Shownotes
The Fall retail conference season is officially underway, with NRF Paris having just wrapped and Groceryshop, Shoptalk Fall, NACS, and many others just weeks away. Omni Talk figured what better way to get ready for the upcoming season than to put together our primer of the must-see tech that has caught our attention this year.
This episode features Quorso’s section of our Fall Conference Season “Must-See” Tech Preview, where Daniel Slowe, Founder & Chief Product Officer of Quorso, breaks down where AI can deliver practical value in retail operations.
Rather than applying AI to broad, open-ended strategic questions, Daniel explains why its sweet spot can be in the high-frequency, labor-intensive tasks retailers deal with every day, from analyzing customer feedback and triaging store issues to recommending staffing adjustments. He also explores the “harness” retailers need around AI to provide context, guardrails and measurement, helping turn AI-driven insights into measurable action.
Be sure to check out the rest of our podcasts this past week and the next few to catch exclusive content from some of the biggest retail conferences!
We can’t think of a better way to get a bird's eye view into the tech shaping the retail of tomorrow, today!
P.S. To explore more conversations from the Omni Talk Ask an Expert Series, head here.
Music by hooksounds.com
Sponsored Content
This podcast uses the following third-party services for analysis:
Podcorn - https://podcorn.com/privacy
Transcripts
Foreign.
Speaker A:My go to for retail understanding around how AI can optimize store operations.
Speaker A:So welcome to the webinar.
Speaker A:Founder and Chief Product officer at Corso, Daniel Slow.
Speaker A:Daniel, let's start off.
Speaker A:Why don't you explain to us what it is that Corso does?
Speaker B:Chris, firstly, thanks for having a webinar.
Speaker B:I love it.
Speaker B:So it's excited to be on it.
Speaker B:Corso is an intelligent management platform that organizations use to kind of orchestrate, guide, connect people's daily work.
Speaker B:So that's a lot of words, but in short, the system ingests a whole bunch of companies, data, events, signals, etc.
Speaker B:Just all that.
Speaker B:And it then determines what actions should be taken by each and every person across the organization.
Speaker B:It prioritizes those, it assigns them out and you know, store teams, field teams, et cetera, they get them on the app.
Speaker B:Um, it guides them to completing those actions.
Speaker B:And then importantly, and this is the key bit, it actually tracks the impact of every action taken and feeds that back into the system to further define future work.
Speaker A:That's a great overview.
Speaker A:Great.
Speaker A:Well, well, really well said too.
Speaker A:And it's a great level set for the rest of the conversation we're going to have with you today.
Speaker A:So, so Ben, I want to ask you, we want to start, we talked about this beforehand.
Speaker A:Our big broad question for you is like, there's a lot of talk about AI in retail.
Speaker A:And so our first honest point blank question for you, Daniel, as an expert on the subject is do you think AI is actually valuable in retail?
Speaker A:My guess is you think yes, but we want your honest take.
Speaker B:You know, I'll give you an interesting answer.
Speaker B:So firstly, can we just, can we just get clarify AI, you know, it's been around for ages and AI includes, you know, machine learning, etc.
Speaker B:So just for the purpose of this conversation, Chris, I think we're all probably talking about LLMs and Frontier models.
Speaker C:Yeah.
Speaker B:So if you're asking the question, do I think LLMs are useful in retail?
Speaker B:The honest answer is I was pretty skeptical.
Speaker B:So for a long time I was like, oh, this is kind of cool, it can help me plan a holiday.
Speaker B:But is this really reliable enough to be able to embed within a complex organization?
Speaker B:And so yeah, I was skeptical.
Speaker B:But I must be honest, having embedded it throughout our own business and seen huge sort of productivity gains, yeah, I can say it's truly transformative.
Speaker B:So yes, I'm now convinced it is hugely transformative.
Speaker B:But here's the caveat and this is where it's interesting.
Speaker B:It's just a technology it's not magic.
Speaker B:And like every technology that's come before it, from sort of steam train to penicillin, the World Wide Web, it has its strengths, has its weaknesses.
Speaker B:If you deploy it in the right place, in the right way and leverage its strengths, it can create enormous value.
Speaker B:If you deploy it in the wrong way, in the wrong place, it can be very value destructive by the way, because the power of it, the value it can create, is offset by the value it can destroy.
Speaker B:Yes, it can be transformative, but only if used in the right way.
Speaker C:Okay, Daniel, that's great.
Speaker C:Let's get into detail then because I want to build on that last point.
Speaker C:Let's go to practical perspective.
Speaker C:What, what therefore are the best use cases for LLM based AI in retail?
Speaker C:And go and tell us what's the worst.
Speaker B:The best use cases are not the most creative ones, it's not the most exciting ones.
Speaker B:Okay, so when the LLMs came out and people go, you know what we can do?
Speaker B:We can put an LLM on top of all of our data insights and it's going to tell my DM exactly which store to visit at 9am on a Thursday.
Speaker B:That's a pretty cool use case.
Speaker B:But it's not very good at that.
Speaker B:You know, when people like, oh, you know, we can do, we can feed in all of our marketing data, it's going to tell us what strategy to run next year.
Speaker B:No, it's not going to do that either.
Speaker B:So I mean these are being slightly facetious, but those open ended, strategic, creative, big bets, those kind of big problems, it's not actually that good at solving those.
Speaker B:You know, humans are, by the way, those kind of.
Speaker B:Because we're very good at the judgment that those things require.
Speaker B:But the kinds of things it's fantastic at are the boring use cases.
Speaker B:I say boring use cases by the way.
Speaker B:It's the stuff that keeps your business running, but it's the kind of high frequency things that your teams are doing every single day in the stores.
Speaker B:So let me give you a couple of examples.
Speaker B:When a freezer breaks down, AI is fantastic at pulling the sop.
Speaker B:I'm pulling the context that's broken twice the last month and telling you exactly to resolve it and potentially resolving it.
Speaker B:You know, it's great when someone doesn't show up of recommending which other shift should be adjusted.
Speaker B:It's, it's great at kind of reading the last 13 surveys and saying this is the most common issue.
Speaker B:So it's kind of, it's, it's, it's Great.
Speaker B:Those use cases, those, those, those things that your teams are doing every single day.
Speaker B:If you think of, like, if you almost like if you, if you think of a store and go, hey, what are the, what are the things that my store manager does every single day?
Speaker B:Can AI, you know, help improve those processes?
Speaker B:The answer is probably yes, a lot of those ones can.
Speaker B:But that, it's, that's the transformation that comes from making every single one of those a little bit better, a bit quicker.
Speaker B:It's not from sort of, it's not from changing the direction of your business, if that, if that makes sense.
Speaker A:It's actually pretty mind blowing, isn't it, Ben?
Speaker A:Like this is because, because Ben and I track this stuff every day and we hear from people on both sides of this question, which is why we do this webinar, to kind of set the stage for how people should think about tech as they go forward, you know, for the next six months or so until we do it again.
Speaker A:Because, you know, I was just talking to a company that was telling me they could do what you're saying it's not good at, you know, I was talking to them this week.
Speaker A:But so then my question for you is if you are just trying to solve, if it is best at solving those boring use cases, those day in and day out things that just, you know, just riddle people with issues, why is it so hard to do it?
Speaker A:Why is it so hard to get it right?
Speaker A:Is it because people don't understand what you're saying?
Speaker A:What makes it so difficult to go from prototype to deployment?
Speaker A:Daniel?
Speaker B:Yeah, and yeah, Chris, a great question by the way, before we move on, I should say that it's those boring use cases.
Speaker B:It's also the ones that are highly tuned, time consuming for people.
Speaker B:People, by the way, slow at reading, AI, very quick at reading, people, very slow at writing AI, very quick at writing.
Speaker B:So it's those ones.
Speaker B:Why is it hard to deploy?
Speaker B:Again, it's sort of, let me answer that indirectly.
Speaker B:The reason people think it should be easy to deploy is because they are able to do it, do a prototype or do it one off on a computer using Claude.
Speaker B:So Chris, do you use Claude or ChatGPT?
Speaker A:Yeah, I use Claude.
Speaker B:Let's take a single use case, which is analyzing surveys to extract the common issues that have occurred.
Speaker B:Okay.
Speaker B:Okay.
Speaker B:So if you were doing this yourself, you would, you would, you know, drop all those surveys into Claude and you say to Claude, hey, can you tell me the most common issues that occurred in this store?
Speaker B:And Claude would come back and it would say, hey, by the way, you know, I've read your tax returns.
Speaker B:And you'd say, well, that wasn't actually what I asked you to do.
Speaker B:I asked you to read the surveys.
Speaker B:And so Claude would then analyze the surveys and it would come back with some kind of, it would come back with some kind of response and maybe the answer would come back with like four pages and you'll be like, actually, thanks Claude.
Speaker B:But John is, I've been more efficient me to read the whole survey.
Speaker B:So you then say, thanks Claude, a little bit shorter.
Speaker B:And it would say, Chris, you're right, I should have given you a shorter response.
Speaker B:And it'll again then give you like three bullet points that are so specific, like fixed store sales.
Speaker B:You're like, it's a little bit, it's a little bit too high level.
Speaker B:Can you give me something new?
Speaker B:So you, you'd, you'd work with it.
Speaker B:You know, it's a collaboration, It's a collaboration.
Speaker B:I think that's, maybe that's the best way to say is AI is best when it works with a human alongside you.
Speaker B:It's doing the bit that you're bad at the kind of reading loads of surveys, the summarization, it's.
Speaker B:But you're kind of making the judgment calls and you could eventually get to an answer and you get a really good answer and, and it probably still only took you five minutes and not three hours.
Speaker B:Okay, right.
Speaker B:The role you are playing in that interaction with Claude is you're kind of being the operating system.
Speaker B:You're giving it guardrails, you're giving it guidance, you're giving it feedback.
Speaker B:You're basically taking this incredibly powerful tool and you are, you are making sure that it's, that it's being used in the right way to solve the right problems.
Speaker B:Now let's put that in the field.
Speaker B:So you, you've gone, you've gone, you come to me and go down.
Speaker B:That's great.
Speaker B:Look, I prototyped this.
Speaker B:It's fantastic, it works.
Speaker B:I'm like, okay, cool, let's put that in the field.
Speaker B:So what does that mean in practice?
Speaker B:It means that as a DM is driving up to a store, they need to be able to ask their phone for the most common issues that occurred in that store of the last, you know, based on the last surveys.
Speaker B:Okay, so then first they need a mechanism, they need a tool, they need some interface, they need to be able to say that something.
Speaker B:Because they haven't got Claude, they haven't got the kind of just Chat thing, right?
Speaker B:The system needs to go and find the relevant surveys.
Speaker B:So it needs to go and, you know, look up the Store Identifier, find what, you know.
Speaker B:Is it Store Audits, by the way?
Speaker B:Maybe they changed the name.
Speaker B:It was called Store Audits, now it's called Store Walks.
Speaker B:That system's already confused, you know, needs to put in the right stuff.
Speaker B:It needs to pull in relevant context, like, you know, the visiting on Tuesday.
Speaker B:The store manager is not there, the store manager's new.
Speaker B:So it needs to, you know, bring all that additional information it needs to.
Speaker B:The system needs to give all the guidance around, actually.
Speaker B:It needs to be, you know, four.
Speaker B:Four bullet points long, and it needs to be this.
Speaker B:And then after providing all that, the system needs to get the response and give it to the dm and it needs to then track whether this was valuable by asking the DM, was this what you needed, etc.
Speaker B:Essentially, everything that you did on your computer needs to be systemized.
Speaker A:Basically, what you're telling me is there.
Speaker A:There is some human cognition that is needed or required to make these systems work and function appropriately?
Speaker B:Almost.
Speaker B:I would say there is.
Speaker A:At least right now there is.
Speaker B:Yeah, there is.
Speaker B:No, there's that.
Speaker B:I'd say that there is a.
Speaker B:They're almost like the operating system that goes around them.
Speaker B:Maybe that's what I think of if AI is this incredibly powerful kind of.
Speaker B:You know, I think it was like, you know, I always think of it as like a dragon, by the way.
Speaker B:I used to.
Speaker B:I used to.
Speaker B:I used to do a webinar before where I actually had a picture of a dragon.
Speaker B:I said, if AI is this dragon, you kind of need to harness it.
Speaker B:Now, when you're doing it yourself, you are acting as the harness.
Speaker B:You're acting as the operating system around it.
Speaker B:When you put that in your stores, you need an equivalent harness.
Speaker B:Otherwise, like, you know, what.
Speaker B:What's telling it, what to do, what's, like, amazing.
Speaker B:It's incredibly powerful model.
Speaker B:But who's.
Speaker B:Who's telling you what to do?
Speaker B:It's just gone wild.
Speaker A:You need a harness.
Speaker C:Daniel?
Speaker A:Yes.
Speaker C:I'm going to move us from dragons.
Speaker C:I want to take us back to stores.
Speaker C:Because I think we get into the nub of this now, which is the thing that people are finding really hard, which is how to deploy AI at scale beyond that point where you've got somebody who can nudge it and help it and have that interaction.
Speaker C:So help us get practical here.
Speaker C:What should companies be focusing on if they really want to deploy AI at scale in their operations.
Speaker B:The SIMP answer, the one word answer is it's infrastructure.
Speaker B:So it's not about the model, it's not about chat, DBT or, you know, they're slightly commoditized.
Speaker B:And by the way, I can say that I don't have shares in any of them.
Speaker B:It's fine.
Speaker B:It is about the infrastructure and it's about putting in place all of that, the systems.
Speaker B:You need to track everything that's going on.
Speaker B:It's almost like you need to be able to track everything that's going on in your stores because that provides the context and the guardrails, which you can then leverage to inject AI, which is sort of what candidly we do at Corso.
Speaker B:So I always, I, you know, the kind of example I give is, you know, if you think of what retail operations is, and I'll try to, I'll try to describe it, you know, retail operations.
Speaker B:When I say retail operations, I mean all the, all the activities being done by store teams.
Speaker B:It's not one problem.
Speaker B:The reality is, you know, it's 100,000 problems every single day.
Speaker B:And that is because, you know, you have lots of different types of work.
Speaker B:You have events and you have tasks, you have orders, you have follow ups, et cetera.
Speaker B:And you have lots of different types of role.
Speaker B:So you have store associates and store manager and department leads and you know, asset protection, regional managers.
Speaker B:So, and then you've got this enormous web of stuff that needs to be done.
Speaker B:And when I say infrastructure, I mean is you need to kind of get a grip on all the work your teams need to do and you need to sort of almost like for every bit of work that someone needs to do, you know, how do they do it, what information needs to be done?
Speaker B:Like, it, it sounds very, it sounds quite amorphous, but until you kind of have mapped almost like a digital twin of your organization, until you know exactly what's happening, you can't really leverage AI because AI needs that information to be successful.
Speaker A:Yeah, you have to build the harness, right Daniel?
Speaker A:Isn't that Chris?
Speaker B:There we go.
Speaker B:You've got, you've, yeah, you've got, you've got it.
Speaker B:You've got to build, you've got to build the harness.
Speaker B:You've got to, you've got to have a really strong grip of what's going your organization or you need any, by the way.
Speaker B:You need, you need the tooling to be able to get a grip over it.
Speaker B:You know, if, if all of this is mixed across 20 different tools it's very hard to inject AI on the correct way.
Speaker B:So you need to bring it into what you need.
Speaker B:Ideally, you bring it to one system.
Speaker A:Okay, got it.
Speaker A:So let's get you out of here on this thing because I'm curious.
Speaker A:This is the last question I want to ask you today.
Speaker A:You know, if you look, because you see all the retailers across the world really knowing the scope of who you work with, what are the leading retailers doing with AI the way you've described it thus far today, to get value out of it that maybe they weren't doing before the advent of AI that we're talking about?
Speaker B:I think the first thing that the retail which are working, which I think are getting really right, is they are directing their focus on building the stuff that's proprietary to them and candidly buying the stuff that's not.
Speaker B:So let me take one example.
Speaker B:It's a very large gas station.
Speaker B:A company which runs gas stations.
Speaker B:They have identified, for example, that in their business it really matters whether toilets are clean.
Speaker B:And so they have developed a really incredibly sophisticated model for identifying when the toilets cleaned.
Speaker B:By the way, this is kind of what I mean by, you know, that, you know, it's not.
Speaker B:It's not the glamorous use cases.
Speaker B:Where's the high vacuum?
Speaker A:I love this song now.
Speaker B:But what they've also said, and we knew they're our customer of ours, they've all said, yeah, fine, so we're going to build this really good model for identifying when the toys and clean.
Speaker B:But we're going to use Corso to make sure that that alert gets prioritized at the right time and goes to the right person and end up in the shop.
Speaker B:You.
Speaker B:It ends up on their tablet when they need it, and they can track whether it.
Speaker B:Track whether it works.
Speaker B:And we can feed that model.
Speaker B:So they're using Corso as the infrastructure, but they're focusing their efforts on building something which is unique to them.
Speaker B:I think that's just one example.
Speaker B:But I think, as with all these things, don't rebuild the stuff you can buy.
Speaker B:Build the stuff that actually is unique to your business.
Speaker B:So that's one thing they're doing.
Speaker B:Because if you try to build everything, there's always that temptation which is everyone goes, let's build everything.
Speaker B:You know what, you're going 10 years down the line before you, before you get anything.
Speaker B:The second thing that they're doing is they are measuring what works.
Speaker B:So I sort of.
Speaker B:It almost goes back to the start of this podcast where I said, I said AI can be incredibly value, can add a lot of value, but also be very value destructive.
Speaker B:You need to know that.
Speaker B:Which means that what they're doing is anything that AI pushes out, that they push down to their store teams they are tracking, was this beneficial?
Speaker B:Did this change what you did and did it change what you did and did it drive positive impact?
Speaker B:Again, this is something that we of course, I think is kind of core to this and it's one thing that's embedded in our platform is the ability to track the impact of every action taken.
Speaker B:So going back to my example before, which is if you're going to push, if you're going to use a model to determine if we should clean bathrooms and you're going to push out through Corso, let's make sure we're tracking if the CSAT score goes up in that dimension, you know, because if it's not, what are we doing it all for?
Speaker B:And, and so they track, they track the impact of every, of everything that they're pushing down to stores.
Speaker B:And then the third thing they're doing is they're kind of, you know, keeping humans in the loop.
Speaker B:Honestly, and we've sort of talked to this throughout, is yeah, I think AI is really powerful and lms, LMS are really powerful.
Speaker B:They're great, you know, they're, they're great at pattern recognition, they're fast, they can read stuff, they're incredibly powerful judgment.
Speaker B:I don't know, it's like, you know, they come to a view.
Speaker B:Is it the same view I'd come to?
Speaker B:I don't know if I would but you know, your store managers and your district managers and your, you know, the rest of your field team, you know, they have a lot of experience and we should be putting them alongside AI.
Speaker B:So the best retailers are also saying this is not a silver bullet.
Speaker B:Whatever we push out, you know, we want humans to verify, provide feedback, essentially train you, train these models further.
Speaker B:Again, it kind of, you know, Chris and Ben, it kind of goes back to the complexity of these systems, which is you now need, you know, you need the infrastructure to leverage the right models to then pump it out to the people, to then track the improvement, etc.
Speaker B:You can, you can see, you can see why there's a lot of, there's a lot that needs to go into getting it to building infrastructure before you can affect people.
Speaker B:AI.
Speaker C:Daniel, super interesting.
Speaker C:We're going to wrap here.
Speaker C:If anybody listening or watching wants to get in touch with you.
Speaker C:And honestly, if you can have a 10 minute conversation in AI and cover dragons and toilet cleaning.
Speaker C:Who wouldn't?
Speaker C:Please can you let us know how can they get in touch with you and The Corso team?
Speaker B:12 Years building AI technology for retail and I've been, I've been summarized as, you know, yeah, the experts on dragons and toilet paintings.
Speaker B:I mean, that was it.
Speaker B:It's what the.
Speaker B:It's what the last 12 years were for.
Speaker B:Honestly.
Speaker B:I think this is my proudest arrived.
Speaker B:Yeah, I think I've done it.
Speaker B:I've done it.
Speaker B:If you, if you would like a conversation, by the way, and I think Ben and Chris, I think all the audience hopefully records.
Speaker B:I love, I do love talking about this.
Speaker B:You can, you can email [email protected] you can also just go to our website, corsa.com.
Speaker B:You know, there you can book a demo, you can get a touch.
Speaker B:You know, I personally, particularly when it comes to this area, I, I'm.
Speaker B:I love talking about it.
Speaker B:So, so happy to have spend time with anyone.
Speaker C:And the team's going to be in Grocery Shop next week.
Speaker C:I understand, Ben?
Speaker B:Yeah, that's absolutely correct.
Speaker B:So we'll be there.
Speaker B:Grocery Shop.
Speaker B:Yeah, Grocery shop next week.
Speaker A:Thank you, Daniel.
Speaker B:Thanks, both.
Speaker A:All right, so, Ben, do you think we covered.
Speaker A:Do you think we covered how to scale AI efficiently inside of retail operations?
Speaker A:What was your takeaways from that conversation?
Speaker C:Oh, I mean, what a fascinating conversation.
Speaker C:I mean, I took so much from that one.
Speaker C:I think Daniel was really interesting.
Speaker C:And look, I think there's two things that struck me and neither of them were dragons.
Speaker C:The number one is context.
Speaker C:I've having so many conversations at the moment which is not about the AI, it's about the context layer and that without the context layer in your operation, in your business, you can't do anything.
Speaker C:And I think what Daniel painted a beautiful picture about is how you make that context layer in a retail store environment, which is critical to be able to tell the power of it.
Speaker C:So that was one of my.
Speaker C:Yeah, I can now start to picture what a context layer for a retail store looks like.
Speaker C:I mean, that's the number one.
Speaker C:The second one.
Speaker C:I absolutely adored the fact he talked about boring.
Speaker C:You know, look, we've got, we've got Grocery Shop coming up.
Speaker C:We know scale retail is a game of efficiency.
Speaker C:It's a game of working out how you do the same process over and over again.
Speaker C:A bit more efficient, a bit more efficient each time with an incredibly high quality of outcome.
Speaker C:That can be boring.
Speaker C:And I think we'll.
Speaker C:Daniel told us is they are perfect use cases for AI and retail.
Speaker A:Yeah, I think the.
Speaker A:That's funny you said that, Ben, because my big takeaway from that was actually.
Speaker A:And Daniel kind of came back and mentioned this in terms of what he meant when he said boring.
Speaker A:But, you know, the word I took from it was laborious.
Speaker A:It's best at doing the jobs that are truly laborious that we just don't have the time to do.
Speaker A:Like going.
Speaker A:Like with Georgina as an example, like going through survey data, like Daniel said, to survey data.
Speaker A:He said it many times, like, that's a very laborious task.
Speaker A:And so that's where it's best used.
Speaker A:But to get the power of AI, you have to provide context, and to do that at scale.
Speaker A:This was a key point for me that I never thought about.
Speaker A:You do need a harness, you know, I mean, you made fun that you talked about the dragon analogy, but it was really good because you need a harness to help you steer the ship.
Speaker A:You need some.
Speaker A:Or a steering wheel or something, whatever you want to call it, but.
Speaker A:And that, in today's day and age, has to be provided by a software system.
Speaker A:That's the key.
Speaker A:There has to be a software system that's providing that harness or that operating layer of context.