Online Grocery Substitutions - Phish Food or Fish Fingers?

Cleaned transcript

Introductions and the online grocery substitutions problem

COLIN:

For online grocery, most orders around the world are typically picked from the store. When the picker finds a gap on the shelf where the item has been ordered, there’s a challenge: they need to quickly find an appropriate substitute—if the customer has said they would accept one.

Online grocery substitutions was the topic, the challenge, and the opportunity we covered as a theme for one of our recent meetings.

In this podcast, I’m joined by Professor Daniel Corston. We’ll explore the discussion points that came up from the meeting: how to not mix up your fish food with your fish fingers, and the different approaches retailers are taking to understand what preferences and substitutions might mean—while also preventing the risk of a poor substitute using advanced data analytics.

It’s a fascinating topic and a huge productivity challenge for retailers. Roughly, for a typical order of 15 or 20 items, maybe 25% of the time can be spent looking for substitutions. It’s a compelling topic, and there are some great insights from the good professor.

What the working group covered

COLIN:

We had our on-shelf availability working group meeting—our seventh meeting now. We’ve got a really good cohort: mainly retailers and a few consumer goods producers. The focus, and the second time we’ve looked at it, was online substitutions.

By that we mean: when the picker gets to the item that’s been ordered as part of an online grocery order and they can’t find it—what happens next, and could you prevent that?

We had three speakers: a retailer from America—one of the biggest grocers out there—then our friends at West Monroe, who do time and motion studies and productivity analysis, and then Federico from Everli, providing a platform perspective. We also had contributions from retailers like Tesco and others on the call.

So, what were your key takeaways? At a high level, what should a retailer take away strategically?

The cost of online grocery substitutions and the productivity hit

DANIEL:

Thank you, Colin. It was a great session. To me, there were three themes: the cost of substitutions, some solutions—where we realised that apart from improving on-shelf availability, which is a no-brainer, personalisation is a big thing—and then some more conceptual, high-level thoughts towards the end.

Starting with costs: remember last year we did the survey to find out the cost of a substitution. Thanks to the group, we estimated that for about every dollar in margin you can gain through substitution, retailers pay about $2.40, $2.50 in cost. Shocking.

COLIN:

It’s still shocking.

DANIEL:

And those numbers haven’t budged. This year’s session added another way of measuring it. West Monroe said it takes, on average, 26 minutes to pick a basket of about 25 items—so in principle about one minute per SKU.

But what was interesting is they assume that a basket with 25 items has about two stockouts—somewhere around the famous 8%, which is apparently an industry… I don’t know who said it, but a brilliant person probably.

Jokes aside, they said those two items take about 20% of the picking time—five minutes. In other words, picking productivity in store is largely determined by the out-of-stocks. If five minutes of 25 minutes is stockouts, you can imagine what improving this would mean for picks per hour and picking efficiency.

COLIN:

And that’s why we’ve heard about people putting on an extra picker who comes around and picks all the ones they couldn’t pick first time. We also heard about flashing lights on electronic shelf-edge labels—potentially speeding that up and maybe negating the need for that second shift. We haven’t reached conclusions on it, but those were improvement opportunities being discussed.

“What should it blink?” and why personalisation matters

DANIEL:

That’s where it gets interesting. We have technology—shelf labels that can blink—but what should it blink? What is the best substitute?

This is where we went back and showed the survey results from last year. It turns out that many retailers, when asked to provide a substitution, will call if they can—which, with all respect, is a wonderful waste of time. It’s wonderful because if you can reach the customer, they love it—or if they don’t love it, at least you know.

But when retailers don’t call, substitutions are either random—because about 50% of the time the picker provides something from gut instinct—or they provide “what most people do.”

And what we really need is personalisation. Electronic shelf labels might direct the picker to the next generally accepted alternative, but not necessarily your preferred one.

Everli’s approach: asking customers in advance

DANIEL:

We had two interesting talks. One was from Everli in Italy. They developed an algorithm for personalised substitutions, partly because they face challenges in their model.

Retailers don’t necessarily share valuable information with platforms that pick in their stores. They may not share locations. They may not share inventories. They may not allow them into the back room. They may be late and delayed. So in other words, these guys are picking blindfolded because they don’t really know where the stuff is.

So what Everli did was this: when their forecasting suggests that when the picker arrives the product might not be there, they ask the customer, “Just in case this happens, what will be your preferred substitute?” If the picker arrives and the product is there, they don’t need it. But if the product isn’t there, they know exactly what to offer.

Federico reported improvements: net promoter score increased, churn reduced, and customer lifetime value was clearly better.

Using basket data to predict better online grocery substitutions

DANIEL:

Then there’s work where I participated: with Professor Srikan Chakrabutla from NYU, with whom I supervise a PhD student, we developed an algorithm which we tested at Everli. Our algorithm is also personalised. It looks at past basket composition and what people bought when a certain product was out of stock.

We were able to simulate and improve results. It shows that if you do this in a platform like Everli, you can increase sales by reducing lost sales by about 1%. I think that is the future—with algorithms, Gen AI, and all these opportunities. We should be better at personalising our substitution decisions.

COLIN:

You sound very optimistic—and it does sound like the right direction as an industry. We certainly have the power of the maths. That’s a very good and inspiring thought.

“Perfect baskets” and moving from picker-centric to customer-centric

DANIEL:

This brings us to other ideas. There was one thing the retailer from the United States said that sounded impressive, but it made me think. Remember what the lady from the American supermarket said?

COLIN:

I can’t remember the exact words, but the customer received a note that said, “Happy days. We picked everything.”

DANIEL:

Exactly. What happened was: when customers received a delivery with substitutes, the retailer would say “you have one substitute” or “two substitutes” or “three substitutes.” But once in a while, someone received a delivery without any note, and they worried—thinking, “How many were substituted?” So the retailer realised they should write down that nothing was substituted.

I love that it’s a way for retailers to engage. But as an industry, we should strive for perfect baskets—perfect basket fulfilment—from the customer’s perspective. I love first-time pick rate, but it’s still internally centric: picker-centric and SKU-centric. What I loved about this meeting was the interest in becoming more customer-centric—moving from SKU-centric to customer-centric, from picker-centric to customer-centric—because otherwise we just create internal measures that are easy to compute, but don’t show how we’re permanently frustrating customers.

First-class and second-class customers

DANIEL:

There’s one more thing. You were listening in to that conference we sometimes attend at Wharton and Harvard, where there was a thesis that we have first-class and second-class customers.

We treat our online customers as first-class customers because we go out of our way. We spend 20% of picking time finding something for them. They get a delivery and all these wonderful things.

But 85% of the people are second-class customers—that’s me going shopping like in the old days. I spend my time trying to find a substitute. I walk around. I talk to associates: “Where can I find this? What is good?” I spend all of that time.

As an industry, we should become more customer-centric—not just online-centric. Eventually, we should be able to provide the same service to offline customers as for online customers. That should be our vision.

COLIN:

That’s a great vision.

Inventory accuracy as the foundation for online grocery substitutions

COLIN:

It all comes back to discipline around inventory record accuracy. That’s the heart of what’s going wrong, at least in part. Some of it could be in the back room, some could be up on risers. We’ve talked about sessions where we found a lot of inventories always in the back—it should be in the front.

There’s work to be done around inventory management: knowing where it is in the store and knowing that it’s accurate. That becomes the platform to deliver a customer-centric approach to substitutions.

And that Harvard presentation a few years ago—provocative, delivered with passion.

DANIEL:

I don’t think it was controversial. It was provocative, but most people felt it was right.

COLIN:

Yes—provocative, and delivered in a provocative way. And it was linked to “why am I paying more for my groceries to subsidise the first-class shoppers when I walk in?” That added a bit of spice.

Teaser for the next session: self-checkout and availability

COLIN:

Next session is coming up after the summer break in September. We’ll be looking at the relationship between self-checkout and on-shelf availability. There’s a sense that missed scans lead to items not being sold, which messes up inventory files and should, in theory, lead to on-shelf issues.

We want to see where retailers have explored that and what data there is.

DANIEL:

Wait—didn’t you have a teaser? Wasn’t there a question we asked about whether the audience thought this had an influence?

COLIN:

Yes—71% was the number.

DANIEL:

Don’t hold that back. This is the teaser for the next session: 71% of participants confirmed that they believe self-checkout messes up balance on hand and inventory records. That’s big. It not only messes up the sales record, but the inventory record. Then you get problems where the picker can’t find the item because they thought it was there.

We may have a retailer speaking about a solution or intervention—auto-correction—to address this, but we also want to prove the case, and hopefully more retailers can share their experience.

COLIN:

Daniel, thank you again for sharing your thoughts. Fantastic as always—and getting on your soapbox. I love that. Bye-bye.

DANIEL:

Thank you. Bye-bye.