AI is reshaping the logistics world, from route optimisation to fleet efficiency, demand forecasting to customer satisfaction. But beyond the hype, what does it actually mean for retailers today?

In this practical conversation, Satalia CEO Daniel Hulme joins science & tech communicator Maren Hunsberger to unpack the role of AI in grocery delivery fleets: where it drives real impact, where retailers are currently falling short, and what the future could look like…

MH: How do AI tools help us overcome problems, especially in things like the last mile and these really complex and expensive parts of the delivery process?

DH: First of all, we need to acknowledge that this problem is very important. Customers are expecting not just next day delivery, but now same day delivery. And AI has a massive opportunity to be able to address lots of frictions across last mile delivery to make sure that you’re fulfilling that promise to the customer.

So maybe we can start out by reminding ourselves what the fundamental delivery problem is, which is if I have 24 points on a map that I need to deliver packages to, there are actually more possible solutions or more possible routes around the map than you might imagine. In fact, if you had a computer that could check a million routes a second, it would take 20 billion years to go through all the possible routes around 24 points and say, “This one that I looked at 10 billion years ago, this one’s the shortest one.”

And if you add another point to that map, it becomes 25×24 billion years. If you add another point to the map, it becomes 26×25×20 billion years. So, these are what are called exponential problems. And of course, companies don’t have billions of years or gazillions of years to solve these problems. They have seconds or milliseconds. And they’re not just delivering to 24 points. They’re delivering to sometimes thousands, tens of thousands, or in some of our clients’ cases, hundreds of thousands of points on a map in a day.

So, first of all, you need to solve the fundamental delivery problem and then you need to start to solve problems across that landscape to make sure that the decision is the most effective it can be.

MH: And we have so many variables to consider here, right? We have maybe sustainability if that’s a factor for the client. We have delivery window, optimisation of the route, so many things to consider. How do AI tools help us consider all those variables?

DH: Yeah. So maybe again to geek out a little bit, there are broadly four optimisation problems you need to solve for in last mile delivery. So, if you go to a, you know, retailer’s website and you fill your basket and you say, “Show me the slots that you can deliver to my house tomorrow or next week.” You have a few seconds to calculate the slots to show the customer. That requires deep, deep optimisation algorithms to calculate the routes, to figure out: “can I deliver to you at 9am, 10am, 11am across different days?” That has to happen in milliseconds.

Once you’ve chosen a slot, you then have a few seconds between the next customer coming in to optimise those schedules so you can show more slots to the next customer. Sometimes there are unfortunate events where vehicles or depots go down where you have to re-optimise very quickly lots and lots of deliveries.

And there’s a concept in some retailers called post cut-off, which is, during the day you want to be showing as many slots to customers as possible. And then after you’ve stopped showing slots to customers, you want to optimise the balance of those deliveries across vehicles so that vehicles are leaving the depots in a staggered way. They’re coming back in a staggered way.

So those are kind of the four key optimisation problems and there’s lots of machine learning problems to solve. How do you calculate how long it takes to get from A to B at a certain time of the day? How long is it going to take to deliver to this customer, based on the number of pallets or the type of house that customer lives in? Even driver behaviour—all the drivers behave and drive differently. So, understanding how to profile drivers. There are many, many opportunities to apply AI to solving these frictions, making it more optimal.

But what you’re ultimately trying to do is you’re trying to reduce the amount of cost, i.e., number of miles that a vehicle drives, which has a massive carbon improvement. You’re trying to unlock more capacity from your fleet to be able to deliver more stuff. You’re trying to make sure that the drivers have breaks at the right time to comply with regulation, and you’re trying to also make sure that those drivers are coming back on time so they can see their families.

MH: It sounds like an almost impossible problem to approach. How are people doing it now and how will it be different if they were to integrate AI-driven tools?

DH: There’s been a slew of technologies that have been developed over the past several decades and you know there are solutions out there that I would argue probably do, you know, 80% of what a client needs. What we’ve found over the past several years in Satalia is those clients have had to tune and adapt those solutions and ultimately those solutions are not solving 100% of what the client needs.

So where Satalia really saw the gap in the market was: can we model 100% of the client’s problem? Can we understand all of the constraints and considerations that a client has when delivering packages and can we build algorithms that can then solve that problem in a way that’s differentiated? And the answer is yes. We’ve built solutions for Tesco, for DFS, for Woolworths in Australia, for many organisations where we’ve built custom solutions that have really moved the needle for those organisations, reducing the amount of carbon, unlocking more capacity. But then we’ve also built reusable assets that we can combine together to be able to serve new clients.

And as far as I’m concerned, we’re sitting on perhaps the best last mile delivery solution in the world. And the beauty about this technology is you can compare and contrast. You can get your clients’ schedules. You can run those schedules against the technologies that we have and we can show that our technologies are superior.

MH: Amazing. And what kinds of improvements are we seeing when these clients have implemented some of Satalia’s solutions?

DH: We’re seeing dramatic improvements. With most of the types of projects that we do, we typically reduce the amount of carbon by 10, 20, even 30%, which is a huge opportunity for organisations, and it really also empowers and excites the team because we feel like we’re having a material impact on reducing carbon. As I said, one of the big challenges is how do we make sure that the drivers are delivering and making sure that they’re getting home on time to see their family.

You know, a lot of organisations are using antiquated solutions or they’re trying to solve these problems on pen and paper, and I guess when you’re going through the journey of trying to adopt and embrace these technologies, if you’ve already got an existing solution then the problem is: how do you do a heart transplant? How do you replace the existing heart with a new solution while keeping the wheels on and the vehicle moving? If you’re sitting on solving this problem on paper then there’s usually a much bigger opportunity to make an impact because algorithms can usually solve these problems significantly better than human beings.

MH: Yeah, that transition is always going to be an interesting time.

DH: Yeah. And the transition also requires you to really consider change management. You know, you’re affecting people’s lives that are in depots that are driving these vehicles. And so, you know, Satalia’s approach is to not just engage with senior stakeholders that typically care about reducing costs, unlocking more revenues, reducing carbon, but also you’re enriching the lives of the planners, making sure that they can now create multiple scenarios rather than just one.

And that we’re also enriching the lives of the delivery drivers that now have much better working arrangements and experiences. So we make sure that we engage with all of the different stakeholders across that supply chain and we take them on the journey to understand that these algorithms can actually make the solutions better not just for the business but for them, and I want to dive a little further into that difference between some clients that may already have an existing very large historical database that they’re working with and some clients that may be transferring over from manual or paper-based systems.

MH: What’s the difference there and how can something like Satalia implement solutions for both of those kinds of clients?

DH: So I guess for more mature organisations where they’ve got an existing solution, I think that the reason why they’re looking for an alternative solution is because their existing solution either can’t take into consideration all of the different complexities that they need when planning, or it’s not suitable to allow their organisation to grow to the next level to unlock more capacity given the fleets that they have. So they go to market and they look for differentiated solutions like Satalia.

And again the nice thing is you’ve got an existing set of plans that you can then run across our solution and show a differentiation. For companies that are still doing this on pen and paper, again that’s not a bad thing. I think ultimately we see companies investing in the wrong technologies and then having to remove and replace those solutions in a few years’ time.

MH: That’s almost harder than going straight from paper and pencil to something new!

DH: Exactly. Going straight to Satalia, hopefully! But you know the challenge there really is creating a solution and training those people to use something completely new and also getting them bought into the idea that AI and algorithms can typically create solutions better than human beings.

You have people that have been doing this job for 10, 20, 30 years, and it’s hard for them to imagine that a technology can actually make these decisions better than them, but that’s a journey that we’re used to taking people on.

And what happens is we empower them not just to create single plans, but multiple plans that might benefit the customer or the business or the employee. So it turns them from being tactical people to strategists, and that’s a journey that we’ve seen many people go on and enjoy.

MH: Empowering I think is the key word there. Now we’ve talked a little bit about the carbon footprint improvement that we can see with some of these solutions, but I want to talk specifically about electric vehicles because we’re seeing a lot more integration of electric vehicles into fleets. What are some of the challenges that distributors face when they’re trying to integrate?

DH: Well, I think electric vehicles do unlock lots of opportunities to reduce carbon, but one of the clients I can’t talk about, unfortunately, is that we’ve not only solved their last mile problem, but we’ve solved their middle mile problem, which actually I’d argue is significantly more complex than last mile. And from what I understand from that project, we have managed to reduce the equivalent carbon to three long haul flights a day. So, we can now optimise those schedules that are reducing the amount of carbon equivalent to three long haul flights, which is thousands of long haul flights a year, which actually has a material impact on carbon reduction, which is something we’re incredibly proud of.

The transition from petrol vehicles to EV vehicles again unlocks more capacity, more opportunity to reduce carbon, but it also introduces much more complexity. When you’re dealing with electric vehicles, you have to consider different types of routes.

So, for example, electric vehicles charge when they’re going downhill, but their battery really runs out when they’re going uphill or steep inclines. And so, you would send a petrol vehicle on a different kind of route than you would an EV. There are also lots of interesting ancillary problems associated with electric vehicles like driver behaviour, but also infrastructure planning.

So, where do I place my depot? Where do I place my stores to make sure that I’m maximising the capacity of those electric vehicles as well? Those are big decisions that companies need to make before they even solve their last mile problem. And we do a lot of work with companies where we do infrastructure planning—using our technologies to run scenarios about what the optimal fleet requirements would be and optimal infrastructure to maximise and service the needs of their customers.

MH: I had no idea it would be so complex. I think when you think about it, you think you should replace a gas vehicle with an electric vehicle, but there’s a lot that goes with that.

DH: Yeah, absolutely. I mean, it introduces another level of complexity. Also, a lot of the decision-making that you need relies on accurate map data, and we work with a number of partners that give us an understanding about the types of roads and the road restrictions. We also have to work with our partners to understand the nature of those vehicles, how they perform in different scenarios.

So every minute that we’re able to save on a delivery translates typically to tens, even hundreds, of millions of dollars or pounds of savings for our clients. And if we talk about machine learning, one of the projects we’ve done for some of the clients I can’t name is they historically would have algorithms or rules that would try to predict how long it’s going to take to deliver to your door. So based on your postcode, based on the number of pallets or bags that I’m delivering to you, how long do I think it’s going to take to deliver to your house? You might predict, like, 7 minutes. We can use more accurate algorithms, more interesting data sources to now make much more accurate predictions.

And again, if you can shave off a minute from those predictions, then it does translate to literally hundreds of millions of dollars of savings.

MH: And then even aside from the cost, how does using an AI tool to help integrate a fleet with more electric vehicles—what kind of impact does that have on the client in terms of a sustainability outcome?

Yeah, I think everybody—it’s not just the business, but all of the people, employees, customers—are conscious and conscientious now about the impact that technology, or the impact that these deliveries, have on the environment. And based on our experience over the past almost two decades, we know that if we apply the right algorithms to the right problems across supply chains, if we help organisations adopt new types of innovations like electric vehicles, we can significantly reduce the amount of carbon, allowing us to actually get control of our ecosystem, which is a very exciting opportunity for AI.

So when talking about the complexity of this problem, it occurs to me that creating a custom version of this for each new client would be an incredibly difficult and expensive process, maybe, but how has it been overcome when customising different solutions for their clients?

So I guess what we don’t want to do is fall into the same trap as some of our competitors where they built a solution that only really serves 70 or 80% of the client’s needs and then they can’t customise it enough to solve 100% of their needs.

So again, we’ve built assets that allow us to 100% solve the problem and make the solution operate in a much more flexible way. What I mean by that is that customers or clients recognise that last mile is a key touch point with their customers, and that touch point might be fulfilled by contractors. It might be fulfilled by their own delivery fleet. It might be fulfilled by drivers that have certain characteristics. Or it might be fulfilled by vehicles that have different characteristics like electric and petrol.

And when you start to take all of these different considerations into account, the problems become exponential. Fortunately, we’ve built assets and technologies that allow us to solve that problem at scale, making sure that you’re satisfying not just the business needs, but the customer and the employee needs.

MH: Sure. So, is it accurate to say that it’s a little bit of a plug-and-play—like we have this tool for this problem, this tool for this problem, this tool for this problem—and if a client has that suite of needs, we can apply that suite of tools.

DH: Indeed. And to forgive me for geeking out a little bit, but let’s imagine you had a rule that delivery drivers only do a certain amount of deliveries on their birthday. Now, to build that constraint into an off-the-shelf or an existing solution might be very tricky. We have built technologies that allow us to create these constraints in a very rapid way. It actually takes less than a day to build these constraints and considerations that allow us to customise these solutions 100% to the client’s needs.

So all of those nuances, all of those complexities, those rules that might be different from one company to another, we can now take into account and we can also solve at scale.

MH: Yeah, it allows you to be incredibly versatile. Talk to me a little bit more about the middle mile. So what is it and why is it so problematic and how can we apply some of these AI solutions?

So last mile is typically delivering goods from stores to people’s homes and they’re usually homogeneous vehicles—vehicles that look the same and are able to do a certain number of deliveries—and the drivers tend to be quite homogeneous as well. When we look at middle mile you have these large HGV lorries that might have a chilled compartment, a fresh compartment and an ambient compartment that can also change in terms of size. These vehicles have to drive over many hours or long distances.

So the time, or the traffic on a particular road in one part of the UK, is going to be different according to the time of day, which you have to take into account when planning. There are also many more road restrictions. You’re not allowed to drive HGV lorries around different parts of cities during certain times of day. There are council restrictions. So whilst the number of vehicles is typically smaller than last mile, the complexity—the rules associated with middle mile—is significantly more. Middle mile is moving goods from depot to stores.

So last mile is stores to depot; middle mile is goods from depot to stores. And the complexity of these types of problems is significantly more. There are companies that we’ve solved this problem for where we’ve built assets that we can now repurpose for other organisations.

MH: And are you guys looking at working on applying these solutions across the entire supply chain, not just the last mile?

DH: Yeah. This goes back to the concept of a digital twin. If I’ve modelled and solved my middle mile problem and I can understand accurately what’s going on across my middle mile, and I can maybe see that I’ll have a certain demand on a store at a particular time of day, I’ll then be able to project that onto my last mile solution. So I can start to connect my last mile and my middle mile solutions and then build a much more intelligent co-op to my solution.

And I think that’s really where the exciting opportunity is. With DFS, which is one of my favourite clients, we’ve been solving problems across their supply chain for over a decade where we do their last mile delivery and their middle mile delivery. We predict footfall in their stores and we allocate staff in their stores. We do supplier confidence prediction so we know when goods are going to be coming into their stores. We allow their suppliers to book slots into their stores.

So when you can start to get visibility across the supply chain, we can then create these digital twins and simulate: what happens if this supplier defaults on their supply? What does it mean in terms of the number of people I need in this store or the number of drivers that I need on this day to fulfil that promise to the customer? And really that’s the power in stitching all these technologies together—being able to run those scenarios and adapt your organisation to a rapidly changing world.

MH: So when we talk about creating a sustainability impact through the integration of EVs and optimising them with AI, whose responsibility is it to make that integration happen and to oversee that transformation? Do you see that as a private industry responsibility or should it be more government-led?

DH: Well, you know, we’re seeing the government setting mandates to make sure that organisations are becoming much more sustainable, minimising their carbon impact.

I think organisations have made promises to the market, to their customers, that they’re doing that, and we’re seeing organisations go on that journey. When I engage with the employees in those organisations—and certainly when I engage with my employees—they are really motivated to use these technologies to reduce the amount of carbon. I think ultimately it’s everybody’s responsibility: it’s government, it’s business, it’s employees, it’s Satalia—to make sure that we’re using these technologies in a way that reduces the amount of energy required to create and disseminate goods.

And I hope that if we do that really well and continue to do that for our clients and create digital twins, we’re able to not just unlock more capacity, but reduce significantly the cost of creating and disseminating goods which then, you know, gives people access to more abundance that hopefully enriches their lives.

MH: And what I see here is maybe the opportunity to reduce the cost of taking more sustainability measures, which makes that more accessible to more parts of industry everywhere?

DH: Absolutely. I think we typically solve problems that have never been solved before, and that might be many months or even a multi-year journey where we’ve built an innovation that really moves the needle in solving things like a last mile or a middle mile or a warehousing problem. But the nice thing is that once you’ve solved that problem we build an asset that’s reusable, so it can be used to benefit other industries—hospitals, or routing ambulances. One of the things I’m really passionate about is not just sitting on the innovations that we’ve developed but leveraging them to actually benefit other industries. When we talk about these AI logistics solutions, are they typically cloud-based? And if so, how do we keep things like customer addresses secure?

The first project that we did with Tesco was helping them build their last mile delivery solution, which was one of the biggest initiatives that they’ve ever taken. That was a cloud-based solution. Historically, a lot of these last mile solutions have their own hardware. But that was the first ever solution, I think, that Tesco built at large scale on the cloud, which is overwhelmingly successful.

Our solutions are typically cloud-based because they have to scale according to the needs of the client. You know there are demands during Christmas, there are demands during peak seasons where you need to ramp up more compute. And when deploying technologies on the cloud you have to make sure that things are secure and safe and all that kind of stuff. But again, those are problems that we’ve solved, and our clients are very, very happy.

MH: Can you detail that out for me a little bit? Like, what does that security look like?

DH: When developing these technologies for clients, you can make sure that they’re multi-tenanted or even ensure that the data is not kept on the same virtual servers as others. We also make sure that we tend not to use PII data. So of course you need to have lat/longs or addresses, but we don’t need to have any information about the customers—their names, etc. We just need to know where to deliver, how much the weight is, and we need to make sure that we fulfil that promise to the customer.

MH: So the least amount of personal information that’s necessary to be able to implement the solution?

Yeah, absolutely. In fact, most of the data that comes to Satalia is either anonymised or encrypted.

MH: I don’t think most people would think that that was the case; that you don’t need someone’s extremely personal information to enact these solutions?

DH: No, indeed. I mean, you could argue that if you just had somebody’s postcode and if that’s the only house in that postcode, then you can isolate that person, but that’s a very rare event and, as I said, typically we use lat/longs anyway. So that’s a problem that we don’t really face.

MH: So it sounds like we’re already in a really exciting place with it, but where do you see these innovations going in the next 5, 10, 20 years?

DH: When we’ve developed our algorithms and infrastructure, we’ve really tried to make it as future proof as possible. We envision a world—timing uncertain—where you have hybrid autonomous or non-autonomous vehicles that can be reconfigured to solve different problems in different parts of the day. So you might have a vehicle delivering a certain type of goods during the morning that might be different to the evening.

And how can you have vehicles that can do different things that are autonomous, that maybe even can come together while they’re driving along motorways or roads, and share goods and distribute goods? So you have these mobile warehouses. We’ve factored in pretty much all of the different kinds of flexibilities and considerations in a world where you have the ability to have mobile warehouses that can distribute any type of goods. And we think we’ve solved that at scale.

So, what I’d like to see over the coming years is not just the migration to EVs, but also a migration to autonomous vehicles, and making the most use of those vehicles. There’s a concept in logistics which is we want to minimise the shipping of air. So if you’ve delivered some goods, how can you then use that vehicle to collect trash or rubbish when it’s going back to the depot? That’s what’s really exciting to me—identifying those gaps, identifying the inefficiencies and using creativity to actually add more value.

MH: Yeah, it’s exciting. It’s envisioning this future where all of these futuristic solutions come together to amplify each other’s effects. So when enacting these solutions, how do we measure the success?

DH: Traditionally, when deploying these types of technology, you would focus on one KPI which might be cost reduction or unlocking more capacity—more deliveries. I think we’ve recognised over the past almost two decades that if you focus on one KPI and you build the right solution, you can move the needle significantly in that KPI, which can sometimes cause problems or harm elsewhere across the supply chain.

So when we’re looking at building or deploying solutions, we don’t just take into account the business needs—money, revenues, costs—we also take into account the employee needs, making sure they’re getting back to their families on time, having breaks, even having breaks with their friends. We also take into account the customer needs, making sure that we’re fulfilling that promise to the customer, and even the environmental needs and supplier needs.

So you’ve got this entire ecosystem of KPIs to consider. When we build solutions, we allow planners to tune and optimise across those different KPIs. So if, for example, over the next three months there’s a real need to engage and enhance customer experience, then we make sure we’re always fulfilling that promise to the customer. If we want to retain staff to reduce churn or improve employee well-being, we make sure we’re solving for that potentially more than the others.

The nice thing is that AI will typically lift the needle in all of them. You’re not necessarily negating one to improve another—it can elevate all of them. It’s just a matter of which one you want to elevate more.

MH: If Satalia has a client whose only priority when they come to you is decreasing cost, how does Satalia model for that client that we also need to be considering these other areas that we are going to look at?

DH: We typically haven’t had that situation where a client is only focused on cost reduction. There are usually other considerations like sustainability and customer satisfaction. I think companies recognise that they have lots of different stakeholders that they need to manage, and if you just focus on one then you’re going to disappoint others. So I think our offering, our solution, the way that we work, the way we think, really gels with how organisations are trying to grow their businesses.

To speak to a Satalia expert on solving your delivery/logistics problems, get in touch here