How Edge AI is Reshaping Retail - and Creating an Innovator's Dilemma

Good morning. Clayton Christensen's book Innovators Dilemma introduces the theory of the “Dilemma Zone”—the crucial moment when new technology emerges and growth in older technology starts to slow. It’s the perfect framework to understand what happens when fresh innovation shakes things up.

I mean, look at Walmart back in the day—they watched Amazon take over like, “Nah, they sell books. We got this.” Or Intel looking at NVIDIA saying “we don’t need fancy graphic chips”. Now NVIDIA is practically printing money.

You can almost hear these companies collectively ask, “Wait… what just happened?”

They are stuck in the Dilemma Zone.

At Tech Tales and Tactics, we love to look at the new—the red curve (in the chart below). We seek new opportunities for businesses and individuals empowered by new technology.

In today’s episode, we examine the retail industry and its dilemma caused by a technological advancement: edge AI. This advancement requires an established industry to take risks and invest capital in a technology that might not show immediate ROIs or other benefits.

Hint: Amazon is already killing it in this area, again taking on retail…

Let’s dive in.

6-minute-read

What is Edge AI, and Why is it Important?

Cloud technology has disrupted on-premise servers and fast access to computing resources. It has enabled entrepreneurs in the software space to start businesses without too much commitment and has provided a quick way to scale if ideas work out. It has also paved the way for better data analytics by combining data from all kinds of different data-sources. Cloud technology has changed the business world in many ways.

With edge computing, we somewhat reverse the move to the cloud. We bring information storage and computing closer to the devices that users interact with. Complex operations are done directly in our phones, cars, and cameras. Unthinkable until a few years ago, we can now even bring highlight complex AI processing back down into the devices - that’s what we call edge AI.

Why is this happening? Can We Not Make Up Our Minds?

One reason to move things to the cloud was that cloud computing was much more powerful. Big data operations could not be done on smartphones until a few years ago. All the “crunching” was done on the much more powerful computers that were way too big to put in our pockets. But things changed.

Rapid hardware development is enabling a new path. Edge AI can only be practical with advanced processors that are small and efficient enough to handle complex AI operations. A new generation of chips can now take complex operations back into devices. A few hurdles like energy efficiency, size, and heat have been overcome so that these new chips are small and powerful enough to take on AI work in our pockets - or any small box - for example, connected to a camera.

Edge of Glory - Who Makes These Chips?

The competition for these chips is fierce. It’s no surprise that NVIDIA is well positioned here. NVIDIA’s Jetson series already powers robotics, autonomous technologies, and other IoT applications. If you are itching to start building Edge AI applications - you can order one on Amazon for $499 right here and make all your Edge AI dreams come true.

Apple also has some high-performing chips in this race. The Apple Neural Engine (ANE) powers Apple’s A-series and M-series chips.

Lastly, the most popular smartphone chip is the Qualcomm Snapdragon. The Snapdragon, for example, powers voice recognition, object tracking, and photography enhancements. If you have an Android phone, that’s likely the chip powering all the good stuff.

Why consider edge AI?

More speed—fast decisions: Most of the services we use are hosted on a server that our phone connects to. This server can be thousands of miles away. This introduces latency, the time it takes for pages to load the content or for ChatGPT to answer our questions.

Imagine a self-driving car would have to avoid a collision only by asking a remote server how it should react. These are real-time decisions with no time for latency. For this, self-driving cars need to be equipped with the right computing power to handle these decisions within the car.

Data security—local data: AI can make great decisions for us, but only if it has a lot of data available. Uploading data to the cloud creates security and data compliance challenges. If decisions are made on the device, there is no need to upload and store data on remote servers. For a decision to be made, it can stay local, eliminating many risks.

Personalized experiences: Again, large amounts of data are needed to run recommendation engines. If I am, for example relying on a shopping companion that guides me through a store, I don’t want to wait or the service to stop when I have bad connectivity in a shopping mall. If these algorithms run on a device, the interactions can be real-time and happen without interruption.

Some industries can greatly benefit from bringing fast, secure, and personalizable decision-making closer to the user and supporting real-time interaction. However, getting it right requires massive investments, risk-taking, and lots of research and development. Thus, Clayton Christensen must be all over it because it is a prime example of an Innovator’s Dilemma.

Let's take a look at the retail industry as a case study of how edge AI can change the game for retailers if they are willing to take the chance.

The Pioneer: Amazon’s “Just Walk Out” Stores

Amazon has built an impressive tech stack for these new “Just Walk Out” stores. As customers walk through the store and pick up their products, their purchase receipt is updated as they fill their carts.

Enabled by sensors across the store and local AI models combining and analyzing the data. The models can take multiple inputs (multi-modal). Shelves have weight sensors, and cameras are on every corner to track customers across the store. Computer vision analyzes their interactions, and a central model combines these inputs. All of it has to be done in real-time.

Amazon has deployed this in a few test stores, such as the micro-stores in the Seahawks stadium. This has got to be a wonderful testing ground for their engineers. Imagine getting a computer vision model to track 20 people wearing the same Seahawks outfits accurately.

Aside from that, relying on a cloud to do the computing would be too slow and risk latency issues, making tracking unreliable. Edge AI is the enabling technology that minimizes cloud dependency.

Check out the AWS blog for a deep dive into the Machine Learning tech deployed.

The Competitors: Key Players in Edge AI for Retail

Standard.ai

Standard AI unlocks real-time AI powers in retail. The San Francisco-based startup leverages computer vision and turns it into real-time analytics. Standard AI has raised $266 million and is currently valued at $1.5 billion.

It seems to take a similar approach to Amazon. However, it highlights the benefits of real-time analytics. For example, its VISION model can provide predictive insights about the impact of placing specific goods in different places in real-time.

Zippin

Another notable mention in the retail segment is “checkout-free” competitor Zippin (as in Zip in. Zip out). Zippin focuses on small-scale stores and provides a cashier-less experience. Zipping is well established in the sports industry and equips stadium shops across the US, from Brooklyn (Nets) to San Francisco (Giants). With reportedly $44 million in funding, they are already generating a revenue of $23 million annually.

AiFi

Competitor AiFi, which provides autonomous store checkout systems, already partners with ALDI, a German grocery chain. The cool differentiator I found here was that AiFi also helps stores optimize their layout based on its advanced analytics. They analyze traffic across the store to help model it appropriately.

They also have some modular stores that can be dropped anywhere and run fully automated. Current valuation ~$390 million

Something dawned on me while researching these solutions and where they are already actively deployed. Interestingly, they all got a foot in the door with sports teams and the little shops within stadiums. Improving check-outs must be the driving pain point.

During halftime, the stores must funnel as many customers as possible through the store. They want to buy fast and return to their seat as the game or show continues. If its too slow, they might not buy at all. Speed absolutely matters, and this is where the ROI stands out.

However, what if the ROI is not so obvious for other retailers (yet) before its too late)?

The Innovator’s Dilemma

Let’s apply our book smarts to what is happening here.

Looking at the curve below, we can put traditional retailers on the blue curve. Existing technologies sustain the market, and not much has changed in our retail customer experiences over the last decade.

After reading about what Amazon and the other startups are doing in this field, I would plot them on the red curve above. They are disruptive innovations and are currently in the exploration phase. Some stores are fully operational, but only in select locations.

The dilemma arises. The time has come for retailers to realize that this is happening. It’s cool technology, but there are still many risks that they have to balance.

What causes risk for them?

• Significant investment - Ramping this technology requires a lot of investment in hardware, software, and talent.

• Completely changes workflows - Change management across the company to establish new workflows.

• Cultural shift - How will existing employees react to this? Autonomous stores put jobs at risk. Can they embrace this new technology?

• Complexity and errors - I can only imagine how many edge cases evolve. In other words, in which situation would this technology fail because it needs too much data and training to understand what is happening before it works with near 100% accuracy?

• Clear ROI - Any investment a company makes must have a clear ROI to its shareholders. Retail margins are thin. Do they see enough ROI yet?

How can they overcome these risks and decide if the bet is right for them?

These are the practical steps the book suggests:

One pattern we noticed when we looked at the market was that the edge AI startups found a footing within sports venues. The biggest pain here is the check-out time. It must turn customers away when waiting 20 minutes to get a $20 bag of cotton candy. Unless their kids are as patient as mine, unfortunately, it’s totally okay to wait and miss the second half of the game. But they lose a lot of customers because the check-out is the bottleneck.

Once the automated solution convinces, the owners will probably look at what other improvements this fully automated store can bring, such as better personalization, optimizing store layout, real-time coupons, and buyer recommendations.

These are nice to have and might not show a clear ROI at the beginning.

A personal wishlist item would be a guided shopping cart across the store to purchase items I need for a specific recipe—maybe even a recipe that fits my personal taste. Or general in-store navigation based on my shopping list. Connect the shopping cart to my personal shopper profile and guide me around. Maybe a startup opportunity?

Conclusion: What’s the Right Path for Retailers and Edge AI providers now?

Edge AI provides opportunities to create better shopping experiences. From the simple examples, we talked about to guided AI shopping companions, magical AI mirrors can be style guides, and AI-guided store optimization based on very specific store and customer traffic patterns. There is a lot that can be done.

However, not for all of this seems to be a big enough pain point. Retailers can work closely with startups to bring up their biggest pain points and let the startups figure out how edge AI can solve them.

The ROI seems clear to support fast check-out, which is a huge pain point - stadium venues need to get as many people through in half-time as possible. But that’s just the beginning.

Retailers that don’t have these huge pain points face the dilemma of deciding whether to invest in this technology or risk becoming obsolete because they can’t create competitive shopping experiences in the long run.

Retailers can offer real value and position themselves as forward-thinking leaders, but startups in this space must work closely with them to identify the pain that can be solved.

This analysis is a bit different. We looked at new technology and then applied the Innovator’s Dilemma framework to understand what is happening in a specific industry. How did you like it?

Have a great rest of the week,