In this YouTube Q&A, Brandon asks Vishal Kapoor about two core marketplace challenges: whether to display items that are predicted to be out of stock and how to handle pricing for scarce physical goods. Vishal explains that a machine‑learning model predicts item availability, and the platform often still shows the item to give shoppers a chance to substitute it in‑the‑moment, but only up to a reasonable substitution rate to avoid overwhelming the shopper. He notes that the decision involves tuning thresholds for how often out‑of‑stock items are shown and balancing substitution opportunities against a poor experience. Regarding pricing, Vishal says the platform does not apply surge pricing to individual items because it could damage retailer trust and create liability; instead, any surge is applied at the order level (e.g., paying more for faster delivery). He acknowledges that limited markup may occur for highly seasonal or exclusive items (e.g., Thanksgiving turkeys), but this is an exception rather than the rule. The discussion highlights the interplay of predictive inventory, shopper experience, and pricing strategy in modern grocery delivery marketplaces.
Questions this video answers
The platform often still shows items predicted to be out of stock to enable real‑time substitution, but only up to a reasonable degree; thresholds are tuned to balance substitution opportunities against shopper experience.
No, surge pricing is applied at the order level (e.g., for faster delivery) rather than marking up individual items, to avoid harming retailer trust and creating liability. Limited markup may occur for highly seasonal items like Thanksgiving turkeys.
Chapters
I had a question you mentioned around inventory and the negative customer experience if you guys don't have the inventory that shipped. Have you guys decided not to show that inventory, but to show that inventory as not available? And if you have decided, what went into that decision? And secondarily, with, with physical items, similar to, to with Lyft, if there's a physical item that's out of supply, let's say it's, it's one of those situations similar to out of a ballgame with Lyft, do you do surge pricing with that item to try to bring the, the market back into equilibrium? And what data points, if you do, go into what pricing you make for that popular item at that time?
Yeah, this Those are great questions, Brandon. I can clearly see that you are somebody who has your, you know, the rubber to the road. So I appreciate the questions, first of all. So I think the first question was—
You show the item or do you show it as—
Do we show it or not? Yeah, right. I would— so my response is that I don't work on that side. So don't quote me on it, but I'll tell you what I think happens. There is a machine learning algorithm that tries to predict whether or not an item will be in stock. That I know. But what is the outcome of the algorithm in terms of what does the customer see? So I'll cover that in a second. There are 2 parts to this problem. One part is we, within a reasonable degree, we can allow— we should actually show, even if we think that something will be out of stock, At that point in time, we should show it. And the reason being that a lot of this shopping exchange, at least today in the way product is designed, both for Shipt as well as Instacart, a lot of that actually happens in the moment. So if an item is not found, so let me give you a very specific example. You know, there's a specific, let's say, brand of 2% milk that somebody has ordered. right, that you have ordered as a customer. Now that brand of 2% milk is not available, but another one is, and it's just a buck higher. Now in the moment, right, so the fact that we allowed you to order it gives us the opportunity or gives the shopper the opportunity in the moment to ask you the question, I don't— I'm not finding X, can I actually substitute that with Y? If we never took that opportunity, if we never showed that, then we wouldn't know that you wanted milk in the first place. So up to a reasonable degree, that makes sense. It doesn't make sense if, you know, you place like, you know, 10 items and all 10 of them are like the shopper is asking you to substitute all 10 of them. That just doesn't make any sense. And then, you know, where do you draw the line? Like now do you start showing items which don't even exist in the store? Like where do you draw the line, right? For example, because they were pulled from the aisles or something like that. So I think that up to a reasonable degree that the algorithms can predict that this item will be available. or there is a reasonable substitution that could be made. It makes sense. And also, there is, again, as I said, there is a fine line between how many of those chances do we want to— out of 10 items, do we want to do it 20%, 30% of the time, or do we want to do it 100% of the time? So both of those things are— it's very analytical, and there's a machine learning algorithm that tracks all of this. But both of those things are sort of what we— Keep in mind before making— before showing something to a user, to a customer. And then the second part was, do we search for a particular item? It doesn't quite work like that, Brandon. I mean, searching on particular items, what is— I think where you are coming from is if an item is exclusive and everybody, let's say, you know, take a simple example, turkey on Thanksgiving, just make it more, you know, concrete. If it is so exclusive, do you want to like mark up the price of turkey by something? Some markup happens. I'm not gonna deny that some markup happens for some of the items, but the pricing generally, the surge pricing actually works on an entire order level. It doesn't work on an individual item level because we are not in the business of policing, at least right now, we are not in the business of policing what should the base cost of an item be. And the other risk is if we do mark up the price of an item, think about it like this. If we do mark up the price of an item, you know, 2x, 3x, 4x, it actually might make the retailer look bad because they might think that, right? Because they might think that my product, somebody went into my store because the way the shopping experience works is a customer goes into a store, into a retailer store, let's say Whole Foods, And, you know, then picks up an item in that store because the shopping is happening at the store. Now, if you mark up something 4x, 5x, you as a customer might think that Whole Foods is actually the one who's— because you don't know what the underlying price is. So you have to be very careful about this. I mean, it could also, you know, open you up for some liability if you do that. And the other fact is, you know, people will just not trust your system and just stop coming and and shopping with you. So the way you do it is you generally have a price. When you do the price, you know, the surge pricing, the right way to think about it is it's more like Amazon, for example. It's more like an Amazon or a USPS, which is you basically pay more for instant gratification, which is if you want a 1-day delivery or 2-day delivery or, you know, a few hours delivery, then you actually pay up a premium price. You don't actually do it on the contents of what is inside that delivery. Does that make sense?
Yes, that makes sense. For some reason, I thought that there may be items that you need to reset the equilibrium in the market for, sort of like how you reset—
Yeah, turkeys on Thanksgiving. Like, that's an example, right? There's other, you know, other things like, you know, like apple pudding or something else, for example, that people are ordering. You want to probably mark up the turkeys on Thanksgiving. You want to, yeah, yeah.
