What is the role of data and technology in Duffl's inventory and demand management processes?

This is actually our core competency. In order to deliver in 10 minutes, the products have to be five minutes away before the customer orders them. So it's fundamentally a data challenge. How do you predict what people want before they want it? I think this is why having a store that's built entirely on a digital substrate is very advantageous. 

The fact that you can track every single view, every single click, every single Add To Cart, every single Remove From Cart, every single conversion, every single time a person comes back for the same product, what time they come back, who they are, the demographic data, and all of that mostly right now, the frequency (but slowly as our models improve) of every single component like their search, the wording they use, their delivery address – all those factors are going into the supply chain algorithm to decide what we buy on any given week from any given supplier.

So the way it works today is we click a button, and it generates a list of products to buy from every single supplier. And that's the software we give to our store managers and admirals (the students running the store).

We use the data we collect on the consumer app to inform the supply chain algorithm. So we've abstracted that entire merchandising process that people have to do. But I think the next step is, number one, having a really good SKU cycle.

What I mean by that is, how do you experiment with 15 to 30 skews per store systematically every single week? What are the evaluation metrics? Is it search conversion rate, retention rate, rate of sale? How many sales per hour? Or is it an internal democratic rating system by the entire community? And then how is that data being used to generate the next set of products, and how are they deactivated? And how are they resurrected after they’re deactivated? 

So having that fluid pipeline will be very powerful so each store has a unique selection that’s curated for that demographic. I think the UC Berkeley store is going to look fundamentally different from the Arizona State University store with different people. 

And overall, how can you then personalize on the customer side using state-of-the-art machine learning to be like the TikTok of snacks, basically? How do we get to the point where they trust us to show them what they actually would like? And that's a lot to build, but that's where we're headed.