Discover more from Gad’s Newsletter
Out of Stock, Out of Mind
But this is not a finance newsletter, so I would like to focus on the operational aspects.
One of the most interesting features of the firm’s operations is that, unlike many other ultra-fast grocers, Instacart picks up and delivers customers’ orders from local stores. In fact, you can choose which store you like to order from.
For example, in Philly, on a Saturday morning, as I write this article, these are the options I see on my screen (and I’m not an Instacart customer):
Similar to DoorDash, which delivers from local restaurants (even though it has started delivering groceries from Dashmart —its local fulfillment center), but quite different from Amazon, which delivers primarily from WholeFoods and its own warehouses, and GoPuff, which also delivers from its own warehouses.
The main advantages of fulfilling from other stores is that you don’t have to maintain a warehouse or carry inventory, both of which are very expensive and require significant capital. But not carrying inventory also has disadvantages. One of them is the fact that you can’t really control the inventory. And I don’t mean only the product assortment a grocery store carries, I also mean whether the product is even available or not.
Growing up in a world of Google and Facebook (that track each and every click), you may be wondering how it’s possible that Instacart doesn’t know if an item is actually available, and you might be shocked to learn that retail stores have inventory accuracy issues.
I always find the following fact astonishing. In a research paper published in 2008, Nicole DeHoratius and Ananth Raman found inaccuracies in 65% of the nearly 400K inventory records they observed across 37 retail stores of an unnamed retailer. You may find this odd and exaggerated, but the consensus among others is that the number hovers around 60%. In fact, the belief is that this level of inaccuracy drives almost 15% of stockouts.
Why there’s so much inventory inaccuracy is a different question, but I always tell people to do the following experiment. Next time you go to the supermarket, buy four flavors of the same yogurt brand. Choose strawberry, raspberry, vanilla, and plain (it has to be these 4, don’t ask why, these are the rules). If the cashier scans one of them 4 times (as many usually do since it’s easier), they will have messed up the entire inventory system for the entire day. The same happens if you take an item and return it to the wrong shelf, when you decide not to buy it.
And it’s one thing if this happens at the supermarket —if the product you are looking for seems to be missing, you can ask a store associate to help you or to check if they have more of said product you want in the backroom, or you can look around and find a similar product to the one you want.
But when everything is online, things are harder. You can’t really look around and there are no store associates available. Things become even worse when you are on a platform that offers its customers the promise of reliable service: the promise being that customers won’t need to go to the store and do another round of replenishment because an item was not available.
So now comes the big question: how transparent do you, as a platform, want to be on the fact that your partners may not provide “accurate information,” and really don’t know whether an item is available or not?
It may be less of an issue for a platform like Amazon or Uber, who have millions of third-party sellers or gig workers. But for a platform like Instacart, that has but a few partners, and each carries a significant share of supply and customers in each area, it’s a pretty big deal.
So the question is, to what extent should a platform be transparent with customers on:
how little control they have over inventory,
how bad inventory systems really are (the inability to know whether something is in stock or not), and
how inefficient their partners really are.
A recent paper by Dmitry Mitrofanov and collaborators from Instacart, shares an interesting experiment they conducted. From October 2021 to April 2022, Dmitry and Instacart ran two different field studies on an inventory information sharing policy:
“Over the course of these field experiments Instacart appended a ‘Likely out of stock’ label to an item’s description on its storefront if Instacart’s machine learning model estimated the probability of the item being in stock as falling below a certain threshold.”
You can see an example below:
First, they ran a month-long field experiment with the goal of estimating the impact of this information-sharing policy on customer purchasing behavior and customer satisfaction.
The results? Items that were “likely out of stock” were ordered less often than items that were in stock. In fact, there was a 25% relative reduction in the proportion of items ordered that were likely to be out of stock.
This is expected since people want to order things that they are likely to receive. One may theorize that these items are going to get a boost from the fact that they are in demand, but while I am not sure grocery shopping is where we see significant scarcity effects, the paper shows that this is indeed the case.
How does Instacart usually address stockouts? If a customer orders a product that isn’t available, a replacement is offered (a similar product from a different brand, for example), and if a replacement cannot be agreed on, a refund is issued. Instacart loses in both cases. Replacements mean that the customer will be somewhat dissatisfied with the lack of fit (and the acknowledgment of inventory inaccuracy), and refunds mean a loss for everyone involved: the customer (as they aren’t receiving the product at all), Instacart (as they risk losing a dissatisfied customer and the potential revenue associated with the refunded product), and the grocery store (as they are losing a sale).
In the experiment, “Instacart observed a 3.29% decrease in the proportion of items replaced, and a 2.6% decrease in the proportion of items refunded.” So the short term impact is positive: people react to the signal and order less of the product that is potentially stocked out. When they do request that product, it happens less often, so there are fewer losses to the firm from refunds.
But this raises the question of the long-term impact. Maybe people buy a different product a few times, but then once they realize that stockouts are so pervasive, they just stop using Instacart. This is a fear I see with many firms (a point we previously discussed in the context of in-store signage) —that they may actually hurt their image of high quality service if they disclose information about their “lack of efficiency.” The customer may say “you are very trustworthy, but now I realize that you are not very efficient.”
That was the goal of the second experiment by Dmitry and his collaborators: to understand the implications of sharing item availability information on longer-term revenue. That required a much longer study (which lasted 6 months).
The results are very reassuring. They observe a 5.33% relative increase in total revenue per customer over the course of the six-month period, and specifically an increase of 4.9% in order frequency.
The rationale: People indeed value trust and don’t punish the firm for its clear lack of accurate inventory information and stockouts. And while they may not buy more when they buy, they do buy more often.
A 5% increase in frequency means that if a family orders once per week (i.e., 50 times a year), they will order around 2.5 more times a year. This may sound very little, but remember two things: this is a very simple intervention (you are alerting people to potential stockouts), and this is a low margin business based on volume and density. The more orders you receive from existing customers (amortizing CAC), and customers of the same area, the better off you are.
Now, everything has a limit, so if you are constantly out of stock, your customers may ultimately stop ordering from you. The following graph from the paper, illustrates this trade off:
At the level of 15%–20% searched items that were “potentially stocked out,” we see the biggest increase in customer spending. At 30% (and this is during COVID when stores experienced significant product shortages), there was no increase, but also no reduction in spending.
The reality is that these issues are quite pervasive in many platforms, but not only. And since you don’t have much control on the supply side, how much information are you willing to share?
What Information Should Firms Disclose?
Long-time readers of this newsletter know that I am a huge proponent of information sharing. My motto: Tell the truth. Tell only the truth. But you don’t have to tell the whole truth. It’s ok to keep things ambiguous if this ensures that information is: actionable, digestible and interpretable, and finally, transmitted truthfully.
Actionable: the information should allow customers to make better decisions so they may improve their own experience as well as that of others. For example, when some customers didn’t order the “potentially stocked out products,” they increased the likelihood of those that really wanted to get them.
Digestible: help customers build the correct expectations about their experience. More realistic expectations make for overall better decisions and higher customer satisfaction.
Transmitted Truthfully: neither side has incentive to lie or intentionally mislead.
But again, you don’t have to disclose everything. Telling customers how many units of a product you think are left, or that it’s the last one, or similar information, may backfire if the customer gets the impression you are bluffing (trying to induce them to buy) or if it’s too much for them to process.
A few years ago, a student told me that a mobile provider informed him he was number 98 in line and asked me what they meant.
They were subtly trying to tell you that they are, in fact, clueless.
Thanks for reading Gad’s Newsletter! Subscribe for free to receive new posts and support my work.