The Business Value of Data

The Business Value of Data-Driven Cultures

August 23, 2023

Christina Cardoza

Data-Driven Cultures

It should come as no surprise to anyone in the IoT landscape that data is key to business success. Computer vision has opened a whole new view into business operations, which has led to a need to collect, manage, and analyze all that data, which in turn opens the door to AI that offers insights and can lead to valuable changes. Now the elements are in place for what we might call the data-driven culture of not just one industry but many, from manufacturing to smart cities to dining.

But what exactly does that mean, “data-driven culture”? And does it just create more complexity and additional challenges for businesses and organizations trying to implement it? We talk to two people who know a lot about getting the most out of data-driven culture: Atif Kureishy, Founder of AI and automation retail solution provider Vistry, and Saransh Karira, Head of Engineering at Awiros, a video AI OS and marketplace company (Video 1). They discuss its benefits and challenges, and how a data-driven culture can connect different aspects of a business to create real value.

Video 1. Awiros and Vistry explain the value of creating data-driven cultures and the strategies to successfully implement them. (Source: insight.tech)

When we talk about data-driven cultures, what does that really mean?

Atif Kureishy: Data-driven culture is really about making decisions that are evidence based—decisions that are grounded in the understanding of data coming from your enterprise, and being able to trust that data, analyze it, and derive key understanding from it. Then, ultimately, making decisions that drive strategic advancements and strategic initiatives.

The first generation of data-driven culture was really about data acquisition and data understanding. The second phase of that journey, going on for the last decade or so, was then starting to do prediction on top of that, which introduces a lot of concepts in the machine learning space. And now I think we’re on the third generation with the introduction of large language models, LLMs.

And rather than having very human data science, or data-engineering-intensive activities, now we’re moving towards AI-based systems that tend to be smarter than us. And so how do we share a large corpus of enterprise data with those LLMs in a trustworthy way to make decisions that are informed in the enterprise.

Saransh Karira: At an earlier point, data policies were like an umbrella term for any kind of data. But in the last three to four years we have seen tremendous changes in the landscape, and now people are becoming aware that the amount of data that you give to the system is the amount of precision that you get from the system.

How are computer-vision AI applications making data more valuable?

Saransh Karira: It’s those changes in data policies: They make the data a lot more accessible. The raw data is the first step, then once you have this raw data, applying intelligence on it. But now let’s say you have thousands of hours of data—even when you have access to that data, it’s not really accessible; you cannot sift through it. So that’s where the systems come in—the intelligence systems, the machine learning systems. It’s all changing very rapidly.

And because of that, a lot of infrastructure is being built for integrating a lot of data. I think the value of data is when you can connect a lot of different types of data. So, if you take each data as a dot and then you can connect them together, the sum is more than the parts. A lot of our customers are connecting data throughout their different infrastructures or their different divisions.

One use case—but it extends to a lot of different organizations—is that we work with government extensively, and what we are seeing currently, for instance, is that they are connecting vehicle preregistration with cameras and then with passports. The interconnected data becomes much more valuable than one system that is standing just in a silo.

What kinds of challenges face the businesses you work with?

Atif Kureishy: At Vistry we are focused on the restaurant-hospitality space. It is a very people-oriented business, high velocity and relatively unsophisticated. Those businesses are starting to make a lot more technology investments, but historically that’s not been the case. So any type of capability that gets deployed and scaled across a large number of locations has to be very cost-effective.

And a lot of the things that we are tracking are objects in the kitchen, which makes for a unique environment. For sure, our training infrastructure has to be robust to be able to detect and track and understand the activities that are occurring in that environment.

“#Data-driven culture is really about making decisions that are evidence based” — Atif Kureishy, @vistryai via @insightdottech

This is where I think Intel especially has brought a unique value proposition, in the sense that you can run on commodity compute that’s right there in the restaurant. Or potentially deploy next-generation compute and have machine and deep learning models that can run effectively there at the edge. Some of the technologies around OpenVINO^™ and deep learning tools that the Intel group has provided have helped tremendously. So we can run our inference workloads on Intel Atom^® tablets, on i7 Tiger Lakes, on the new Alder Lakes very easily, and can optimize runtimes effectively. That’s been incredibly useful for us and for our customers.

How are you creating data-driven cultures and strategies for those businesses?

Atif Kureishy: Let’s take the example of production control—and a restaurant essentially is a mini manufacturing site. In a manufacturing sense, you have measurement of inventory, and you have QA and oversight of work products. And so if you apply that into a restaurant space, imagine that you have orders coming in through digital, through the drive-through, through dine-in. And when those orders get acquired, they get consolidated into a kitchen that needs to essentially manufacture the orders correctly, if you will.

Now one of the areas where AI and ML are coming into play is that you can create a production schedule in a quick-serve or fast-food restaurant where certain products are premade and then held. That is the ideal scenario because it allows food to get out as quickly as possible. So you build and manufacture those menu items most efficiently by predicting how many and what type of inbound orders you’re going to get. This also allows the kitchen to be much more efficient not only from a labor perspective but also from a food-waste perspective.

Another aspect where we’ve been using computer vision is with inventory management—having cameras that can look at a bowl or a pan and do volumetric estimation of how much product is in those pans to help inform that production schedule. And that, from a lean-manufacturing perspective, is sort of like the just-in-time concept. So, modeling demand and then using AI to ensure that the supply is there. That is how the optimization of the restaurant is becoming more data-driven.

If we think about what the culture of a restaurant was 20 years ago, it was really reliant on people—managers using their intuition: “I expect a lunchtime rush today. There’s going to be a field trip coming in on top of the usuals, and here’s how I’m going to place people.” And, by the way, there are a ton of restaurants, especially small restaurants and local restaurants, that still run that way. But when you look at the larger brands, they’re absolutely moving to more of this data-driven culture.

I wanted to highlight what the historical culture is in the restaurant, because I think it’s important to understand that, and then to understand how it makes sense that we’re now using data to serve the customer more effectively.

Saransh, what are the use cases you’ve encountered at Awiros?

Saransh Karira: One use case was a deployment with multiple different campuses, and for each campus there were multiple different access points. The original implementation was just to see how many people were coming in and how many of them were visitors—basically, how many of them had access to the site and how many of them were there for the first time. That was the initial use case.

But the customer then used that information to change the configuration of their security personnel depending on where people were—where there was more of a crowd, they added security there and reduced it from the other access points. So that was very interesting to see.

We have also seen a lot of what we can call meta-analytics use cases, especially in retail. For example, our customers can improve store layout and operations by being able to see patterns in foot traffic. Where meta-analytics comes in is to basically generate a heat map to visualize where the footfall is more and where it is less, and depending on that data our customer can change the configuration and placement of products.

What is the value of working with partners like Intel to promote data-driven cultures?

Atif Kureishy: We are very thankful for our partnership with Intel. It takes a village, or it takes a broad ecosystem, to make this all work. I would say it’s around ODMs and OEMs that are providing the Intel base compute, and also working with the systems integration teams that ultimately need to place edge devices and sensors at the locations so that this processing can occur.

And, of course, having a cloud-based infrastructure, we work very closely with AWS. And so Intel is a key part of facilitating the dialogues and interactions with that larger community.

And then, of course, there’s the robust set of tooling and infrastructure that’s provided around OpenVINO. That’s been great for us. It allows us to optimize the types of processing that we’re running on CPU or on the iGPU—integrated GPU. There’s also good support in working with the open-source community and the various deep learning frameworks that are out there. That has been wonderful.

Saransh Karira: With our platform at Awiros, we are trying to create an ecosystem of video-intelligence applications. Basically, it starts with the hardware, it goes to use cases, and then it goes to the marketplace. The hardware is where Intel comes in. And then on top of that there are different use cases that are being developed by different researchers or any of the third-party developers. And on top of that there is a layer of marketplace, which is what is visible to the end customers.

I think at the edge Intel is very cost-effective for us, first of all. And its libraries have helped us a lot in optimizations, be it for inferencing—the actual part where the AI runs—as well as the decoding part of the video, and many other things. Also, the support is very, very wide.

Final thoughts? What does the future of data-driven culture look like for businesses?

Atif Kureishy: We, like everyone else, have gotten on the GenAI bandwagon, if you will, and have really worked extensively with models like GPT-4 for the last several months. A lot of our focus for the first couple of years was generating, let’s call it dark data. How do we apply computer-vision workloads at the edge to create a data stream of physical observations?

And then that data needs to be stitched into a larger baseline or foundation of data that’s coming from the point of sale, coming from inventory-management systems, coming from time-reporting systems. And so we’ve been looking at LLMs to really interact with a larger and broader set of data and make sense of it. The ability to do that very quickly is really fascinating and phenomenal.

So if I were to leave this audience with something, it is that beyond ChatGPT and getting recipes and looking for travel itineraries and generating poems—which I’ve done with my kids, and we have a lot of fun doing that—this new wave of AI really does have big implications for the enterprise, and we’re excited to be a part of that journey.