Image recognition and retail: seeing is believing but it’s not enough

Share on email
Share on linkedin
Share on facebook
Share on twitter
Share on tumblr
Image Recognition in retail is now becoming more common but it’s just one piece of the puzzle in the quest for perfect retail execution

Even in an age of digital retailing, reconciling the digital record with physical reality can be a challenge. An extreme case is clothing retailer Ted Baker which in January asked accountants Deloitte to investigate its inventory and found it had £58 million (€68 million) of phantom stock. It was there on the balance sheet but couldn’t be found in the real world. So imagine then the challenge of knowing what’s on shelf in a supermarket at any point in time. Is every item in-stock, in the right place, presented in the right way and priced correctly? Out of stocks alone are a multi-billion dollar problem for the industry. Traditionally this has been a very labour intensive task both for the retailer and for vendor field sales teams who need to ensure their own products are well presented. No surprises then that the industry is investing and experimenting with technology to help and creating a new branch of retail analytics in the process. No surprises either that the Covid-19 pandemic is accelerating this trend as manual tasks are affected by the need to keep social distance and wear protective gear.

At the centre of these developments is Image Recognition; the ability of Artificial Intelligence and Machine Learning to identify which products are on shelf and even compare the shelf image to a planogram to identify what’s missing and what’s in the wrong place. To be effective, image recognition solutions need to be deployed as part of a wider system that includes how images are collected, how the system is trained and how analytics feed into workflow to enable actions to be taken. What follows is a selective overview of some of the technologies and emerging companies appearing in this space.

Retail product recognition - training the system

Training the system and keeping it up to date is critical for computer vision. Just like a human, the system won’t recognise a product it has never seen before. Therefore, a complete and up to date product library with images, descriptions and ID codes is essential. Across the whole store this is a huge exercise and with thousands of new products launched each year keeping it up to date is even more of a challenge. Enter vendors such as Nielsen’s Brandbank which systematically captures high quality images and detailed product descriptions. Much of the content is syndicated which helps to spread the costs across the industry and the images and data can also support e-commerce and supply chain operations.

Who holds the camera? From phones to drones

Early deployments of image recognition used in-store staff and field sales reps as data collectors. The logic was that these people were typically collecting data manually anyway, so Image Recognition was part of automating a process and taking advantage of the high quality cameras on mobile devices. However, time was still being used for data collection that could have been used to fix the problems. Also, many FMCG companies were concerned that their sales teams were not an independent source of data.

For FMCG companies, putting image recognition into the hands of a third party data collector was a potential solution but the costs of using full time market researchers was a problem. A solution emerged in the form of the “crowd:” Consumers can download an app that enables them to undertake store observation missions and earn a little extra money for it. Our own business is an example. This solves the problem of securing an independent source cost effectively.

Retailers have the advantage of being able to control the store environment which means they can deploy new kinds of technology. Instead of having your staff take pictures why not get a robot to do it? French start-up Qopius is experimenting with a robot that can wander through the store able to capture a shelf image throughout the day. The robot is potentially intrusive and you can imagine retailers having concerns about using it in a crowded store. There has also been the recent news that such roving robot trials have been discontinued by Walmart in the US. So how about having your robot fly? Pensa Systems has attracted significant funding to do just that. Autonomous drones hover above the shelves, detect when an aisle has no traffic and then descend to shelf level to collect a shelf image. 

But aren’t stores already full of cameras for security purposes? Couldn’t they be used to power image analytics in retail? That’s exactly where some retailers are heading. Take Walmart’s Intelligent Retail Lab or Store #8 as an example. It goes beyond cameras to include shelf sensors and integration with other systems. The Amazon Go stores are an obvious example and Albert Heijn is also experimenting. Emerging vendors like  SFD Systems and AWM Smartshelf are developing smart shelf technology solutions that can enable dynamic pricing and promotions, shelf-edge advertising and checkout free stores as well as providing feedback for retail analytics. If this sounds experimental keep in mind that more than $200billion per year is invested in retail technology and the direction of travel is clear: stores are going to be smarter. For manufacturers that also means that in the long run retailers are going to be in control of the data for image analytics in retail.

Making a difference with data

The out of stock won’t be closed and the planogram won’t be corrected if there is no bridge from the analytics to action. The problems and remedial actions need to be included in the workflow of the people who can do something about it. For retailers Yoobic is a great example of this. An app on the in-store employee’s smartphone indicates what actions need to be taken, what “good” looks like and allows them to notify head office when the issue is fixed. The app can also integrate training content so it seamlessly fits into the employee’s working day. At BeMyEye we do the same for manufacturer field teams. Our Field Compass app shows the rep which are the stores on their route that most need attention and dynamically optimises the rep’s route to maximise the impact they can make during their working day.  It also highlights and prioritises the actions to take when they get to each store. Across the team data is fed to dashboards in a Retail Execution Cockpit, enabling sales managers to allocate their resources to maximum effect. Using an API, data can also be integrated into Customer Relationship Management (CRM) and Sales Force Automation Systems (SFA) systems. Integration with systems like Business Intelligence, Enterprise Resource Planning and data lakes is also possible.

Image Recognition and the associated technologies are part of a revolution in improving retail execution. To make a real difference and deploy it effectively you need to think about a series of critical steps; product data and training the system, who or what will collect the data, what analytics will you use and how will you change your people’s work pattern and integrate the data into their workflow. The trend is clearly towards more automation, so if your in-store employees or sales team are still spending time collecting data, it’s probably time to think about making the next step.

If you would like to learn more about data sources and technologies for retail execution, you can read the full outlook in our white paper “Perfect Sales Execution: The ultimate guide to data sources and technologies” here.

Neil Preddy

Neil Preddy

Neil has more than 30 years experience helping FMCG companies. He is a Six Sigma Executive Blackbelt and an advisor to BeMyEye.

Don’t forget to share!

Share on email
Share on linkedin
Share on facebook
Share on twitter
Share on tumblr

Want to reach perfect store execution?

See how BeMyEye helps brands track their retail execution in real time so they can take corrective actions:

cookie

This website uses cookies to ensure you get the best experience on it.