Smarter retail requires a better way to scale AI
Imagine shopping in a future grocery store. Scan yourself when entering with your palm. As you move around the store, put your items in a smart shopping cart that lists your bill and you can leave the store without spending a second at a checkout.
Sounds like a utopian dream? For most shoppers, this is still the case, even though Amazon, Walmart, and other retail giants have started making huge investments in the technology required to run such smart stores. What does it take to make it a reality?
Retailers need smarter AI, and the artificial intelligence needed to address these challenges needs to be more accessible for retailers large and small. Otherwise, cashless stores could never really go mainstream.
Here’s where smart retail will stand in 2021 and how retailers can take smart stores to the next level.
Current use cases
Amazon Go is currently the king of smart retail with its “just walk out” technology that relies on in-store cameras to track shoppers’ items and bill their Amazon accounts accordingly. But with fewer than 30 active Amazon Go stores in the US, the experience they offer remains rare, and given the expensive investments required to run such stores, scalability remains a big question.
This is due to the high price associated with it. In order for these intelligent cameras to work in real time, they need a lot of computing power and the algorithms have to be very precise. To save costs, enable scalability, and operate efficiently, retailers often have to choose between performance – how quickly the camera can detect goods – and accuracy, which is not an ideal compromise when most retailers are already operating on very low margins.
It’s not just retailers who need to consider smarter cameras. Some retailers, like Choice Market, a Denver-based chain, allow shoppers to check-in and check-out using a mobile app, and even allow customers to customize their shopping experience based on dietary needs or specific recipes.
[Related: Retaining the New Consumer Through the Power of Technology]
Other stores do not fully automate the shopping experience, but use AI-based heatmaps to better understand customer behavior and to better target promotions, product placements and customer communication in so-called hot zones. All of these technologies require hardware to process huge amounts of data efficiently, where latency is a no-go; otherwise, buyers can shop elsewhere.
Given this complexity, the financial barrier to entry has been increased. It can only be reduced by finding better ways to scale the AI algorithms responsible for these tasks to run on cheaper hardware without compromising performance or accuracy.
The technology to make it happen
Currently, many businesses rely on cloud infrastructure to process data – and while this approach makes it easier to scale, this approach can create high cloud costs, delays in processing times when transferring data to the cloud, and potential security-related privacy issues bringing with them data in transit.
In-store servers that process data at the edge address these concerns, but retailers must purchase the necessary hardware themselves and upgrade the associated infrastructure to incorporate more sophisticated, complex deep learning models and other functionality.
While it is all too easy to get bogged down in debates about whether cloud or edge processing is a better solution, those discussions miss the point.
Instead, retailers should focus on leveraging their existing hardware rather than investing huge sums of money into new infrastructures if they want to innovate. Optimizing software can help save valuable time and money while delivering the same benefits retailers seek when they make large hardware investments. And this is where really efficient AI algorithms could make a crucial difference.
Skilled data scientists will be the linchpin of this smarter approach as they bring in a deep understanding of the nuances it takes to solve algorithmic problems and fine-tune retailers’ AI systems.