7:00 PM or something like that. Whereas if you're like distributing food to cafes and restaurants, it could be way tighter. It could be, oh, you need to be here on my loading dock between one and 1:30 PM. And don't come at 12:00 PM 'cause that's my lunch rush. And you know, all, there's very tight time window expectations that are typically enforced on the delivery companies.

And we have access to all of that. And then there's usually a timestamp to say when did they actually deliver it? And did that meet the time window or not? 

Deep: Okay, so you've got this, I don't know what you call it, maybe historic delivery data. Historical data, 

Richard: yeah, 

Deep: historic delivery data. Okay. So that doesn't tell you who's available right now, who should take the next thing, what's your immediate problem is your first problem, like analyze historical patterns and try to figure out something from that?

From that? Absolutely. Yeah. Okay. So maybe we'll walk us through that first and then we'll get to the next problem. 

Richard: Yeah, absolutely. 'cause different archetypes and different industry verticals have different problems and their fleets run in very, very different ways. Okay. So that's usually the triage at the beginning is, okay, which one does this company fit?

Are they parcel distribution? Are they metro, are they regional or, suburban? Are they food delivery where there's a cold chain component and really tight time windows? Are they something else? And we've worked with companies, all different kinds of companies. Everything from garbage pickup and and garbage can delivery to recycling, container deposit, scheme delivery.

I mean, they're all quite different and they all have their own requirements that they need. So we've been at this for a few years and we now have a sense of how to put them into a bucket and say, okay, cool. Your core problem is gonna use. A type of model that we've already got, or we're gonna need to extend a model that we already have to take care of your unique, circumstances.

And then we'll work through that with 

Deep: them, them What's that first analysis you're doing though? You've got historic delivery data. You can look at transaction, you can look at delivery volume, maybe delivery volume by time, by location, maybe by vehicle type. And then you're presenting all the stats and then you're doing what with it?

Figuring out what, 

Richard: This is where it starts to get interesting is you look at all the historical data and you say, okay, cool. That's how you've been running. Then we have a bunch of optimization solvers, no ai really required for these more operations models, right? You then start to do some optimization experiments and say, okay, well how could you have done that day of work better?

Could you have done it better? And you always could, especially depending if they were using manual tools. So, okay, cool. What's the, from to, what's the ab comparison of how you did it to how you could be doing it? And then that's the key moment where they say, oh, okay, cool. There's a a minimally disruptive set of changes that I can make in my planning using ad's tools to be more efficient tomorrow.

Stuff that's easier to do. Like, like the reason I described that example earlier about the flower pedal routes versus the highly clustered types of things is out of the box without any kind of training. There's certain things that you can implement pretty quickly and get a fast return on investment and start to improve those delivery operations.

Right? And that's, how do I get more on time in full. 

Deep: Because you're looking at a visualization of the, of the patterns. And then you do a scenario, like what if we change the roots from this pattern to that pattern or these patterns to those patterns, and then you still don't have any of that other traffic data or any of that.

So you don't actually have a model yet. You have, I guess you have a model, but not one that involves those extraneous variables. So how do you get from historic, trajectory data with delivery and vehicle types to different patterns.

Richard: Yeah. Well then they have to start running the new routes, right?

Because again, garbage in, garbage out. So if we start to train models based on the stuff that they did historically, but that was inefficient, then we're gonna get bad models. What we do is create this new baseline to say, well here's how you can start being more efficient. And so, adjust your fleet to have this many vehicles instead of what you did before.

Have them at these distribution centers and you'll be able to increase your on time in full delivery rates by x percent. So typically, if you're a parcel courier and you're 85% on time in full, that's a reasonably okay number. But we have customers that are 99% on time in full and that's what an Amazon type level of service is, is gonna get you.

So that's just a non-AI related operations algorithm. Change that we can help them be more efficient straight outta the box. The real magic starts when they start running. Those delivery routes that we've prescribed using the kind of non-AI operations model. 'cause that's where they, we get the real good baseline.

Deep: So, okay, so I have questions about how you actually determine those suggestions, but I get the gist. We can kind of, you know, you've got a historic baseline. You have to a first establish the baseline because otherwise they're not gonna be impressed. So now you establish the baseline, I imagine a constraint to your model is like, figure out how to do this with less vehicles is probably something in there.

Richard: Well it's, it's a branch. It's a branching type of algorithm where it's looking for the lowest costs, like the mini and the maxima of the solution set. 

Deep: And what are some like, intuitive examples of the kinds of things you can find in that analysis phase? 

Richard: Well this is one of the reasons that we really focus on the minimum number of vehicles.

'cause there's, you can assign a cost. To any of these variables. And so cost per kilometer that you drive, cost per waiting time that you sit, waiting around in idle cost for the labor resource, cost for insurance, vehicle maintenance, all this stuff. And so all of that is, is built into our non-AI algorithms at the very front end.

Yeah. To make it more efficient. Right. And always adding a new vehicle to the fleet to do work is gonna be the highest single cost that you can encounter. 'cause suddenly now you've got to pay for either a contractor or your own vehicle, plus all the labor and all these other things. So the algorithms are really successful at minimizing the fleet required.

And then, basically that becomes your starting point is to what I call right size the fleet. You're taking what you did yesterday and you're saying, actually I can do more work or the same work with less vehicles. Let's do that first. It is based on payload of the deliveries, the weight, the volume, how much can you fit in the actual trucks that you have, those types of things to then come up with a new fleet plan.

And that's where they, the, becomes the input to the AI models. 

Deep: But that only works if you have the same stuff in or like a representative sample at least of the kinds of stuff from prior than, than you have today. . So you make that assumption and now you have a guess.

You go out and you maybe run the new guess for like a couple weeks or something, gather some new data and then repeat this process for a bit. And that's, 

Richard: That's exactly right. But let's unpack the key steps that make it better when you use ai? 'cause otherwise you're just gilding the Lilly, so to speak.

But the real differences that we see in the real world are that you can create these beautiful route plans that are a big jump forward from what they were doing before. There's still so much more to gain because people will, again, and organizations will again, have operational constraints that don't necessarily make sense.

And, and I'm not saying it's not our, role to pass judgment because these are big organizations that have a lot of resources and a lot of capital constraints. You know, they've invested heavily in warehouses and vehicles, and if we tell them you're gonna save X amount of dollars, if you get just, let's look at electrification, we're gonna, you're gonna save more money by getting rid and selling this, gas powered vehicle on the secondary market and replace it with an ev, they'll just reject that outright and they'll say, no, no, no.

I'm gonna sweat that gas powered vehicle until it dies, and then I'll consider buying an EV because they have an just emotional 

Deep: attachment to it or something. 

Richard: Yeah. It's just, there's a lot of. Change management that you have to go through as a big organization to get rid of a vehicle and replace it.

It's not like some simple thing like you can do if it's your consumer passenger car where you could just sell it and you could buy a new one. There's a huge process in the regulations and all these things. So it's not as simplistic to walk in as outsiders, as consultants and say, oh, just take all these steps and you'll be way more efficient.

They'll say, yeah, we're gonna go piece by piece by piece. And that's 

Deep: where, yeah. So maybe you're giving them smaller knobs and dials to turn than the ones that are too big for them. 'cause they maybe don't yet have the trust and confidence in you yet at this stage. 

Richard: Yeah. Well, it tends to, your question from before, it's like, so they start implementing some of the changes, the kind of low hanging fruit changes that we prescribe, but they're still gonna make irrational changes to our plans.

Deep: Oh, so they still, yeah, they're not even following your plans completely. 

Richard: Correct. And the drivers will do something different. You can't get a hundred percent compliance from the drivers, you can't get a hundred percent compliance from the warehouse managers, and nor can you always predict changes because we could make this beautiful route plan for tomorrow, and then one of the line haul semi trailers carrying all the parcels gets delayed in Omaha and it isn't gonna come in.

And suddenly all those plans are thrown in the bin and you have to regenerate them. So those are the types of changes as well, that if there are repeated cyclical problems, what our models can start to pick those up and then automate out those changes before the humans have to. Does 

Deep: that make sense?

Do you run into, yeah, I mean it does do you run into like social resistance? Like the drivers want you to fail because they know that at the end of the day you're cutting drivers out. 

Richard: I mean, it's not because we're cutting drivers out per se, it's because humans are creatures of habit. And if you've been doing something the same way for 1, 2, 10, 20 years, you're just always gonna be resistant to change.

You don't want somebody coming in and I don telling you that they know better. 

Deep: Don't turn that road. I don't like that road. I like this road. Or, 

Richard: oh, it's funny, like, you know, being in the industry now, you get resistance. You know, people say, well, I don't wanna do that route because it doesn't pass by the cafe that I go to for lunch every day.

Or I go visit my girlfriend and my boyfriend, you know, after work and this route ends across the city. But I usually end at the, you know, this part because that's where I go and see my partner or my family. 

Deep: Honestly, this kind of surprises me 'cause I, I sort of had this impression of the industry.

Maybe it's 'cause I'm in Seattle and we're so colored by Amazon, but that the humans were just, cogs in the machine and the machine was telling them where to go down to the left and right turns. I just sort of assumed that, that you were operating in that world, but it sounds like that's definitely not the case.

You're still operating in a world where you've got humans making human decisions. The joke around here is like the Amazon guys, you know, carry around water bottles in their van 'cause they're not even allowed to go pee. 'cause the bot's just telling them exactly what to do.

That does not sound like what you, you're talking, it sounds to me like almost what I envisioned from like 20 years ago. You know, where, where drivers back when I moved furniture in high school, even maybe like 30 something years ago. Like, I remember those guys were, they weren't gonna go anywhere that, you know, that they didn't feel like going.

And it definitely involved a lot of utterly unnecessary stops along the way. 

Richard: Yeah. And I did it too, you know, and I used to, I I, my first proper job was a paper route and I had to figure out how to deliver hundreds of papers through 30 apartment blocks in my city, and you know, was it efficient?

Yeah, it was fine, but like, it was all just in my head. And if somebody else came, I'll, I'll give you an example. Like, because you're right. Amazon colors all of our perception. I, there was this meme that even back in my medical device days, 'cause I did cardiology devices for 15 years and it was this meme of like.

I'm sure you've seen these types of memes, you know what people think a medical device manufacturing facility is like, and it's all robots and 

Deep: Yeah. 

Richard: People and you know, looking really clean and slick. And then it's like what a medical device manufacturing facility is really like, and it's, bins with stuff strewn over the counter and it's just the perception is always gonna be different than the reality.

And then when you extrapolate out over the different verticals and sizes of companies, now you've got an enormous variety. Because you could have a super high tech courier that has two vans. It's like him and his wife delivering stuff, but they're using every piece of kit and technology and they're automated things, right.

You're saying 

Deep: there's, there's like a spectrum here, but still like 17 years after the, it's 

Richard: trillion dollar market. 

Deep: 17 years after the smartphone. I mean, I'm kind of shocked that at a minimum you don't have GPS on all these devices and that people aren't running an app. And even if the driver just have a app in their pocket like that, that's telling them.

Richard: No, no, no. I didn't, I didn't say that. They don't have GPSI just said the GPS data is not necessary. Necessary. Gotcha. And it's often the harder data to use as the input to your model, because it's just hard to clean. Yes. I mean, in the, in they have to have GPS because of chain of responsibility laws, making sure driver fatigue and not exceeding miles and things like that.

Um, and maintenance for the vehicles and all that. So yeah, they have data available, but it's, again, it's an old school market. You know, these are people who 

Deep: Yeah, 

Richard: yeah. Technology wasn't their first thing. 

Deep: So now like, walk us through the full life cycle with the customer. So, okay. So now we got through, we started up a new customer.

We got those first few weeks, we got recommendations coming in. The humans are like taking 'em out for a spin. Maybe making some changes, like what's the next phase of evolution with Adona. 

Richard: Well then it's about then training and segmenting. So every city is gonna be different.

It's gonna have different, ways that we parameterize the models and the core things that we do to create this kind of hyper-specific model for a company in a region or a city is we have to look at what we call facilitators or enablers and and barriers. And those tend to be things that are hyper geographic specific.

So bridges, rivers other water boundaries that you have to go around 'cause you can't cut across 'em like a lake or something like that. Natural mountain range barriers, where highways run up and down. Is it a grid city or a not grid city? Those types of, very specific geographical barriers of things that, again, any good company that's been around for a while knows in their head.

And their allocating team and, routing team would be working around those things all the time. I've got a customer in Western Australia that won't send his drivers down a specific highway because at night there's too many sheep and he might hit a sheep, and people, their drivers have hit sheep.

And so if, a routing company like us comes in and just is looking at GPS data and mapping data And supply and demand, of course, if that's the most efficient route, we're sending that driver down that road. We don't know how many sheep there are. And there's, there's no data source that I've been able to find that has a sheep counter on it.

So imagine that at scale, every person having their 

Deep: and other types 

of issues. Yeah. 

Richard: Oh, there's thousands of types of issues. 

Deep: you're, you're basically designing a system that's complimentary to the human's intuition. 

Richard: Correct. Correct. Yeah. And because these people are in an aging workforce and they're retiring out, if you've got somebody that's got 20, 30 years of experience and knows where all the sheep are.

They retire. Good luck replacing that person without some automation and, and without ai. That's something that's held the industry back for a long time is that tribal knowledge transfer. 

Deep: So that sort of implies that almost everything you do is suggestive, you suggest and you're probably even compartmentalizing the suggestions and making them smaller scale where possible so that they can have option A, B, or C and we recommend A, but if you really wanted to do D, maybe B or C is good.

Richard: Yeah, you're absolutely right. It's always in consultation with the human. And the human always has veto power. So 

Deep: yeah, 

Richard: we're constantly giving the best suggestions, but on the day, if they have to deviate from that, well we're certainly not gonna be the ones to stop them. 

Deep: So one thing I didn't quite grok was like the carbon connection.

So I get the macro problem, but all the way down. Most of our listeners probably understand the global regulatory environment and the new carbon reporting, stuffs. But maybe if you can summarize that quickly, that'd be nice. But it sounds like , what level of control are you trying to give your fleet operators?

Are you trying to give them like, Hey, whatever you drive, you drive. It's all based on, efficiency and car reduction for the most part, but we'll report the carbon that you are consuming, or are you trying to facilitate them making more carbon reduction centric decisions in some way?

Richard: Yeah, a absolutely. It's both. So, , again, because of this diversity of different types of companies at different points in their journey, some of them are very big, but might still be relatively low tech. so they're gonna be on the hook for reporting, and that's where, as you mentioned, this regulatory change around the world led by the EU

but now, in, other markets like here in Australia and in the US and Canada companies of different sizes are being forced to report their emissions. And the emissions are broken into three buckets. Generally. Scope one, two, and three. Scope one are the emissions that you create yourself.

So if you're a manufacturer, those are things that, you are responsible for. But if you're not doing your own transport, those become scope three, meaning the fleet that's doing the deliveries on your behalf, it's their scope one. It's a bit complicated, but basically you have to report all the way downstream to anything that touches your product or service.

Deep: But like statistical sampling and guesstimation is not allowed because like this is very detailed reporting to get down to the vehicle, the type of vehicle, it's gas mileage, all that stuff. And then track that all the way through to, items delivered is very detailed. You could imagine a much cheaper maybe old school way of just sort of saying, well, I got 30 cars, they're driving 70% of the time they're Yep.

You know, they got this average gas mileage. Like, just put that number in 

Richard: Which a lot of companies do, but think about this, right? If you're a big company, and this is a real example, a real company that we worked with that is a telecommunications company and they sell a lot of mobile phones, right?

That's part of their brief if you're a big telco. And so they're shipping millions of mobile phones around the country. Okay, cool. They. Have a net zero target. They have a 2030 target. And in order to facilitate them to get to that emissions reductions target, they're spending a lot of money on carbon offsets and carbon credits.

And that's a huge market. When they go to their transport partners the actual companies with the vehicles that are delivering the handsets for them, and they ask what is the emissions profile for those handsets? They get back, basically what you just insinuated is like a really kind of back of the envelope 

Deep: Yeah.

Richard: Calculation that is sort of maybe accurate but really one of the bigger problems is that it's an average across all of their parcels because they don't, you know, they're consolidating in all these deliveries from different clients and then they're just saying, well, that truck that had all sorts of stuff on it, it could have had, pallets of Coca-Cola on it, which are way heavier than a, handset, a mobile phone.

Right? Yeah. But those two things are gonna get averaged in the truck, and then they're gonna get an average value that they're gonna then pay for in carbon credits. If I'm them, I don't want some average value. You're gonna lean, you're gonna lean the actual value. 

Deep: Yeah. Especially if you have, a couple of products that are in the same ballpark, maybe price wise and attribute wise.

that's interesting. Take on it. Yeah. Okay. But I'm still having a hard time understanding, like, down to your operator. what percentage of their calculus is efficiency oriented and what percentage of their calculus is emissions oriented

Richard: Yeah.

Look, it's a really great question and, I want to be really clear and transparent to people when we talk about these things. Most of these companies cannot afford to invest in sustainability. They don't have the margins. They don't have a bunch of capital sitting there where they can say, oh, we're gonna invest in a bunch of electric trucks.

Now obviously, some companies do that. Ikea, PepsiCo, these are great companies, very sustainability focused for their brand, and so they've chosen to invest a lot of money in emissions reduction, but that is the edge case. That is definitely the exception rather than the rule. Most companies, they don't have money specifically for sustainability, so that's the great thing that I get excited about.

What we do is we can create these financial efficiencies. If you're using less vehicles and traveling less miles to do the same amount of work, now you're emitting a lot less for the same amount of work too. And then they can get excited about the sustainability component, but it is very much financial first, emissions second.

Deep: That's a really fundamental point that you're making, that I need to marinate that a little bit. But it's not a decision between like, you got 30 trucks and buy a rivian, uh, an electric truck or something. It's like, no. Well, yeah, that would be nice, but if I cut out a truck period, then I'm doing even better than the electric Exactly.

Truck. Right. So, because that's, you got it. One less. Assuming that, and then you do drive by an electric truck. Yeah. 

Richard: You know, but then now you're in a better place because you understand, okay, cool, I've gone from 10 trucks to nine trucks, and now which of those nine trucks, now that I know I'm being super efficient, which of those should I replace with an electric van?

It gives you a much more efficient path and, companies that are really shrewd about it can actually then say, well, I've saved 150 K running this truck every year the 10th truck. And so now I, I can I actually apportion some of that budget towards these other things rather than just chalking it into margins?

Deep: maybe jump up a level for me and maybe jump out of the adona, CEO hat for a moment. 'cause I feel like you're have a really interesting perge. There's a myth, amongst a good set of folks that these, Carbon tracking mandates are just not effective. But it feels to me like you are in a position where you can comment on whether that's actually true or you're seeing that or not.

Is it true? Are they not effective? Are these folks who you mentioned or it can't afford sustainability, don't really care? It doesn't really change their behaviors much? Or is it the other case where the folks that are I think you mentioned Akia, if I'm not mistaken, that Yeah.

Where they're buying offset credits and they're forcing change down through their supply chain. 

Richard: Yeah. 

Deep: What are you seeing and how fast is it moving? Because, you know, the clock's ticking. I mean, we got, I don't know, five years, maybe 10 or something until that glacier breaks off an Antarctica and we're all screwed.

Richard: Exactly. Well, it, it's a di it's a very difficult thing to measure because it's just like your personal health, right? If you eat a cheeseburger every day and you get sick then you've got a confirmation bias that eating the cheeseburger caused you to get sick. But if you don't eat a cheeseburger every day and you still get sick.

You might assign causality to something else. It's a very difficult thing because it's so complicated to, to measure. And this concept of do the emissions regulations like reporting tracking, reporting, reg regulations work. I think they absolutely do, but it's just very difficult to understand what is causal from those versus what is causal from Other things, macro market movements and technology shifts.

Well, 

Deep: Maybe causality aside in your data, you should be able to put a timeline up and say what the net emissions are per kilogram of product moved. Something like that. And is that going down? I'm just curious. And how much is it going down?

Richard: Well, I can't tell you how much it's going down because of the net growth of e-commerce.

This is one of the biggest problems, like if you look at PepsiCo is a great example. Okay. And it, I'll answer two of you questions with this story around PepsiCo. So one of them is, the EPA in the US has a voluntary program where fleets, so any fleet owner can voluntarily submit their emissions data, meaning what kind of trucks they run, what the fuel type is, how old they are, and what the calculated emissions per kilometer and per kilogram are.

 And some of them have done that strategically because they've got better than average emissions for their class or their type of organization. And companies like PepsiCo actively go in and we'll pay more for transport that's more sustainable 

Deep: Okay. 

Richard: From those types of companies.

Now, again, what will be the net benefit? Well, Pepsi's got the biggest private fleet in North America. It's enormous. And so if they shave even 1% of their emissions off, that makes a tangible difference. To emissions generally. But are you gonna see that? Like, is it, is it easy to calculate? No. And then the kind of second part of 

Deep: the Pepsi, wait, I actually, I'm following you.

Why can't you just simply look at the emissions per kilogram moved. Are you arguing that because the net volume of products moved is a moving target that you can't compute that? 'cause I think you can still compute it. Yeah. Yeah. But, or are you just saying like, yeah, it's going down, but I don't know if that's the right measure because so much more stuff is being moved.

Richard: Exactly. Yeah. Thank you. That's e Exactly right. So I can measure it for that one organization and you can measure it for lots of organizations. But then, um, and the reason I brought up PepsiCo is because they've got these, they've got their own net zero goals, but their business, their revenue in, Southeast Asia, Vietnam, Cambodia, Thailand, Indonesia is growing like.

Some ridiculous rate. Right. 7% year on year or something. Right. Just 

Deep: So many people are being brought into the kind of western consumption way of life or whatever, so. 

Richard: Exactly. 

Deep: Tell, tell me, what do you see though? Like, okay, I, I'll take that caveat, but I still wanna know what kinds of percentages of efficiencies in both emissions and, I mean, I guess they're, the two are correlated, right?

Like, you're making that argument pretty well here. Yeah. But , what kinds of efficiencies do you see when somebody comes on? Is it like a couple of percent? Is it like five, 10%? What kind of, kilogram per you know, emissions, amounts per kilogram mo. What is it looking like?

Richard: Can, it can be enormous. So when we first started working with one of Coca-Cola's fleets here in Australia, one of our first real implementations of these models, we were able to reduce their average delivery time, like the on-road time, how much time they actually spend on the road delivery by 48%.

It was enormous. Oh yeah. And the total number of kilometers reduced. The distance traveled to do the same amount of work was 36% and the emissions were more correlated to that 36%, reduction. So same amount of work, same amount of product, delivered 48% reduction in time, 36% reduction in vehicle distance.

And they were able to then reduce the fleet size [00:47:00] significantly , to do that. So that's probably the most extreme that I've seen. But then it typically fall somewhere between and other end between five and 40. 

Deep: Five uhhuh. And how much of that is in those early phases where you're kind of coming up with maybe some of the obvious suggestions and what does that curve look like?

Like over a year, 2, 3, 4 years out? 'cause it seems like at some point you're gonna hit some diminishing returns. So I'm curious, how much of it comes, and how early. 

Richard: Yeah, no, you're absolutely right. It's a long tail problem. So you get a, an initial benefit that depending on how efficient they were or if they were using technology before what can be quite substantial.

And then they'll always be a long tail that as the AI model for them matures, they're gonna get incremental improvements. But it's not gonna be nearly as dramatic. 

Deep: My guess is the first few weeks or month or two, they're getting a lot of improvement. 'cause you're identifying a lot of the low hanging fruit from those human biases.

But is, don't you run a risk of them just saying, okay, we're gonna terminate your, our relationship now because we've figured out these new routes and now we don't need you guys anymore. , Does that happen? Or, and what do you do about that? 

Richard: No, that's, it's a great question because, we do have to focus more on what we call dynamic customers, meaning customers, and this is why the parcel sector is really good, is because you just don't know what you're gonna have tomorrow.

It's all gonna be different addresses, it's gonna be different locations. And, um, you also have massively fluctuating volumes throughout the week. And throughout the seasons you've got huge seasonality around, holiday season where volumes go bonkers after Black Friday. Yeah. And then they drop significantly in January and February and you know, you get the idea.

And then they're also across your seven days of the week, they fluctuate massively. And then if there's a holiday and you know, you know, there's lots of other use cases that using our models will allow them to maintain efficiency and a stable driver pool, which reduces a lot of the other kind of ad hoc costs they would go through.

Deep: Yeah. 

Richard: Keeps it sticky. 

Deep: So what we usually like to cover three things on the, on your AI injection, this podcast, one of the things is like what you do. I think we've covered that pretty well. Mm-hmm. Um, the other one is like how you do it. I think we've covered that, fairly well. The third thing though is should you do what you do?

And that sort of sounds odd but it's sort of coming at this ethical question of often like second order effects. So let's say you're tremendously. Successful at what you do, what are the like sort of ethical questions that you think about or maybe don't think about, but feel like you should think about and you know, what are those?

It seems like you have a few off the bat. If you're like incredibly exceptional, maybe, you know, we're down to like less human drivers, so that's probably one of your ethical, but like, what are they, what does that ethical landscape look like for you?

Richard: Yeah it's a great question because there are a few things that we've thought about heavily throughout the journey, and some we continue to and some not so much. When we first started doing optimization work, when I met my co-founder, he's a researcher, PhD in optimization ai.

And when we first started together, there were a lot of companies that were leading us trying to lead us down different paths. And they want us to optimize sales cycles. They wanted to optimize the placement of skews of products in a grocery store. But, um, things like that where we said that there's.

Only really a net benefit in increasing sales and profits for consumer goods. Companies don't really interest us. There's no sustainability benefit or social benefit from those, but logistics we gravitated to because, okay, cool. Now there's a sustainability benefit and there's actually a human element benefit too.

And let me just give you a really quick story because one, the first ride alongs I did, I went out on a truck with a guy and he was doing routinely 11 to 13 hour days, five days a week. And he hadn't been on vacation in three years. The reason he hadn't been on vacation is because he had such a complex delivery route with.

All these little details and post-it notes all over his truck like with phone numbers for customers and access to a door and not just the routing, but the whole thing was just so complicated that he couldn't get anybody to cover for him. They would never be able to actually do his route. And as a contractor and owner operator, it's his little business, it's his microbusiness.

Even though it may have a different logo on the side of the truck. And this is another thing that people need to understand is a lot of these guys out on the roads, in the Rivian trucks, they have an Amazon jumper on, but they don't necessarily work directly for Amazon for Yeah, they're contractors maybe, or, exactly.

And there's this whole network of Amazon delivery service partners, which are small to medium companies that have a fleet of drivers. So you have this abstraction from the company. So this guy, three years, no vacation, kids can't take his kids on vacation, 13 hour days, coming home tired. And I just had so much sympathy for him.

And how hard that was. And so it's a big piece of what we do, looking at the ethics around can we improve the human lives of every touchpoint in this chain? of technology that we create. And that becomes a really guiding principle for us. It's part of our North Star. If it's not influencing emissions reduction and it's not gonna make those people's lives better, we don't typically wanna do it.

It's just not in that ethos Now. You asked a big question too around that second order effects and one of the 

Deep: ones that probably goes Before you get to that, I just wanna like, 

Richard: yeah, sure. 

Deep: Kudos for that. Because like if Zuckerberg or some of these other wankers had thought this way, maybe we'd have like way less problems with our, 14-year-old kids and their suicidal ideation and the shithole democracy.

Mess that these guys have created and these social media companies. You know, I think it's rare the fact that Zuckerberg didn't let his kids on Facebook and Jobs didn't let his kids. Yeah. Doesn't Havet tell all something? Yeah. Don't, 

Richard: that's 

Deep: everything 

Richard: we need to know. 

Deep: That is everything we need to know.

I think on some level and the fact that they just, that he just looked the other way when a million, when rumors were running rampant and Burma and people were getting murdered. Like, he's like, I don't know. Not really my problem. 

Richard: Not my 

Deep: problem. My pet peeve within the tech industry is, you know, everybody is sort of looking at like, I mean, not everybody, obviously you guys aren't in this bucket, at least not based on my conversation with you, but it's, there's a natural tendency to sort of view everything as a morally agnostic hammer and engineers are sort of naturally wired this way, largely.

We're just intellectually curious. We just wanna build some cool new thing. We don't think of second order effects. And I think like, going out of your way to like define a clear north star both carbon and human wise. I mean like. That good for you? You know, like, that's pretty cool. Thanks. That's pretty cool.

So anyway, 

Richard: thanks. I appreciate that. It is something that, , is also the benefit of age, I guess. You know, this isn't my first business and it isn't my, first career even. So I think, the older I get, the more in tune I am with technology, not just for technology's sake, but also having a really well thought out delivery plan and transition plan for the humans that you are using.

Technology, I find to be so much more important. 

Deep: Yeah. I'm more fulfilling, right? Like we have limited time on the earth if we can spend it, improving lives either directly or because, we all get to survive on the planet. Like, either way. That's good. 

Richard: Exactly. Well, exactly. Like if I'm gonna, I'm lucky to have choice to do what I want now.

Deep: Yeah. 

Richard: And so if I'm gonna have choice, I'm gonna exercise it in a way that suits my ethics. 

Deep: Yeah. Well, I think I feel like you, oh, sorry, I interrupted you. You were about to make a point and 

Richard: Oh, I was just gonna say that the one thing that still worries me that I haven't fully come to grips with morally with what we do is, does it accelerate e-commerce growth even more because we're, enabling these companies to be more efficient, they can make more money, and therefore consumers can just order more crap from Temu.

Yeah. Yes, arguably, we are problematic in that, and I'll be honest about it, 'cause it does bother me, but at the same time, we're not the pioneers, shall we say. Like, we're not Amazon. Amazon is really driving this change, and you can't put the genie back in the bottle. And I'm not saying, or the 

Deep: phone is driving it,

I mean, you know, like at the end of the day, , Amazon's reacting to what consumers want, and consumers want to wiggle their fingers on a phone and have something show up their doorstep, you know, ideally, you got it. 

Richard: You got it. 

Deep: Right, exactly. So, I mean, I don't know. I feel like yes, that's an ethical quandary for the space, but that the test that I use is like, if you are not doing this and nobody's doing this.

Is it better or worse? And it feels unambiguously worse, so that seems like you're doing something right. 

Richard: Thanks for that. Honestly, I've never distilled it down to a really clear sentence like that, which I think is 

Deep: really wise. I think that's, yeah, I mean, I don't know. Like it's pretty straightforward, right?

Like we all operate in an ecosystem and we can't be completely responsible for the entire ecosystem. That's meant to say we should wash our hands of it and not play , our individual roles to try to impact and influence the ecosystem. I mean, absolutely we need to, but I think it's easy to just not be able to get out of bed if you are like, overwhelmed by the ecosystem, right?

Yep. Like you have to,, you have to also have some level of realism and be like, yeah, that's actually outta my control. Like, people are buying stuff, people are delivering stuff. How am I gonna cut emissions and improve life quality within that context? 

Richard: Yeah, well it's, it comes down to almost the, that concept of, allowing me to control the things that I can control and accept the things that I cannot control.

Deep: Yeah, and I think 

Richard: guiding principle, 

Deep: and I think probably, you know, when I look at your business, the biggest ethical quandary I would have is the job loss reality. I wouldn't know how to wrap my head around that exactly, but, you know, and I think like a lot of these threads, like they're, when you pull on them, they're kind of complicated, right?

Like you said, you're in an industry that's understaffed right now. You also have the reality of automated stuff coming out, you know, that's gonna happen with her without, Adona and so, I don't know. I mean, I think that's where we're all sort of like, you know, I was on a guest on a podcast and you know, was accused of being a techno optimist and I'm like, really?

I feel like I spent all my time being a techno pessimist, but Okay. Because, and it was really came down to the fact that I sort of accepted the reality of the ecosystem that we're operating in and a lot of these folks don't. So. Yeah. But I don't know. Anyway, it's been awesome having you on the show.

Uh, you know, I feel like. I've learned a lot about, supply chain management about a company. A bunch of friends of mine used to work at, I don't know if you know Convoy, but they did something kind of similar in this space. Yeah definitely unfortunately the biggest blow up disaster in Seattle, I think VC history, so I'm not sure what the full story is there.

One of these days I'll get one of my buddies on the podcast to explain what the hell happened there. 

Richard: You need to postmortem on that. Yeah, I would 

Deep: be 

Richard: curious about 

Deep: that too. I, I do not know the answer to what happened there, but it, it, everybody was, they were like getting really late stage money and we thought it was a billion dollar, like a unicorn for sure.

I mean, they were raising, I think they last raised was at a 10 billion valuation. Somehow the whole thing still blew up, which usually doesn't happen like once you make it that far. Not at that stage. Not at that stage. Right. Like things have been vetted. So I don't know what happened exactly, but. But anyway, you know, I remember those guys making this argument, but I didn't have time.

We were drinking beers and talking about other stuff, like to dig in. But I do feel like I get it and I think I think it's cool. So anyway, thanks so much for coming on the show. I always end with one last final question. Like, what's your prediction five, 10 years out, however far out you're comfortable going, what's like, the good forecast for the world and what's the apocalyptic forecast for the world?

Uh, with respect to, you can take it up a level or two AI in general, but maybe anchored in your worldview, 

Richard: Yeah, that's a great question. I think the optimists in me is also gonna predict that we are gonna go through a massive pain in the labor pool. That bottom rung is being ripped out.

Including in our industry, in supply chain. It'll make things a little bit more difficult, but it'll put a premium on the physical world environment. Again, I think actually it's a good thing. The fact that people are being ripped out of the digital world by AI are blocked from it almost. 'cause it's gonna put them psychologically back into a much more tangible place and look at the world around them and say, wow, actually, , if all of these opportunities in junior level accounting and IT and software development are no longer available to me, what am I gonna do with my life that AI can't do?

And suddenly they have to look at the natural world. So the, that real me. Well, that's a 

Deep: really, that's an intriguing comment there. I'm gonna think about that one because I think marinate 

Richard: on that. 

Deep: Yeah. I mean, honestly I hear this from a lot of angles, like, my son's a a finished carpenter and he and his girlfriend have a business like, fixing high-end sailboats, like with their carbon fiber stuff.

And at first I was like, ah, you know, go back to school more degrees, blah, blah, blah. You know, it's such a wonderful life staring at a computer all day. And then I'm like, no, you know, like, actually I think this is probably the last dominion of AI's going to get to, like, it's a far, he's a long ways away. So then I'm like, okay, keep doing, do it.

It seems like the, like, but that's all. Within five years I've had to reassess my, you know, and now I've. I mean, I've got these kids coming, like friends of friends whose kids got, you know, a master's degree in CS and AI and ML from great universities, great degrees, great grades, can't find anything.

And I'm like, what is going on? So, you know, like we're going through that shift already. Maybe those have been around for a while. Don't feel it as directly. But yeah I think that's good inspiration. The other riff on that and that we do have to wrap up, but I'll just say like, me personally, I play a lot more guitar now because I'm like, nobody's gonna automate a flamenco guitarist.

'cause nobody wants to say robot playing flamenco guitar. Like no one on earth cares about that. 

Richard: People wanna see it once, just out curiosity, but they don't wanna come back and see it. 

Deep: No, no. They'll hate it immediately. 

Richard: That's the holograms. Remember like everybody was like, oh no, we're just gonna be looking at hologram singers from now on after the Tupac thing.

And then of course it was a fad that lasted a few months and nobody cares anymore. 

Deep: Yeah. Awesome. I think we can end on that note. I didn't get your nihilistic take, but that's probably okay. Maybe we'll call this an optimistic episode. 

Richard: It's about how deep the pendulum swings into the bad bit before it swings back.

That's typically my version of pessimism is I always think it's gonna swing back to the good side, but it's just how deep of a cut will we get , when it's in this mid-swing. 

Deep: That's that. That sounds reasonable.