IPLAND helps top brands score in-store sales with Microsoft Azure AutoML

Using IR to innovate and modernize retail environments

Every retail store aisle is a contest among brands to see which display can successfully coax shoppers to add a product to their cart. Fast-moving consumer goods companies like Nestlé, Coca Cola, and PepsiCo look for every advantage they can get when it comes to amping up their products’ shelf appeal and they turn to IPLAND for help. Ukraine-based IPLAND uses Image Recognition (IR) technology to help brands analyze and optimize their in-store presence to best appeal to browsing customers. 

“IPLAND is a solution for brand and sales management on store shelves, where the first decisive moment takes place–the customer makes a choice, and the product becomes a purchase,” says Oksana Koba, Deputy Director at IPLAND. “We increase the efficiency of the supply chain, from manufacturers and distributors up to the stores’ shelves in fast-moving consumer goods and pharmaceutical industries.” 

IR technology provides an innovative way for product brands and retailers to monitor in-store inventory and optimize sales performance in a complex environment. To do that, IR detects, recognizes, and reports product orientation and facing on a shelf, product availability, prices, competing products, share-of-shelf, and more. Here’s how it works:

1. A field worker or merchandiser takes photos of a stocked store shelf using their smartphone or tablet. 
2. The photos are uploaded from the worker’s device to the cloud.
3. IPLAND’s cloud-based IR service then analyzes the photo for brand compliance and key metrics.
4. The field worker or merchandiser receives results to review on their device.
5. Further analytics are available to retailers and client brands on the web portal.
   
   

IR helps brands provide on-demand display and compliance feedback to retailers and merchandisers so they can make real-time corrections. It also tracks data on key performance indicators to help brands gain insights on retail customer choices and trends. IPLAND provides clients with a full-cycle service which includes cloud access, monthly service updates, staff training, data administration, technical support, and advice from business analysts for further development. This is a game-changing service for brands who want to better understand how their products and in-store marketing methods perform in the field.

Succeeding at scale

 As IPLAND grew its list of clients—which now includes over a hundred major brands—the company’s original IR service infrastructure struggled to keep up with demand. Recognition speed is crucial in fast-paced retail environments, where any delay can lead to lower productivity and missed sales goals. If the feedback takes longer, the retailer or merchandiser takes longer to make necessary changes, which can limit how many displays they’re able to adjust in a day. 

“Initially, we worked with a service configured on virtual machines in Microsoft Azure using the TensorFlow framework and the Mask R-CNN libraries. That did not allow us to scale the recognition service in the required volumes. As the number of clients using the Image Recognition module grew and photo data arrays increased, we encountered unacceptably long latencies in obtaining the recognition results,” says Koba. “Moreover, the possibilities of its administration were limited and had inconvenient interfaces. In addition, the recognition rate was not what we desired.” 

Optimizing the scale and speed of its services became a top priority for IPLAND, so it set out to find a machine learning platform that could scale its cloud operations effectively and parse the increasingly large data sets rapidly.

After comparing competing options for a powerful IR service with a low-friction end-user experience, IPLAND elected to move forward with Azure automated machine learning (AutoML). The new solution with Azure AutoML includes cost-saving efficiencies—it requires less maintenance and provides better management over testing and training models. With AutoML, the new IR service will continue to improve after deployment, helping clients more accurately catalog how products function in a live retail environment, which informs their future packaging and distribution decisions. 

Quality work, done quickly

To better support field workers with real-time adjustments, IPLAND wanted to increase the image recognition speed to 15 seconds or less—a time threshold that can get results to workers promptly while remaining scalable as the workload on the IR service continues to increase in the future. 

“The Microsoft product team helped us to conduct research on the types of Azure virtual machines that could support the highest cluster performance and the least photo processing time, for both training and inference,” says Koba. After discussion with the Microsoft team of architects, IPLAND decided to implement Azure AutoML for Images, which gives the company full flexibility in choosing the model architecture and hyperparameters. IPLAND also has complete access to the model after training.

Improved photo recognition sets a fresh standard

Testing on the new IR service has shown that recognition performance results exceeded 95 percent accuracy. “In the field, IR now has an average recognition speed of eight seconds per image, and the processing time of any photo does not exceed 10 seconds—beating the 15-second goal by 33 percent,” says Koba. This has made it possible to scale and perform recognition in as many streams as necessary. 

“This was immediately appreciated by our clients,” says Koba. “We have already received several requests for IR service scaling.” 

IPLAND is still in the process of fully rolling out the Azure AutoML solution, but these early results of increased performance and positive reception from clients are encouraging. Field workers can get better results at a rapid pace using IPLAND’s service, giving the retailers and brands a grounded view into consumer habits.

IPLAND estimates that the total cost of running on the new solution is lower than it was prior to the new solution. The service administration has been simplified as well, allowing for a more streamlined approach to ensure quality in its systems. By freeing up time and resources while delivering superior results, IPLAND allows clients to redirect energy away from gathering and analyzing data and instead focus on putting insights to use, making business decisions, and finding new ways to delight shoppers.

Find out more about IPLAND.