RETAIL

Christina Cardoza: Hello, and welcome to the IoT Chat, where we explore the latest developments in the Internet of Things. I’m your host, Christina Cardoza, Editorial Director of insight.tech. And today we’re talking about developing AI retail solutions that matter with two special guests, Ria Cheruvu from Intel and Nicole O’Keefe from Pathr.ai. I’m very excited to dive deep into this topic. But before we get started, let’s get to know our guests a bit more. Ria, I’ll start with you. What can you tell us about yourself and what you do at Intel?

Ria Cheruvu: Sure. Thanks, Christina. Hey, everyone. I am an AI Software Architect at Intel Corporation and an AI Evangelist. I’m very passionate about being able to speak about, develop, and build cool things in the AI space. I’ve also got a master’s in data science, so, pretty much interested in all things AI.

Christina Cardoza: Great, excited to hear what you have to say about the retail space. But before we get into that, Nicole, what can you tell us about yourself and Pathr.ai?

Nicole O’Keefe: Hi, everyone. Thanks so much for having us on the show. I’m Nicole O’Keefe, I’m a Senior Product Marketing and Operations Manager at Pathr. I lead our marketing efforts here at the company and also project manage all of our deployments. And really excited to dive deep on this topic.

Christina Cardoza: Perfect. Excited to have both of you here with us today. Like I teased in my intro, talking about retail solutions that matter. You know, AI comes with a lot of opportunities and a lot of promises and benefits. So sometimes I feel like businesses or organizations can just be wanting to implement AI, to implement an AI.

But today I want to talk about really solving the pain points and the challenges that we see in the retail space, and how we can implement these solutions in a smart and meaningful way. So, Nicole, I’ll start with you with the conversation. Obviously the retail landscape has changed quite a bit over the last couple of years. Customer expectations have evolved. So, curious from your side of things, what are you seeing in this space? How is it creating challenges for retailers? Where are their biggest pain points on retail floors today?

Nicole O’Keefe: Yeah, I think that’s a great question to start us off with, Christina. You know, as customers we want that seamless checkout experience—I think short queue lines, short wait times. Imagine going to the store, you just finished shopping, and you’re faced with long queues at the checkout. And I think a lot of us have been in that position before, and you start to get frustrated. You think you’re going to spend even longer just waiting in line, and it may come to a point that you just abandon your cart and leave the store altogether.

Now, for retailers that’s really the last thing they want for their customers, because cart abandonment can lead to lost sales; it can affect their customer loyalty. And on top of that they’re concerned with rising labor costs and labor shortages in the workforce. And all of these factors are really contributing to the challenges that we’re seeing in today’s retail and in the retail world. And it really presents the need for them to create an efficient checkout experience for their customers.

Christina Cardoza: Yeah, absolutely, and I certainly agree. Nothing is more frustrating than wanting to run into the store to just grab something really quick and then having to wait on line. It puts a bad taste in your mouth, sometimes, with that store, that retailer.

So I know we also have self-checkout solutions that have been implemented, but that doesn’t really solve the problem. Because sometimes you don’t scan something right; you have to wait for a store operator or a cashier to come help you. Sometimes they’re helping with a different member. So there doesn’t really seem to be a good solution right now, or it’s a pain point I can definitely see in the retail space.

So, Ria, I’m wondering how does AI come and start to address some of these pain points for retailers and try to make this more actionable?

Ria Cheruvu: Definitely, and I think, Christina and Nicole, as you shared earlier it’s just so fascinating to see how these pain points turn into actionable solutions that we can propose. I think one of them, Christina, you just mentioned is basically the integration of different solutions that we would need to apply.

So, artificial intelligence and the solutions that it’s powering are definitely very helpful there where you’re able to integrate multiple solutions and develop models around automated self-checkout, intelligent queue management, identifying and better understanding customer experience, and integrate that into a system that’s actually providing you with valuable insights across multiple stores, customers, and scale that. So we’re really seeing AI helping with that scaling effort, as well as integrating all of these different functionalities together.

Christina Cardoza: So we know sort of what we want to implement in the retail space and why we want to do it, but how do we actually go about doing this? A lot of times retailers, they don’t have the development team on staff that are able to build these AI solutions. AI, it’s becoming more accessible, but I think to implement these and to really make it valuable it’s still a little bit more specialized the way that we do this.

So I’m curious, Ria, if you can expand on how you’ve seen developers build these solutions, or how developers can successfully build these type of applications and work with a retailer to implement it in a way that not only solves the retailer’s problem but they’re building these applications that address the customer expectations, privacy, and security that some of these users may have.

Ria Cheruvu: It can be a challenging question both from the technical limitations of these models and also the types of use cases that they’re trying to satisfy, and then bringing in the different domains like privacy and other elements. I think, just speaking to the first two first, there’s definitely a lot of models and solutions out there that can be put to the test when it comes to these use cases.

We can think of being able to identify the number of items in stock on a shelf, or identifying when someone picks an item off the shelf, puts it into their basket. Even something that really fascinates me is the way that we can integrate AI in the different options, whether that’s a smart shopping cart, smart shelves, smart robots that are kind of monitoring and zooming through the store and multiple different elements. And really a lot of these times we can leverage models that are off-the-shelf and also use technologies to train and build our own models, which gives us a lot of flexibility as to how much time and commitment we want to put into this.

Just briefly speaking to privacy and anonymity and a lot of these security and other elements as well, I think it’s a really critical debate and conversation around how exactly do we incorporate those themes into these algorithms, and also how do we use these algorithms for those purposes as well. For example, whether that’s blocking individual’s faces, respecting their privacy with regard to items they’re purchasing, and basically maintaining that form of anonymity while also extracting the insights needed to continue to improve these algorithms.

So it’s definitely something of a major theme around this, but we are seeing with the improvement and the number of AI technologies coming out that this is becoming easier and easier and allowing us to have broader conversations for the retail space.

Nicole O’Keefe: Great point, Ria. So, protecting consumer privacy is really top of mind for retailers today. Here at Pathr we have our tagline: you can learn a lot from a dot. And every dot is a shopper moving around the floor plan, and that dot has no personally identifiable information attached to it. So retailers can really leverage these privacy-preserving insights to make business decisions in a very unbiased way.

Christina Cardoza: One of the things that I loved what you said in the beginning of your response is we’re solving queue management with these intelligent queue management systems and these AI solutions, but it doesn’t stop there. You can take a lot of these algorithms and apply it to different use cases and improve other areas around the store. So, want to get into that a little bit more.

But before we get there, Nicole, I’m curious I know, Pathr.ai has experience and a history of working with retailers to implement some of these solutions. So what are the business value and benefits you really are seeing when you work with customers, how you guys are implementing AI?

Nicole O’Keefe: Yeah, no great points, Christina. And I think Ria touched on some really key thoughts earlier, around how are retailers and developers implementing it today. And I think one of the ways they can do that is through spatial intelligence. And spatial intelligence is all about measuring how people move and behave inside physical stores.

Here at Pathr we leverage retailers’ existing camera infrastructure, and we’re able to provide insights throughout their store and in particular around their checkouts. Understanding how long are the queue lines? How long are people waiting? And so for retailers it’s really important for them to kind of narrow down to: I want to reduce my operational costs; I want to improve operational efficiency.

And when they leverage these insights they’re doing so in a very data-driven way. They’re leveraging AI to make these decisions that really impact the growth of their business. Anything from allocating their resources more effectively, reducing unnecessary staffing costs, and they can even understand from a register point of view how many registers are being used in a day. And if they’re not being used often, well, maybe it’s an opportunity for them to turn that space into the sales floor and add more merchandise there.

Christina Cardoza: Yeah, those are some great points. And I love that we’re not only—with spatial intelligence you can not only see these queue lines and deploy more workers, but this can be used throughout the whole story, really creating a data-driven culture and end-to-end experience that you can see where the foot traffic is, where the products are at stock or low, or where it makes more sense to put certain products. So you can really start improving just by adding one piece, and starting with intelligent queue management you can really start building onto it and improving retail operations.

And of course I have both Intel and Pathr.ai on the podcast today because this is not something that I think one company can do alone. At insight.tech we always talk about this theme of “better together.” I think it really takes partnership and ecosystem expertise from other organizations to really be able to do this in a meaningful and successful way.

So, Ria, I’m wondering how Intel not only helps developers build these applications but how do you guys work with partners to make sure that companies and partners like Pathr.ai can really deploy these for their customers in a meaningful and valuable way?

Ria Cheruvu: Absolutely. So, our teams at Intel are very passionate about being able to build out technologies and also provide foundations for our partners like Pathr.ai to then take that forward and then innovate on top of. Some of the approaches that we’ve taken are definitely in terms of the software approach with Intel OpenVINO^™ toolkit providing a number of different optimizations and options for being able to build and deploy your AI models.

We’re also looking very closely at the end-to-end stack and how Intel hardware can definitely help accelerate a lot of the pipelines and large computational requirements that can be required for these types of use cases, especially at this scale.

So those are the two elements I’d say that we’re definitely very, again, passionate about and happy to help our partners with taking that further and then innovating with their customers and for their use cases.

Christina Cardoza: And I know Intel, you guys also provide a lot of resources and notebooks for developers to get started with this. So where would you point developers? How could they get started building an AI solution that is really going to be implemented in a scalable and flexible way across businesses—but where do they start? How do they start learning and practicing and building these algorithms and models and working with OpenVINO and getting more familiar with that Intel hardware that you talked about?

Ria Cheruvu: Sure. I definitely point them to the OpenVINO notebooks GitHub repository, which has a wealth of information regarding how to get started with OpenVINO, how to build these applications—production grade, because it’s always fascinating and very convenient to have both your learning journey and also your production, deployment, and scaling journey integrated into one experience.

So we’re really excited to be able to share that resource with the audience, with developers who are interested. And basically the way that we’ve designed these types of reference kits, tutorials, and notebooks is for you to basically take it, run it, get started, see the result. And then use it as an inspiration or a foundation for you to then check out additional models; try it for your use case; deploy it on the edge devices, for example, that you prefer; and really take it on from there.

Christina Cardoza: Awesome. And, Nicole, I’d love to hear from Pathr.ai side how the value of working with partnerships, how you guys have been working with Intel—are you guys leveraging any of the technologies that Ria mentioned? And what it’s been like on your side developing these intelligent queue-management applications and implementing them with retailers with Intel as a partner?

Nicole O’Keefe: Yeah, absolutely. Intel has been such a valuable partner for Pathr as we’re scaling spatial intelligence in the retail world, and we leverage Intel’s CPU-based edge servers and OpenVINO, like Ria mentioned, for our computer vision. And we’re able to deploy queue insights at scale in a very cost effective and efficient way. So Intel has been there from the beginning. So, excited to continue as we scale spatial intelligence.

Christina Cardoza: Awesome. And what is the relationship like from Pathr.ai when you guys work with end users and retailers? We have Intel being there providing the technology and the hardware, being able to really help make these solutions and developers build these and deploy these and make these possible. So, where does Pathr.ai come in? When you’re working with a retailer how are you guys developing solutions and really implementing and deploying them across multiple stores?

Nicole O’Keefe: Yeah. I mean, we work closely with our retailers to really understand what are their challenges that they’re facing right now. And so, around the checkout experience retailers want to understand how often are registers being used? How long are queue lines? And how long are shoppers waiting? And so all of these factors really help them gain even more insight into, how are your store operations running? And really the goal is for them to run like a well-oiled machine, make the experience for customers as enjoyable as possible.

Christina Cardoza: Great. Now, we talked about using spatial intelligence to do some of this. Some of the use cases, Ria, you mentioned, I’m assuming you use object detection and other AI capabilities and algorithms. So I’m curious if you can talk a little bit more about what are the AI algorithms or the machine learning models that make this possible? And, like we talked about, how does this go beyond intelligent queue management? How do we start building on the success that we see with one implementation and really spreading it across the store?

Ria Cheruvu: Sure. I think there’s a number of different elements that we can do to build on these existing pipelines. We’re seeing the emergence of really popular and powerful object detection and classification models. But I would say that it even extends beyond that, Christina, and I’m sure Nicole can also share additional insights on it. But there’s additional models coming in, like pose recognition and activity recognition, that are helping us better understand how individuals are walking through a store, what they’re doing, that are helping add to the insights that we’re able to get with these models.

In addition, I think that there’s also a world that we really need to think about around these AI models in terms of the pre-processing and the post-processing that we do. For example, once we get those detections, what’s the type of information that we can extract—including attributes and specific types of characteristics that we can get, trends that we can form from these models as well?

And I think, really zooming out of that, the bigger picture is taking a look at how we can assemble all of these models as part of pipelines, or, again, interacting with each other for a lot of different use cases. Whether that’s just checking and validating each other’s outputs across a multiple-camera type of setup, or, again, as I mentioned earlier, being able to append the outputs from each of these pipelines as part of a dashboard for easier visualization.

But, overall, just to summarize what I’m saying, the emergence of new models, the kind of elements that come outside of it for post-processing, extracting insights, or even the things that we can do to help the models, like defining zones for intelligent queue management or potential parameters that we can preset—these kinds of elements are definitely helping put together the bigger picture for the use cases that we can start to see.

Christina Cardoza: One thing that I think is really interesting when we’re talking about these use cases and these improvements to their retail stores, it is really non-intrusive to the user. They don’t know, they don’t feel the impact—well, they feel the impact with getting the lines shorter, but they don’t have any experience cut out, because they’re now—retailers can see: Oh, I need another cashier open so that we can start spreading out some of the lines. It’s not going to be a hindrance on their shopping experience because—

I just think of sometimes retail implementations that I’ve seen to date, that little robot that’s going around the store following you everywhere that you can’t seem to escape that is just trying to find, like, spills or any hazards on the floor. But these are really adding AI that is making improvements that the customers will feel, but it’s not going to get in their way and it’s not going to intrude on their shopping experience.

Nicole, I’m wondering where you see this space headed: How can we continue the success of AI in the retail space? What’s still to come? Where do the opportunities still lie ahead? And how does Pathr.ai plan to be part of this future?

Nicole O’Keefe: Yeah. I mean, that’s such an exciting question. It’s one of the most exciting things about working in AI in the retail space, is figuring out where you’re headed. A few things come to mind here. So, as customers we’ll continue to want that seamless experience while shopping—short queue lines and wait times.

But for retailers it’s all going to be focused around store operations. How can we optimize our operations in-store? I mean, this could look like reducing their staffing costs, working with real-time alerts, and understanding a real-time scenario. When do the checkouts need to be open and closed? And then really making data-driven decisions in a way that, hey—if registers are not being utilized, well, maybe we can allocate that staff to other areas of the store.

Another interesting trend that we touched on earlier was self-checkouts. A lot of retailers today are implementing self-checkouts alongside the more traditional staff checkouts. And here at Pathr we’re also able to empower retailers with insights around staff checkouts or self-checkouts, and help them understand the performance between these two.

Christina Cardoza: Great. Well, it’s been a great conversation with you guys. I think we are running out of time, but before we go I just wanted to throw it back to each of you any final thoughts or key takeaways you want to leave our listeners with. Ria, I’ll start with you. What should developers know about AI in retail stores? Where is this going? How they can prepare? And where else Intel sees this space going?

Ria Cheruvu: Sure. I think it’s a very exciting space. I think you can see Nicole’s enthusiasm when she’s talking about the problem statement, and that’s definitely something that we share at Intel—the enablement, the strategy, and also the enthusiasm that developers can get when they’re solving problems that they have faced in real life and know that they can create innovations and solutions that can do better.

I think definitely when developers are turning to models and algorithms like YOLOv8 or these types of models for object detection and classification, definitely thinking about the bigger picture and better identifying how their solution is fitting in a real-world environment with all of the challenges and pain points that it can come with. Knowing that AI models are still sometimes prone to failures no matter how performant and powerful they are.

I think these are a lot of insights that I personally learned as part of my developer journey, and something that we’re working towards as a community. I’d say, in terms of the future and where Intel and our teams see spatial intelligence and the retail space going, we’re seeing a lot of these experiences being transformed by AI, and continuing to see that as well. And we’re taking steps, I think, to getting to a point where everyone is comfortable with the way that technology is integrated into environments, and we’re reaching a level of ease and convenience that we want to accomplish with new types of algorithms. We’re using existing types, optimizing them, and accelerating them.

Christina Cardoza: Well, I can’t wait to see what else Intel does in this space. I know that you guys are actively trying to improve your solutions and make things easier for developers, make things easier for partners. OpenVINO just had the 2023.0 release, where you guys really listened to developer feedback and pain points and added some more capabilities that make it easier to develop these solutions. So, can’t wait to see how that continues to expand in retail as time goes on.

Nicole, any final thoughts or key takeaways you want to leave us with today?

Nicole O’Keefe: Sure. I mean, this has been such a fun and engaging conversation with you both, so thanks for having me on the show. I’ll leave our listeners with this today. You know, consumer behavior is rapidly shifting, and those retailers who will just kind of wait and see how that unfolds are going to be left behind. And, really, the time to act through data is now. And one of the ways that they can stay ahead of the game is by using spatial intelligence with Pathr alongside with Intel. I think it’s a perfect combination.

Ria Cheruvu: And one final takeaway that I would add definitely for women in AI and developers who are female and who are driving areas in leadership is to definitely continue pushing forward. There’s definitely a lot of available technologies out there. And with the democratization of these reference kits, implementations that you can go ahead and plug and use, I think that that’s definitely a very big motivation for being able to get started in the field.

So that’s also something we could definitely want to see more of in the AI space. So I definitely recommend developers who are women, who are minorities, who want to really get out there and see their voices being heard, consider driving your innovations in the retail space and getting started there.

Christina Cardoza: Yeah, absolutely. Great point, especially on a podcast where we have three women in different leadership spaces representing their companies. I think that it’s great to see, and I would love to see more faces like us developing these solutions and being part of this AI and retail movement.

So, with that, I just want to thank you both again for the insightful and informative conversation. And thanks to our listeners for tuning in. Until next time, this has been the IoT Chat.

The preceding transcript is provided to ensure accessibility and is intended to accurately capture an informal conversation. The transcript may contain improper uses of trademarked terms and as such should not be used for any other purposes. For more information, please see the Intel^® trademark information.

This transcript was edited by Erin Noble, copy editor.