Kale AI on last-mile cargo bike operations and a greener urban logistics future

A look at how one startup is supporting sustainable and resilient city design by optimizing urban delivery logistics — starting with dispatch apps designed for eco-friendly operators.

In 2021, Esben Sørig, Nicolas (Nico) Collignon, and Soonmyeong (Chris) Yoon founded Kale AI. Their mission: to support sustainable and resilient city design in the face of climate change — starting with urban delivery logistics. We caught up with Esben and Nico to learn about their journey so far. The following captures snippets of our conversation, which has been edited for length and clarity. 

To get us started, can you tell us a bit about the Kale AI origin story?

Nico: We’ve known each other for about 10 years and always knew we wanted to work on something together. After finishing my PhD in computational cognitive science, I joined the cargo bike logistics industry at a startup operating 100+ bikes. That experience helped me realize two things. First, cargo bikes have incredible potential, they’re incredibly efficient compared to traditional vans, and often misunderstood. Second, it is important to help cities move away from more traditional car and van logistics. I noticed that software and technology was a big hurdle for operators to scale and grow and about a year later, I connected with Esben and Soonmyeong to start our journey to Kale AI. 

You mentioned cargo bike logistics are often misunderstood. How so? 

Nico: Operators have been using vans in cities for a long time. So the idea that a smaller, manual vehicle can be used in place of a larger vehicle with a motor is a big cognitive leap. Dense urban centers are extremely congested. Some statistics show that van drivers spend more than 50% of their day walking or with the van parked. This means that even if you have a vehicle that can go up to 120 kilometers per hour (kph), the average speed is around 10 to 12 kph. So you have vehicles that are being used inefficiently and at subcapacity. 

Instead, if you use a vehicle that’s a lot smaller, more nimble, and able to use different infrastructure such as bike lanes, you can see a lot more performance. But that requires a shift in your organization and how you think about logistics infrastructure. 

Esben: To add to that, a cargo bike basically has no parking issues. You can just park it right outside the door. So the whole process of finding parking and walking to the door is basically eliminated. 

Nico: You don’t have a lot of people looking at how vehicles operate, park, and navigate in cities. So we’ve been working with researchers at MIT in Boston, ITU in Denmark, IIT in India, and the University of Arizona to use machine learning to model the performance of these vehicles in cities and get a better understanding. 

What is it you’re building and who are you building it for? 

Esben: We’re starting by building out a dispatch and routing system (plus rider apps) called Cavolo (yes, we like our veggies) to help put better routes in the hands of operators using hybrid fleets in cities around the world and support their logistics flows. Our primary partners are last-mile operators in cities such as PedalMe in London and Urbike in Brussels, in addition to Copenhagen-based cargobike manufacturer Larry vs. Harry.

We’re focusing on cargo bike logistics because that's where we have the most expertise and experience. In the longer term, we want to support other types of vehicles in the last-mile space such as different types of cargo bikes and light electric vehicles.

Can you tell us more about the challenges urban logistics operators are facing? 

Esben: We’ve spent a lot of time understanding the problems different types of cargo bike operators are facing. What they see in their dispatch software is heavily designed toward traditional van-based logistics. For example, the routing software might send a fully loaded cargo bike up a steep hill. This isn’t a problem for the van because you have a combustion engine to propel you up, but it is for a cargo bike. 

Another example might be that you only need to load your van once in the morning and you can go for a full day because you have such a large capacity. This means there’s relatively little change in your operations apart from estimated times of arrival if a driver might run late. But with cargo bike logistics, our partners observe that everything changes all the time and software becomes a bottleneck for them. 

There’s only so much an operator can do with a poorly designed automation and UI. At some point they hit a wall. They can only have so many riders, so many parcels, and so many stops that they can service. With these types of logistics you can't just only add another dispatcher and expect to scale times two.

Software becomes a bottleneck for cargo bike operators and makes it hard for them to stay profitable and scale their operations. You can't just expect to throw more dispatchers at the problem and expect everything to go smoothly. So that's the problem that we're focusing on for them.

How do decision algorithms or vehicle routing problem (VRP) solvers fit into what you’re building? How did you evaluate your options?

Esben: For us, a vehicle routing algorithm is a building block in our system that we use with all of our other tech. We want it to be easy to use, modular, and easy to modify from an engineering standpoint. We don’t want to get into the hairy details of a problem that other people have already spent a lot of time solving. 

We’ve tried some other VRP solvers and we found that you generally need to get deep into the technical details of how the solvers work and the APIs are not very intuitive coming from the outside. I’m sure they’re intuitive if you’re developing solvers, but that’s not our business. 

Our core business is not to write a solver for the VRP, we want to use it as a building block to build on top of and solve our problem in a better way. That’s how Nextmv fits into our stack and where it is really good. It has a nice API and we value the support Nextmv provides as well. 

Nico: I agree, writing VRP solver algorithms is not our expertise. We come from machine learning and AI backgrounds, so we can understand the optimization mechanics at a surface level, but our aim is not to become OR experts.

On that note, how do you think about the relationship between ML, AI, and decision optimization or operations research (OR)?

Nico: Let’s look at urban logistics: they are extremely complex. Machine learning helps me learn from the patterns of vehicles inside of cities. In turn, this helps me create a more useful representation for operations research (OR) tooling to succeed. I’d say AI is an umbrella above the two and it’s about developing intelligent systems and solutions using whatever methods you have available. 

Esben: Our core machine learning models are trying to represent the urban environment as accurately as possible. So being able to say what are the speeds of different types of vehicles on different roads, going up a hill, going down a hill with different loads, service times, and uncertainty about demand. Many of the operators we've worked with are either same-day or next-day, there's a lot of uncertainty about what's going to happen in the next two hours. That's one of those things that goes against the classical VRP that assumes you know everything up front — you know quantities and locations and everything and that's one of the places where our ML solutions come in and you can try to represent that uncertainty. Where is the same-day stuff going to come from? We need to make sure the riders are in the right place at the right time and can actually service that.

How can people best support your mission and get involved?

Esben: If you are interested in this problem space and you're not necessarily technical, you can have the most impact by driving change on a local political level. The long-term impact is going to come from political change in cities that are going to push and incentivize operators in this direction.

I also think that many people think it's completely infeasible where they live to have real bike infrastructure. We both grew up in Denmark, and people think that they've always been bike friendly there. But there were massive political movements in Denmark and the Netherlands in the '70s that pushed for bike infrastructure and that's how it got implemented. So it's not like it's always been the Dutch and the Danish people cycling around. Cities were driven to change because of political movements.


To learn more about Kale AI, check out their website, follow them on X, (formerly known as Twitter), and LinkedIn.

Banner photo courtesy of Kale AI.

Video by:
No items found.