Retail AI Is Moving Fast. Most Companies Aren’t Ready

This Retail Technology Spotlight series podcast is brought to you by the Omnitalk retail podcast network.

OpenAI is kind of Pandora's box to taking that same approach that we've gotten by with for the past 25 years and just putting it on steroids.

You have a whole bunch of different UI interfaces, a whole bunch of different prompts, a whole bunch of different logic.

The duct tape use of Excel that's, you know, hampered the industry for so long could actually become magnified.

Even when they go from A to E on their own, they're going to get varying results.

It does become a mess.

And so now you're introducing that human variability into your system design.

Send us your messy data and we'll see what we can do with it in just a week.

Hello, everyone.

I am Chris Walton, your host for today's interview and interview in which we are going to explore the question that is undoubtedly on so many retail executives minds right now.

That feeling or that question of, yes, God darn it, I believe in the power of AI, so why shouldn't I just empower my teams to figure out how to harness AI themselves?

Because I know if I was still an executive, I would be fighting or even fulfilling my urge to do just that, right or wrong, come hell or high water.

So to enlighten us on the puts and takes of such an approach, I brought in an expert, Judah Berger, the AI product manager at Unframed AI, to educate me.

Judah, welcome and thank you for joining me.

Thank you for having me.

Pleasure to be here.

Yeah.

All right, Judah, so let's get right to it because I'm curious, you know, I.

Let's pretend, let's pretend for the sake of the argument that I'm still an executive in merchandising.

I'm chopping at the bit.

I'm all in on AI.

I can't wait to rush into it.

I can't wait to have my team come and bring me all these cool things they can do with it.

I want them to just be off and running with it.

What's right and wrong with that approach of this?

Telling them to do what they want to do and report back to me with any efficiencies that they might find.

Well, I would say first off, if that's your approach, you're actually on the right track.

Okay.

Yeah.

So I would say ultimately it's a bottom up approach that works best for a lot of companies.

But if you take a step back, there's really four approaches that you can take as an executive when it comes to implementing not just an AI solution, but really any sort of technological solution.

So the first off is you have an internal build, right?

This is where you have control, you got customization, you own the ip, you might be competing for talent against other companies, depends on the size of your company.

But you know, there's a lot of cost involved with this, right?

Like timelines can drag, the maintenance never ends, that sort of thing.

The other approach that you could take is you can have a management consulting firm come in, right?

This is increasingly popular, right?

You got smart people, they make beautiful slides, but you know, a few months later you have a strategy and maybe a partial deployment, but then they go home and the thing breaks and then you don't have the muscle to fix it anymore, right?

I used to work in that space and that is something which kind of ate at me.

It's like, okay, well gave that to them and now what?

Okay.

The third approach is of course off the shelf software as a service, right?

You know, or a no code platform or something like that.

It's fast to spin up.

It often demos well, it's often built for the median use case, right?

Which also means that it's not built for your use case in particular.

And then the moment that you needed to actually understand your SKU intelligence or whatever other retail needs that you have, your reorder rules, just your data in general, you're right back to building again and needing to kind of do that customization, which is a heavy lift.

The last approach is what we do here at Unframe, and we'll talk about this much later.

First we want to talk strategy is the managed AI delivery, right?

So that's where you work with a partner to kind of ideate on what are the issues to deal with, how to, how to approach it in a milestone rollout and just build on whatever it is that you want to implement.

So this way it cascades into this wider implementation as opposed to something off the shelf, right?

And then the firm you work with designs it, builds it for you and runs it as a managed service.

And it's tailored to your business.

So you get the speed often of off the shelf, the customization of an internal build.

But you also don't have to deal with any things where a high upfront cost in terms of, you know, usage or seats.

It's really just often, hey, we pay for what you get and this is what's contained within it.

And we know exactly how it's going to work as like a yearly subscription.

So the risk is often on the company who's delivering that as opposed to on your company.

Got it, Got it.

Okay.

And I want to get into that and I can tell which way you're going to roll already kind of the answer to that question.

But you know, before I do that, I want to.

So essentially.

So those are four very different approaches.

I want to get into as many of them as we can.

And so going back to the initial question though, there's actually a fifth approach too, which is I could just leave my team to just try and do whatever the hell they want to do too.

Right.

And the thing that I struggle with, you know, if I were to do that is there's probably a lot of good ideas that would come from the bottom up.

But the question I always come back to is like, how is I as a manager going to effectively scale those ideas and make sure that they can work over and over again, that they're tested, they're qa, that they're not going to cause any problems in the long run?

Is that, is that a real concern?

Like, should I as an executive be like kind of pumping the brakes to some degree on that approach for that reason or how do you think about that?

Well, like a good product manager, I'm going to say it depends because it really does come down to the situation that you're trying to solve for.

So if it's something where you have a very small team and they have a very repeatable process and you don't necessarily need something that needs to be 99 point, you know, 2% accurate, just something that gets them 80% of the way there and then they fine tune it themselves.

Encouraging that kind of experimentation is how you enable your team to do their best work by strapping them into certain constraints.

And, and of course constraints and guardrails are important, but if you don't enable them to experiment on their own, you don't get that those bottom up ideas that the top level executives don't necessarily always have insight into.

And this can be everything from I have this repetitive process that I spend three hours a week on.

Two, there are much wider things repeatable not just throughout that team, but throughout the organization.

Now I would say for the former, you should enable them to be able to do that.

You know, use Claude code, use all these other more generalized models to play around and you'll.

It's really interesting to see what people come up with.

And I've seen that across a bunch of different organizations where a lot of the best ideas did start out as a seed from Someone's own tinkering and their own experimentation, as opposed to something that is much wider.

On the latter half of that, what you find is that many organizations do have these repeatable processes, and asking one person or an internal tech team to build that, it does become chaos because you can whip up a demo pretty quick, you can test it on 10 deliverables, and it seems to be working great.

But then once you scale that up across an org and it has thousands of instances, you, you don't have the guardrails, you don't have the maintenance in place, it breaks down quickly.

And then we've found that when we've talked to people in the past and they've decided to go their own route, we have reengaged talks with them a few months later after a lot of these issues have popped up, because addressing them does take a lot of resources that people don't necessarily keep in mind when they're first whipping it up and very excited about what Claude can do as an example.

Yeah, and I think it's really important because I get kind of nightmares thinking about, like, you know, when I think about how particularly retail, because that's where my experience lies.

But when I think about how, you know, retail technology has actually developed inside of an organization for the past 20, 25 years.

A lot of it has been cobbled together or glued together by people taking initiative and doing creative things with Excel.

And so now a part of me just worries that OpenAI is kind of Pandora's box to like, I don't want to say Excellifying the organization, but taking that same approach that we've gotten by with for the past 25 years and just putting it on steroids and not knowing exactly if we're going to get to a better place in the long run if we don't think about this approach right from the get go.

So am I, am I right to have those nightmares?

Am I right to stay up at night?

Okay.

Why, Chris?

No, you nailed it on the head.

Because that's actually one of the biggest, biggest things that we've seen is you have people to solve a given problem, have each come up with their own implementations.

And as you said in the past, a lot of this was custom Excel workbooks.

One, one client that we worked with had a whole bunch of different instances where people were basically creating plans for, not to give away too much information.

Yeah.

You know, in, in, in the real estate industry, a plan going forward for, for, you know, from an early stage.

And a lot of it was the.

But at the.

But once they actually apply their own logic, their own formulas, it was really a lot of varying.

A lot of varying results.

And frankly, when you have that on AI, it's just supercharged.

You have a whole bunch of different UI interfaces, a whole bunch of different prompts, a whole bunch of different logic.

Basically a lot of different brains who are all.

And not just one brain like in Excel, but instead potentially thousands of brains through all the different steps of the process that are all compounding to get these different results.

And if you want consistency as an organization and in how you're putting your foot forward in terms of what you're delivering to your clients, that's not gonna be scalable and it's not gonna be an approach that you wanna take.

It's good for small problems and it's good for putting duct tape on small little things to keep things running efficiently.

And there's always gonna be room for that, and there's always gonna be spaces where you wanna encourage that.

But what helps is when you have either someone within the organization or someone you partner with to understand, Oh, I see.

Everyone's using duct tape on the same kinds of seams here.

What can we do to scale this up and have everyone on the same page?

Got it.

So that's really interesting, actually, if I play back what you just said, you're saying that the duct tape use of Excel that's hampered the industry for so long could actually become magnified over the next 10, 25 years if we're not careful about what we're doing.

Well, if you take a standard process that you build this for, let's say it's A to B to C to D to E, and maybe Excel, you had certain custom formulas, but it was still relatively within the same ballpark.

And okay, maybe you have different results, but you deal this with any sort of just letting people go ham with AI, then each one of those steps has potentially dozens of steps or those thinking mechanisms even within the same person's process.

So even when they go from A to E on their own, they're going to get varying results.

And then you compound that across a hundred people doing this.

I mean, it does become a mess.

Wow.

Yeah.

That's really crazy because.

Yeah, one of the great.

You're right.

Because one of the great things about AI is that it interprets human variability and how people want to interact with it.

And so now you're introducing that human variability into your system design if you're not careful.

So.

Wow.

Oh, my God.

Yeah.

And it's Everything from the data you're putting in and the prompts, the instructions, the design, there's really a lot that can go into it.

Yeah, right.

Oh, my God.

So, yeah, you could create a monster if you're not careful.

This is important.

I think I. Hopefully people are hearing this right from the get go.

This is, this is important.

You don't want to take this lightly for these reasons, but.

All right, well, so let's say I'm an executive who has listened to this podcast and now I'm kind of on the realm of which you've gotten to me, to Judah too, of like, holy crap, I need.

I have.

I. I know I need some help now because I don't want to open Pandora's box.

So, so where do I start?

Do I hire a management consulting firm?

What are the pros and cons of doing that versus maybe the other approaches that you talked about?

Let's.

Let's go into that in more detail.

Okay.

Yeah, sure.

So let's start with consulting.

You hire a consultancy, they come up with all those beautiful deliverables, and you will learn things.

These are very smart people who do have cross industry insights or deeply within an area of expertise.

I'm not saying that they're not.

They have no value.

That's not at all what I'm saying here.

With some of the people, some of the consultants I've worked with, I've generally learned a lot from, in terms of how to break down a problem and how to approach it.

But if you're talking about what it is that actually does get stuff done, they're not on the hook for that.

Right.

And that is one of the things that does.

Consulting is not necessarily best.

You know, maybe on a strategy it can, it can help.

But in terms of actually building things and delivering working user experiences for your employees, for your customers, that's probably not the way you want to go.

Both on speed, both on understanding what the actual user experience should be.

And that just comes from, frankly, the iteration of actually putting something on the ground.

No matter how much planning you do, no matter how many surveys you run, no matter how much, you know, analysis you run on your data, until you actually put something in people's hands, you haven't learned as much as you can.

So the speed to putting things in people's hands is something that will teach you more than anything else.

So let me put you on the spot, Judah.

I'm curious, like, is that because you have a product manager bias in answering that question, or do you think that Bias is real and justified.

Well, I think, well, speed can also work against you.

If you don't spend enough time understanding the problem, then you're building an ax for something where you really, really need a hammer.

And you're going to find that, yeah, maybe you might be able to make some dents in hitting a nail on the head, but you can't do it very effectively.

And I think as a product manager, one of the things that is most important is to make sure that you spend your time sharpening your axe and making sure that it's the right tool.

And then once you actually need to execute, you do that with speed.

That's where you kind of mix a lot of the insights that someone like a consultant would have in terms of approaching a problem through a structured framework and understanding what it is that the actual root cause is here.

And I can't tell you how many times we see that where people come with what they think is a certain issue.

But then when you really dig down into it, when you really get your hands dirty with the data, that you start to understand, oh no, actually there's other root causes here at play and we can make things to address those issues that ultimately have upstream effects and make it a lot easier to have the end result that you want.

Yeah.

And for fair play to you for answering that question as well as you did.

And truth be told, to the audience, like, I actually believe in the product manager bias too, because I generally feel like that is the right approach to answering a lot of the problems that come at us every day in life, no matter what it is.

But all right, so what if I try to centralize this?

Like, what if I say, what if I'm a large enterprise organization, I've got a large tech team, got a well paid or paid C suite, executives that are, you know, in charge of this and figuring it out.

What if I say I want to grow this all in house?

Like, what do I need to.

What is right and wrong with that approach?

Particularly so what's right about it is there's often nobody else who's better positioned to either talk from a user perspective, whether it's an internal team or customers, they have, they spent all day in it and they know they often do have a better finger on the pulse than anyone else of what users actually want and what the problem is.

And in terms of the data also, once again, these are, these are databases and schemas that you work with every single day so you know how the data works, where the gaps lie.

And that does often mean that your team can build effective products.

And that's why many teams have internal teams.

Right?

And there's, and there's always going to be a need for, for that.

But the hard part is, is that there can often be a disconnect between what you're able to implement and what you think you need and what you want to implement.

And why the partnership model can bridge that gap is because there are people with certain technical chops and have also seen a lot of those common issues across a given industry, a given role, given certain type of data, and they can bring that expertise when it comes to dealing with whatever it is that you're trying to deal with.

Now, I know in retail there are specific problems that come to, that come to mind in terms of inventory tracking, SKU intelligence reorder, there's a whole bunch of routes you can go down.

But there are a lot of common problems that maybe someone might face in manufacturing or in finance or in real estate.

And without that kind of cross industry pollination, there are certain, I don't want to say limits, but certainly opportunities and opportunity costs in terms of what you can develop.

And that's where a managed AI solution does thrive because they're able to see a lot of those commonalities while also listening intently with an empathetic user perspective on, well, you tell us how your users work, you feed us as much as you can in terms of the context and we'll feed you back the context that we have in terms of the solutions that we've implemented and get the best of both worlds.

So Judah, walk me through that then, like, walk me through that soup to nuts like, so it sounds like, it sounds like what you're recommending is kind of a, is a kind of a hybrid approach to answering this question about how to deploy AI inside an organization.

And maybe I'm using the wrong word there too, and I hope you correct me if I'm wrong, but walk me through like, you know, start to finish, you know, as much as you can in as much detail as you can without, you know, getting too detailed, of course, but what then is the right way?

So if I'm an executive and I'm saying like, okay, I want to go after this, what would you tell me to do?

So I think the first part is when you, when you partner with an organization who does this sort of thing is scoping out what exactly needs to be done right.

And this is the sharpening of the ax that I referred to earlier of understanding the problem, understanding the Data understanding what the situation is, understanding where the gaps lie and really, really getting a hold on and of what the problem is and giving shape to it.

And through that, that's a very investigative process.

Now it can be anywhere from, you know, a conversation or two where it's very well known throughout the organization.

Yep, this is the problem.

Everyone agrees and then once you share the data, it's, yep, okay, we get it.

That is definitely an issue.

And it could be a more circuitous process.

Right.

We think it's A, it's really B, but actually it's B2.

And then, okay, maybe it's B melded with C. Oh, okay, great.

Now we have a feeling of what it is and that partnership is what is fruitful for these kinds of investigations.

And it's not A, we're going to come in and tell you what you need to do.

And it's also not on the only one, one way to us of, hey, this is what we need.

Don't question it.

Right.

It really is bidirectional.

Once you've scoped that out, then it's really understanding the data.

Right.

What does, you know, what do we have available to us?

What do we need to enhance?

What do we need to clean up?

How, how does that look?

What is the schema that we need to be able to be reading for the AI to have a good job at answering it?

Basically, like if you're onboarding someone new, if you showed them your data, what would it be that you actually need to look at to actually get the insights that you do?

And there's always been a lot of idiosyncrasies.

But the, but I would say that that's often imparted during this process.

Then they go ahead and they build something that is basically a first pass at, hey, we're taking this data, we're transforming it, giving you the insight into it, whatever it is that you need to solve the problem.

And then let's test it.

All right, how to do.

And let's compare it to what you think it should be.

Oh, it's not as good here, there and here, there and a couple other places.

Great, let's go ahead and fix it and iterate on that.

And that's those couple of paths of iteration.

And then once you've proved it out, okay, this looks good.

Now we can scale this up across the whole organization.

And a lot of different companies have different models for how they do this.

Some charge up front, some charge only when you're happy.

Well, you know what we do Here is we do a proof of concept where we take a small part of the larger process.

Now that can be just like one example, end to end.

It could be focusing on one stage of a multi stage process and proving out the capabilities there and then saying, if we can get you the confidence that we can do this, we can definitely do the rest of the rest of the, you know, eat one piece of the pie and then we can handle the whole pie after.

So that's typically how these interactions go down.

And of course they can vary from situation to situation.

Maybe there's an RFP that's already scoped out, maybe you're working on a whole bunch of different cases at the same time.

But ultimately it just comes down to instilling the confidence from the user perspective that whatever is being recommended, I understand how I got to this conclusion.

And the AI is explainable, it's understandable, and I can see where this goes.

Right, okay.

Okay.

So, so the, the interesting thing about that, like the way you describe that to me, it sounds, you know, it sounds very consultative in approach, which I would expect.

But where does the tech come into it?

Like, do you need, do you need serious tech chops to be able to pull this off, to do this the right way?

If somebody is reaching out to you or any other partner, you know, in this vein to do this, and if so, like, what is under the hood that is required from a technology standpoint to be able to do this better than someone else, let's say, if I'm taking this approach.

No, it's a great question.

So I'll tell you what we do here at UnFrame in terms of the pillars that we have underneath our managed delivery platform.

So the first one, and this is one of the most important ones, is the reusable building blocks.

So we have proprietary engines for things like extraction, abstraction, structured chat retrieval at scale, predictive modeling.

And we've built this multiple times and we can redeploy them across different clients.

Now, not all of our clients, we can share everything we've built necessarily.

Sometimes we build something for someone we're under a certain contract, but ultimately a lot of this stuff can be reused.

And that's why we end up shipping in days instead of quarters, so you don't have to redo all the plumbing every single time.

So that's the first part is these reusable building blocks.

The second part is what we call knowledge fabric, which is really just the context, if you want to think of it that way, that people build up in their Heads, but we make it, we give this to the AI.

So this is what makes it fluent in your business.

It's a governed data layer that connects to your erp, to your data warehouse, to, to your file systems and it gives all of the AI processes, whether it's agentic or whether it's more deterministic, you know, precise control over what you want to give it access to and what it's it see, it can see and what it can't see.

So let's say I want one agentic process just to focus on the merchandise data and not the financials.

You can draw that boundary explicitly and that's really good for enterprise because a lot of chatbots today that are just bolted onto an existing platform that you're working with don't have that and don't have that insight into other, other databases.

And we do integrate with all, all the data that we need to, whether it's one system or 20, we can do that.

So that's two is the knowledge fabric, the third is, and this is very important, is the auditability and the traceability.

So this goes back to instilling the trust that people need to really throw themselves into a solution once they start using it.

Because if you have every answer that an agent or anything that's using AI gives that it can be traced back to the source row, the document, the source system, then it comes with reasoning and every iteration is logged.

So you can't ask someone to trust a black box.

And that's ultimately how you build a high adoption rate is when I am interacting with this, do I understand how it got to this answer?

And if the answer is yes, then that's how you're going to get adoption and also be able to fine tune it and iterate on it to make it better and better going forward.

So that's the third part is this audibility and traceability aspect.

The fourth is in Agent Studio.

So this is basically the customer facing layer that lets your team adjust, extend chain agents without necessarily having to come back to us for every single little thing.

You can change instructions, you can change the data access per agent, you can plug in your own tools, you can expose it as an MCP endpoint for other tools to use.

So you have this full customization and these full audit trails and this full governance all put together in a way that, oh, okay, I understand how I can orchestrate that.

Now we're not just a SaaS that just throws it at you and says you figure it out.

We work with you to implement These to understand how they work, to understand best practices.

But we also don't have to be at every, every time something goes wrong, coming in and having to fix it and devote resources, which introduces lag time.

So it's that combination of empowering you to have the control over the processes that you're creating while also giving your expertise in how we've implemented these both in your space and in similar spaces or other spaces solving similar problems.

And around all of that is our operational layer.

Right.

So that's the unframe company, that's the bug fixes that we do within hours and iteration cycles every single week we stay on call and it's not exciting, but that is ultimately what's needed to keep a piece of software successful once you've already handed off and as your organization moves and data changes and you know, nothing stays static.

So that's ultimately, and I know that was quite a bit of an answer, so happy to dig into any of those deeper.

But that's really our approach on how things work under the hood here.

Can't speak to another organization like ours per se, but that's how we do things at On Friend.

No, but I think that's a good way to encapsulate why the technology that you're bringing to the table is important in this conversation.

Because the part I never thought about, which is in what you said, and I want you to again correct me if I'm wrong, is if you take this approach with the deployment of AI inside your organization.

The beauty of AI as it stands is actually there should be a much better audit trail to understand what you got right and what you got wrong that continues to execute or doesn't execute as you go forward.

Is that right?

Yes, I would say that ultimately that is one of the most important parts of this.

Right.

Because if you want to understand how it's evolving, what things are changing and if you have a self improvement process in there, what's being changed, especially if it goes off the rails in ways you didn't expect, you need to be able to understand why.

And if you don't have that, then you're going to be grasping in the dark when you're dealing with these non deterministic models.

Mm, mm, Yep.

And again, hence the part where you probably don't want to just be doing this on your own because you're not going to have that level of fidelity too.

I would think if you, if, if you are trying to approach it that way.

All right, so to that point then like I was just Like I literally just said, how do I know it works?

Like how do you know how put your money where your mouth is, how do you know this approach works?

So I mean, what would it help?

Would it be helpful to go into a specific example with, for sure, with one of our, with one of our clients.

So we worked with a major shoe retailer where they were over 40 years in the business.

They had over 140 stores, over 2,000 SKUs across, shoes and bags, accessories, super seasonal.

And they knew they needed AI, but they didn't quite know where to start.

Once we talked to them, that's when we understood that a lot of it was an intelligence issue.

In terms of where are my shoes, how are the shoes doing?

Right.

These are simple problems, simple questions to ask, but the answers are not simple.

So they could really only realistically handle a very small portion of the SKUs.

Right?

So they have to pull the data manually, interpret it, take a look at what this means, compare it against all these different stores.

They might see a couple outliers and try to manage it as best they can.

And let's say they launched a new shoe, it had a great sell through rate of 45% in its first month and which is basically screaming for hey, reorder me now.

Like you need to start rolling this out.

But nobody caught it in time.

So that you know, is stocked out, momentum was killed and they were burning through multiple dozens of thousands of dollars every single month, just physically moving the unsold inventory between all the stores and the warehouses, sometimes even moving around the same product two or three times.

So this is all capital that's being tied up in trucks, right?

What we built for them.

And after really understanding a lot, how do you interpret the data, what are the different thresholds that you would need to, you know, classify X as Y and just really take spending a lot of time there understanding and making sure that when we are baking in this logic that we're doing it correctly.

Then once we did that, and that took a couple of weeks, within 15 days we had taken a proof of concept and built it at a production level ready solution where every single morning they could wake up and they could see all of the different solutions.

Oh sorry, excuse me.

All of the different SKUs that were most important for them to action on.

And I can go deeper into what that experience was like.

But the ultimate outcome here was that they saw over a 40 times ROI per dollar invested, right?

Because they were able to more effectively make sales more able to allocate their resources in a way that that worked for them and this came down to those four pillars.

Right.

So it was the fact that we were able to take these reusable building blocks, implement them quickly.

It's because we were able to have control with the knowledge fabric on where each of these agents that we set up to do this analysis was looking and constraining it to the right data, auditing it and tracing it.

So that when they asked a question in the iteration process they knew, oh well actually I wouldn't define it that way or oh, it was pulling from the wrong store to set or that actually is that maybe it missed out a couple things.

Oh, we need to include this great through that.

Because it was all traceable.

We were able to do that also very quickly and then they're able to make changes themselves where needed using our agent studio.

So that's one example.

And honestly I would love to go into more detail on that one because that was actually a really interesting implementation in the retail space that I think some of your users might be, some of your listeners might be interested in.

Yeah, I mean, let's do it.

Let's do it.

I mean I love whenever we can go a level deeper with the audience to, you know, to kind of prove your mettle, so to speak.

So yeah, let's do it.

Like tell me what, what detail do you want to share about that?

So yeah, what we, what we built for them was basically the, when they first get to the page, think about it, you know, to kind of illustrate it in, in the minds of the, of the listeners.

Yeah, you have a daily brief, right?

It's like you walk in and it's as if you had a team of analysis, like analysis, sorry, analysts, excuse me, analysts going forward and saying hey, this is what you need to, this is what you need to action on and why.

Right?

So there's, you know, the, the category manager and the regional manager would open it up and see in a tabbed experience all the SKUs.

Everything needs to be reordered.

Everywhere is dropping in sales.

Everything that's trending up in fastword.

Everything needs to be distributed across different stores stores due to, hey, it's selling good here, but not, not so well here.

We call local heroes, right?

Is are these, these products that are punching above their weight right at a single store and are ripe for promotion or expansion.

One example that you know, we actually saw recently was there was a 69% sales drop in on, on a given SKU, right?

So it only had 55 days left of inventory at its current Velocity and they had a reorder period of like 40 days.

So they, and they had no incoming purchase orders.

Right.

So pre system that would have been caught maybe in a few weeks and then that'd be too late.

Right.

Because then they, they might put in the order.

Maybe not enough, they're not entirely sure.

But ultimately they wouldn't have been able to act on it.

But with the intelligence that we're able to offer them, they did.

Right.

So with, you know, multiple weeks out, they're able to easily plan, put in the order and distribute it and track the planning for how it does in other stores.

Right.

So this is basically one, it's the action items where you can understand this is what I need to do and why.

And there's also explanations from the AI analyzing and recommending as well.

Hey, this was selling really well here based on seasonal trends that we see, based on geographic trends like hey, this is a great beach shoe.

There's here are all your other stores where located on the beach.

Recommend you, you know, clicking a two but two clicks of a button, boom, you can send it to that store and we fill that out to kind of go into their system really quite easily.

So that was the daily brief and I think that was one of the most important parts of what we were, what we were putting forward.

But much deeper than that, that can't be the only thing, right?

You need really a way to interact with your data in an intelligent way.

And that's where we had the inventory matrix.

So this is a heat map.

It goes from everywhere, from across all the stores by size and color or individual, you know, or individual stores.

So you can see how a given SKU is performing, right?

You can see the sell through, you can see the margin, you can see the transit inventory, everything you need to know about a given sku.

And within that you can go even deeper where you could see exactly predictive trends on how we think it will do based on, based on previous, based on previous performance of the sku.

And you can interact with it with a chat, right?

So you can ask it real questions and get traceable answers, create charts, kind of just create tables, whatever it is that you want to do in terms of sending around that information internally, presenting it to buyers, whatever it is that you need, and getting a level both at the SKU level and at the store level.

So you could see how a store is doing, you could see how a SKU is doing.

And it's seamless transition between them allows you to kind of see things that even this is, we may not even brought up on, on the daily brief, right?

So with these three tools combined and then chat on top of that so you can just interact with it in natural language, it allowed the, the buyers to move on their products so quickly that they really just felt the efficiency gains nearly immediately.

Now, this was a case where the dividends paid off very quick, you know, in such a quick manner, because they were, they were stuck, you know, with, with their previous solution on something that, you know, looks like it came out of the, you know, the late 90s.

But to a degree, we all have these within our organizations and we, we have things that get the job done.

Like we have a car that gets from A to B, but we really, what we really want is a rocket ship that gets us to not just B, but also A, C, D, E and F. And I think in this case, this really did unlock those capabilities.

Yeah.

God.

Oh, my God.

You opened up a whole host of questions for me here at the end of the podcast.

And so I'm gonna, I'm gonna get you out of here on this because, like, you know, like, one of the things that, that to me, like hearing this conversation, if I was an executive trying to decide what to do, the, the thing I'd want to get a handle on, I think most of all is like, what is, what is my process?

And, and where are the gaps?

And one of the things I think about too, in that regard, which I could devote an entire another podcast to, this concept would be like, man, if I'm a retail executive of a current retail, current retailer, I'm scared of the AI native retailer.

Like we used to be scared of the digitally native retailer.

Now I'm scared of the AI native retailer that is doing all of this from the get go.

And so my question for you would be, as I'm trying to understand the gaps and all that, what is.

If you could offer one piece of advice to the executives listening, like what.

What would that piece of advice be?

As they're trying to wade through everything we've just discussed so far.

So that's an interesting, it's an interesting question because just like you, my mind goes in a bunch of different ways.

You know, what could you solve?

And there's never any one right answer.

There's often a lot of good right answers, which is, which is frustrating, right, because you want to also make sure you're moving with the times and things are accelerating so quickly.

One place, if you really want to start and you want to wrap your head around a Problem that seems doable, doesn't seem too scary, is there's really no magic to AI when it comes to implementing it.

Now, how AI works, nobody in the world actually knows.

That's a whole separate story.

But it's more like plumbing, right?

It's understanding that, okay, if we can just put these, if we can put these pipes together and kind of have this flow through here, that's, you know, and kind of, you know, configure it in just the right way.

Okay, great.

Now what problems could you apply that for?

The, the simplest ones would be that if you could explain a certain problem to a new hire or an entry level employee in, you know, given, you know, given the right tools and the right training within like a few hours or a few weeks, AI can probably do that right now.

So it's your job isn't necessarily to imagine like where AI is going to be in 20 years, because that will happen anyway.

It's more about to look at your own organization, ask what are my people doing today that they wouldn't be doing if they could just push a few buttons?

Right?

So it's, you know, if you look at it more as an operational discipline instead of like a moonshot of like, I need to transform everything with AI, it can seem very scary.

It's, it's really just taking one step at a time and looking and examining, okay, what's a repetitive process that's really a pain.

If I gave each of my employees a thousand entry level grads, what could they do that would speed up their process?

And ultimately a lot of people do have something that comes to mind when you frame it like that.

Instead of like, how can AI help you?

I don't know what AI can do.

Is it like a super intelligence that can answer all of my questions at the drop of a hat, or is it like just chatbots?

Like, oh, I only use ChatGPT.

And that is it just chatbots.

Is that everything that you can do?

And I would say when you take that approach, it does seem, hey, I don't know where to start, but if you just kind of take a step back and you think, okay, what's something that, if I had, you know, a whole army of people on this who were, I could explain something relatively easily and clearly they could do this.

They're transforming data from one to another, rooting through spreadsheets and spotting some trends, you know, coming up with emails on how to interface with suppliers or with customers.

These are all things that can be done.

And you do want people in the loop.

Don't get me wrong.

I don't think you should be cutting out people out of this entirely, but more maximizing the efficiency and how long it takes to get to a result and to a successful outcome.

Yeah, that's a great nugget to take away for those listening.

I like how you said that.

As you're thinking about AI, you should be thinking about it first and foremost as an operational discipline, as opposed to a moonshot transformation effort, so to speak.

I think those were close to your exact words.

Yeah, operational discipline versus a moonshot effort.

So, yeah, it gives me a lot to think on, and it actually makes it seem more bite.

Bite size and attainable to.

Right, Judah.

I mean, that's the beauty of it.

Exactly.

And that's what.

And when we do go in an engagement and we have.

And it's more starting from scratch, that's the kind of approach that we take.

And usually within a couple calls, we've landed on something.

Yeah.

All right, well, that was.

That was really interesting.

I mean, you gave us.

Gave us a lot of food for thought throughout that entire conversation from start to finish.

If.

If people want to get in touch with you, particularly Judah, or if people want to get in touch with anybody at UnFrame AI, what's the best way for them to do that before I let you go?

So, sure.

Best place to Find Me is LinkedIn.

So my name is Judah Berger, and I'm an AI product manager here at UnFrame AI.

You can also find us online at UnFrame AI.

And honestly, look, please reach out to us.

We're always happy to walk through what's possible, and after a call or two, you can send us your messy data, and we'll see what we can do with it in just a week.

Send us your messy data.

I love that.

I love that.

All right, well, that wraps us up.

Thanks to the Judah Berger for joining us today, and thanks to Ella siryork for producing today's podcast as she always does.

And on behalf of Ella, myself, and all of us at omnitalk Retail, as always, be careful out there.