Data science, machine learning, and AI have clear applications for e-commerce, and given their relative ease of implementation, most online retailers are already deeply invested in strategies like recommendation engines, dynamic pricing optimization, and supply chain optimization. But so far, aside from the big players like Amazon, brick-and-mortar retail is behind in the move to AI.
The State of AI in Retail
Using data in physical retail is more challenging than online retail for obvious reasons; yet taking a step back, both are still surprisingly only in early stages. Capgemini found that “More than a quarter of the top 250 global retailers are integrating AI into their organizations. However, [they] also found that only 1% of AI initiatives reach full-scale deployment.” This is an exciting time for automation in retail; if smart systems can help leverage analytics at every location of a chain, the business insight will help executives understand local trends and better adapt marketing campaigns.
This seems to hold up, as Capgemini found that sales and marketing were the top use cases for AI in retail, even though most were just POCs. They anticipate that there is $340 billion in increased profits for businesses that can take advantage of AI use cases in operations, such as “AI stock replenishment” and “AI-bots for shelf scanning.”
Challenges for Brick-and-Mortar in the Race to AI
Calculating ROI for data initiatives is hard no matter what the industry or use case. But at least with more digital businesses, experimenting with data initiatives is less costly given that the data is already flowing in from multiple sources.
In brick-and-mortar retail, the largest hurdle to AI is, unfortunately, data collection. Sure, there’s transaction and loyalty card data that can be used for marketing initiatives (like churn prediction and prevention). But to implement machine learning — particularly deep learning — projects in the stores themselves, there’s a chunk of missing data related to the in-store experience. Yet obtaining this data via complex systems is a huge cost hurdle, and given that today, only 1% of retail AI initiatives reach full-scale deployment, implementation is still (understandably) considered a risk.
What’s to Come?
In general, DYI AI is becoming a thing (thanks, Google). But it’s the enterprise — and particularly retail — that has the opportunity in 2019 to move this technology from the realm of gadget to practicality. For example, accessible, cost-effective ways to make intelligent cameras that can recognize objects can benefit everyday companies who can use the technology to jumpstart their AI efforts. This may open the doors of possibility to new, real-life uses cases that would not be possible with a startup price tag of millions (or billions) of dollars to develop and use the technology otherwise.
There are also some up-and-coming technologies looking to provide baseline AI technology for brick-and-mortar retail, like Trax. They’re poised for big things, especially since Capgemini found in the same report that AI has only penetrated 10% of brick-and-mortar stores, as most adoption is limited to digital-native businesses, and Trax is targeted at physical stores, a wide open market with a lot of potential for innovation.
Trax helps stores keep track of stocking needs and performance through a series of automatic cameras and technicians. They take pictures of each display using multiple cameras to maintain a high level of quality. Images are sent to the cloud where different perspectives are stitched together to enable neural net computer vision technology to form 3-D “objects” at the item level. Performance metrics are automatically returned to retailers within a few hours.
This is useful because it equips store owners with the tools they need to anticipate customer needs and enables geo owners to target their strategy to the specific demands and opportunities of each store.
Deep Learning for Retail: How Does It Work?
To better understand deep learning for retail, we did a deep-dive into Trax’s ML patents and found that they solved tricky size issues for computer vision. Instead of transmitting high-resolution images of each shelf in order to determine the stock available, an ideal Trax system is equipped to take both high and low-resolution images.
The low-resolution images are transmitted to the server, and the high-resolution image only supplements when an item in the first image is indecipherable; they never send the whole high-resolution image and delete after a set amount of time. In this way, Trax optimizes both storage and network usage without sacrificing computer vision accuracy.
Trax has partnered with GfK to leverage their proprietary point of sale and catalog data. This partnership has resulted in a new syndicated service called In-store Intelligence, to tackle campaign compliance, product positioning, and in-store execution. While the Capgemini report suggests that sales and marketing are already popular and that the real skill gap is in machine learning-driven operations, since there is such poor saturation of ML in brick-and-mortar stores, this could still be a significant offering.
We've Got Data... Now What?
Of course, once systems are in place that are collecting data that could possibly be used for deep learning projects, the question becomes... now what? How does one go about actually using all of it? That's where data platforms come in.
To scale, data teams need far more than just good data — they also need staff, structure, efficiency, automation, and a deployment strategy; data science tools facilitate these requirements (and much more — learn more). They can also help by bridging the gaps in technical skills required through tools like visual machine learning or deep learning plugins that facilitate simpler image recognition.