Q: Why has online grocery been such a struggle?
Pradeep: eCommerce outside of groceries is done in big warehouses. Doing perishables and food and groceries out of big warehouses far away from people has a lot of operational friction. While they might be optimized for in-person shopping, supermarkets are well laid out for eCommerce. As a result, they wind up adding a ton more fees that you normally cannot tolerate as a customer. This suppresses demand and retention for online grocery adoption.
The narrative has been, “Online grocery adoption is starting to happen.” That turns out to be 3% of the market. Even after COVID, it’s settled at around 5%. We’ve been working since 2016 to figure out online grocery as a last-mile service with the perfect eCommerce experience. That’s brutally hard; inventory control of perishables is a pain, making money while selling perishables is a pain, reducing fulfillment costs while dealing with a real-time last-mile delivery is a pain.
Q: Isn’t online grocery delivery growing quickly?
Pradeep: Yes, for now. Grocers can keep posting 100% year over year growth for four or five years but that’s still miniscule adoption. Then the growth rate will come down because they’ve saturated all customers who are tolerant of fees and quality problems. At that point, they will struggle to unlock the real market — people living in the suburbs with tight budgets and low tolerance for change orders, out-of-stock and unpredictable delivery fees. These are folks have $100 a week to spend. They’re going to Safeway and Kroger. They’re also going to Aldi and Trader Joe’s. Sometimes they go to Whole Foods, sometimes they go to Walmart. That’s most of us. That’s how we shop, across the country. That segment is 70% of this $1 trillion space.
Q: Why has online grocery retention not been higher?
Pradeep: The problem for online grocery customer retention is the combination of three defects: order adjustments, late deliveries, and product quality along with associated refunds. If you screw these three things up customers will leave, guaranteed. There are operational challenges to delivering on these three things out of existing supermarkets. It’s difficult to do defect management when you’re picking out of a supermarket. Inventory controls are critical, so you probably want to digitize your inventory control ASAP. That’s easier said than done because that’s also affecting labor practices and in-store technology infrastructure. Supermarket product quality is usually good, but it can be exposed to challenges during delivery.
The best way to control all these defects is really to vertically integrate and do it in a warehouse environment close to the customer with strong technology integration. But just opening a microsite is not a solution. You need the technology, the integration, and the operational ability to run the microsite to optimize for online conversion and retention. You also must perfect the online shopping experience, which is a challenge unto itself for most grocers.
Most of the people who tried online grocery — more than 50% of the US during the COVID period — did not stick. How do you get them to stick? And how do you make those orders profitable?
Q: I thought COVID gave a huge boost to online grocery.
Pradeep: Most of the people who tried online grocery — more than 50% of the US during the COVID period — did not stick. How do you get them to stick? And how do you make those orders profitable? Offering delivery is not the same as scaling it properly. We have 50,000 supermarkets in the US. Most can support 150 online delivery orders a day. We have 130 million households in the US. How’s that going to work? You need to increase access with more delivery slots and increase capacity. You need to get thousands of orders per day out of a location to really get eCommerce to work.
To deal with the capacity issue most grocers have said, “These orders are not profitable, but we like the incremental revenue, so let’s add more marketplaces.” This is crazy to me. They are basically saying, “We’re already losing money on every online order, but now we’re going open up more chances to lose more money on every order and lose the direct relationship with our customer and quality control. But we do like the revenue that’s coming from it, so I guess that’s okay for now.”
The upshot? You really need a different approach driven by breakthrough technology that lets customers pay the same or lower than the stores, get free delivery, and enjoy a perfect order experience every time. Better tech really offers the opportunity to make eCommerce amortization happen while you’re scaling up online grocery. Technology is a capital-light option. You don’t have to add to new stores that cost a few million a pop. You don’t have to build big warehouses. You can do it the way we’re doing it — turn on a lighter warehouse and go live in 14 days with the right tech. We went live in 10 days.
Q: How long did it take you to figure this out?
Pradeep: The big drivers of adoption and retention may sound obvious, but it was challenging and took us years. There were no comps for targeting a mass market US shopper. We learned by focusing on perishables. The challenge is nailing the SKU mix. Even if you can order 30,000 products, we know from our data there’s only about 2,000 products that really sell. Most of those products heavily long tail and it’s perishables.
Once you identify what the perishables are, you must control the inventory. There are a few unlocks here that are compelling. When you get very good at perishables and managing perishables in a warehouse you can move the needle on overall net margins. Most supermarket chains, they wind up doing about 35% food waste across the board, including their distribution centers. Our DCs are both the stores and the DCs, essentially. And our food waste is under 3% because we have gotten really good at matching procurement to demand, kept quality high and returns low.
Q: How does that change the game?
Pradeep: Food waste numbers that low create ripple effects. Your merchandisers are no longer scared of perishables. This means they tend to procure more perishables to sell and different kinds of perishables. Customers want perishables. The more perishables you offer them, the more they’re going to buy. Perishables are usually 15 to 20 points higher gross margin than non-perishables. We make more on the milk than we do on the Cheetos. If you are offering more perishables to sell and customers are generally buying more perishables than they normally do in online grocery, then gross margins wind up being 10% higher than legacy supermarkets. This means with higher gross margins you can spend more on the labor and driver costs, if necessary. You wind up being able to offer free delivery at a profit.
Q: How did you change your procurement approach?
Pradeep: Software. Since code does procurement better than people, you should never have a person deciding how many bananas to bring in every day. There’s this belief that grocery is a low-margin business. That’s not correct. It’s a low net margin business. It can be a very high gross margin business if you sell the right stuff. You can get to 45% gross margins if all you sell is perishables, but you usually must sell the Cheetos too. You wind up as a grocer selling more non-perishables because they don’t go bad and affect you at the food waste level. As a result, your blended margins wind up being in the low 20s.
That’s really the problem. Making poor predictions means you wind up not being able to keep the right SKUs in stock. You’re not able to increase your margins, which really makes your net margin profile so much weaker. That gives you less flexibility for eCommerce and fees.
Q: What are the most surprising things that you’ve learned when you started to apply AI and machine learning?
Pradeep: It was not on the shopper side but on internal decision making — how much we were underestimating procurement. Before we turned on AI and machine learning for produce, we had a terrific buyer. He had grown things before and knew produce inside and out and had worked in grocery. But we also measure out-of-stock rates, and we saw that out-of-stock rates are crazy high for the products that he was pulling in manually.
We told him, “We value that you are qualitatively incredible at produce but quantitatively, we’re leaving a lot on the table in purchasing.” I don’t think people should ever do purchasing. Across thousands of products, it’s impossible to predict exactly the right amount to buy across so many variables. Software should be doing it. His immediate feedback was, “You’re using AI to replace my job.” We said, “It’s still your job. You’re going to inspect more and make more decisions on what to procure rather than the amount to procure.” We turned it on and realized we were under-buying by 45%. That was a huge change in our thinking, because as demand grows, the format of a supermarket breaks.
Q: Once you get loyal customers, how does that impact personalization?
Pradeep: Everyone’s different. You have 2,000 products you can show them during an eight-minute session. What are the things you should show them? In a physical supermarket, you can’t do personalization. You also can’t do good personalization if you don’t have good data on how a person shops over more than a few sessions. That’s where all the pieces start to come together and create a feedback loop — conversion and retention lead to better data lead to more accurate personalization lead to better conversion and retention.
Once you can do this and have digitized inventory, it unlocks entirely new possibilities. We just released a feature called “Eat This First” which tells shoppers what to eat based on expiry and aging of the items which they have already purchased. You need digitized inventory with very precise sell by dates. We can do relative comparisons and say, “You just bought salmon. You should probably eat this before you eat anything else that you bought from us because the probability that it will not be edible in six days if you forget about it is very high. Next, you may want to eat the salad and drink the milk.” Not only do we know what’s on our shelf at the unit level, but we also know what the sell-by dates for those units are that we’re sending you. This kind of leverage is not something that grocers are used to but it’s something that technology companies are very much used to. We are looking forward to doing more of it.
