Mike Hudack

Well, over time, I think we learned a lot about the importance of selection and all sorts of things. And I think Deliveroo Plus really influenced people's behavior in a really dramatic way, where people feel a loyalty and affinity, a sunk cost with the platform, which brings them back. It was a very... unbelievably important product.

But yeah, you know, it was also the first time that there was an alternative in the market. And so the combination of those things was brutal. How important was exclusivity to winning on the supply side of restaurants? You know, I think we thought it was very important.

But yeah, you know, it was also the first time that there was an alternative in the market. And so the combination of those things was brutal. How important was exclusivity to winning on the supply side of restaurants? You know, I think we thought it was very important.

But yeah, you know, it was also the first time that there was an alternative in the market. And so the combination of those things was brutal. How important was exclusivity to winning on the supply side of restaurants? You know, I think we thought it was very important.

But to get back to the question of like figuring out what to build there and how product is different, you know, you land in that kind of a situation and you're like, holy shit, like... You know, we need to fix this problem. We need to be able to modulate demand properly. We need to be able to make a promise to the customer which we can actually fulfill.

But to get back to the question of like figuring out what to build there and how product is different, you know, you land in that kind of a situation and you're like, holy shit, like... You know, we need to fix this problem. We need to be able to modulate demand properly. We need to be able to make a promise to the customer which we can actually fulfill.

But to get back to the question of like figuring out what to build there and how product is different, you know, you land in that kind of a situation and you're like, holy shit, like... You know, we need to fix this problem. We need to be able to modulate demand properly. We need to be able to make a promise to the customer which we can actually fulfill.

And at the time, the logistics algorithm was very, very, very simple. We had an amazing data scientist, a neuroscientist who knew how to build a great delivery algorithm, but we didn't have it. And so the first thing we did was just open up a war room.

And at the time, the logistics algorithm was very, very, very simple. We had an amazing data scientist, a neuroscientist who knew how to build a great delivery algorithm, but we didn't have it. And so the first thing we did was just open up a war room.

And at the time, the logistics algorithm was very, very, very simple. We had an amazing data scientist, a neuroscientist who knew how to build a great delivery algorithm, but we didn't have it. And so the first thing we did was just open up a war room.

We had an apartment across the street from the office, and we moved about 10 engineers into that apartment and had them sit down and spend two or three days trying to figure out the highest leverage way to start actually delivering on our brand promise properly.

We had an apartment across the street from the office, and we moved about 10 engineers into that apartment and had them sit down and spend two or three days trying to figure out the highest leverage way to start actually delivering on our brand promise properly.

We had an apartment across the street from the office, and we moved about 10 engineers into that apartment and had them sit down and spend two or three days trying to figure out the highest leverage way to start actually delivering on our brand promise properly.

And the first thing that they did was they built a lateness model, a machine learning model, to correctly estimate how late we were going to be. The theory being that we could then accurately tell people how long their order was going to take, and they could make a decision as to whether or not to place the order, and they wouldn't be pissed at us.

And the first thing that they did was they built a lateness model, a machine learning model, to correctly estimate how late we were going to be. The theory being that we could then accurately tell people how long their order was going to take, and they could make a decision as to whether or not to place the order, and they wouldn't be pissed at us.

And the first thing that they did was they built a lateness model, a machine learning model, to correctly estimate how late we were going to be. The theory being that we could then accurately tell people how long their order was going to take, and they could make a decision as to whether or not to place the order, and they wouldn't be pissed at us.

We would say, instead of saying, oh, it's going to be 15 minutes, we'd say, oh, accurately, it's going to be 45. And then we would deliver within the 45 minutes, and they would be happy customers instead of us having said it was 15 minutes. Still delivering 45, they're pissed off.

We would say, instead of saying, oh, it's going to be 15 minutes, we'd say, oh, accurately, it's going to be 45. And then we would deliver within the 45 minutes, and they would be happy customers instead of us having said it was 15 minutes. Still delivering 45, they're pissed off.

We would say, instead of saying, oh, it's going to be 15 minutes, we'd say, oh, accurately, it's going to be 45. And then we would deliver within the 45 minutes, and they would be happy customers instead of us having said it was 15 minutes. Still delivering 45, they're pissed off.

And so that team just built a very simple kind of regression model to estimate lateness, and that dramatically improved lates very quickly. And most of the work in Deliveroo for a really long time was actually in what we called the delivery organization. We created an organization called Delivery,