Automating real world retail with virtual products
Out with the Old, Innovate with the SKU
The year is 2021. With a few, simple clicks of a computer screen you can order any product you desire and, in most cases, have it in your possession in less than a day’s time. That’s the best case scenario that retail customers are met with today but, more often than not, their experience is far from perfect.
Despite the marvel of technological and logistical advancement that has occurred in the past two decades, one only has to wander down to their local supermarket to witness the abundance of literal gaps that exist in the system in the form of Out-Of-Stock and Out-Of-Shelf products. These gaps should haunt the nightmares of retailers everywhere as each one represents a significant and preventable loss in revenue.
In fact, from the moment a product or Stock Keeping Unit (SKU) is manufactured to the moment it meets its end (ideally in a recycling centre) there exists bountiful revenue-generating potential. This potential is unnecessarily robbed by broken processes and imperfect systems at every stage of a SKU’s lifecycle.
This problem is no more prevalent than in in-store retail. Today, 5% of all sales are lost due to Out-Of-Stocks, with, on average, 8% of SKUs remaining Out-Of-Stock. The biggest headache here is that the issue simply does not have to exist with the right use of technology.
The best solution starts with the complete digitisation of your SKUs. This is the first step towards automated, real-time shelf monitoring that puts an end to Out-Of-Stocks using a powerful new technology called Synthetic Computer Vision.
Beyond 20/20 Vision
Switching from manual, time-consuming, and costly monitoring of shelf stock by in-store staff to a fully automated, real-time solution is a no-brainer and yet the majority of today’s commercially available technology isn’t up to the job.
Most computer vision (CV) solutions on the market rely on a process that requires massive amounts of costly data as input. This makes it impossible to adapt and scale to the demands of the average retailer. Simply put, traditional computer vision, the most widely used type of commercially available computer vision, is dying a slow but inevitable death and wasting plenty of retailers’ time and money on its way to the grave.
For a field of study which has been the hot topic of attention of fervent AI researchers since the 1960s, mainstream advances in computer vision have not been as drastic as the lofty expectations that were promised. Attempts to democratise the technology for widespread commercial use have been throttled by failure time and time again to optimise its largest dependency, the sourcing and preparation of high-quality data.
Data is Key to Success
Currently, it is estimated that only 1% of AI research is focused on sourcing and preparing data for AI models with the other 99% focused on AI model training and algorithm optimisation. This is in spite of the fact that the data input component of computer vision takes up largely 80% of a developer’s time, when implementing a solution that relies on an AI model, while 20% of their time is spent on the model itself. This disconnect between where a developer spends their time versus where advances are being made, in terms of efficiency, presents a very big problem for the future of this type of computer vision. On the flip side, it presents a very big opportunity for those who are willing to innovate with a more sophisticated and capable solution.
Rather than having a human painstakingly collect real data to train a computer vision model, Neurolabs is using Virtual Reality engines to generate the data - this makes the process faster, more cost-effective, and truly scalable.
For retailers, we use digital 3D scans of their SKUs to train our computer vision models. By doing so, we can shortcut the development cycle to a fraction of the time it would take using conventional computer vision methods. This approach also has the benefit of being truly scaleable and so easily adaptable that our real-time shelf monitoring solution recognises new products before they even hit the shelves.
From Lab to Production
Neurolabs is one of the very first companies in the world that is using fully synthetically trained Computer Vision models in large-scale production environments.
We’ve detailed one implementation for you below: from the rollout process and technical considerations to the proven benefits after a successful implementation. What started as a proof of concept to validate whether Synthetic Computer Vision was a retailer’s best choice when it came to on-shelf availability quickly turned into a large-scale rollout for the supermarket chain UVESCO.
Synthetic Computer Vision in Action
UVESCO needed a solution that could augment their staff’s efforts to maintain optimum on-shelf stock levels at all times. Humans are excellent at complex tasks that require reasoning but when it comes to monotonous, repetitive tasks such as shelf monitoring, human error and apathy, work to a supermarket chain’s detriment over time. This leads to persistent availability problems, inventory inefficiencies and, ultimately, lost revenue.
While inventory omniscience is an impossible expectation for human staff, camera-based, real-time shelf monitoring is the next best thing, providing complete visibility of on-shelf stock levels at all times. With a live view of on-shelf availability, the store staff are empowered in their efforts to maintain optimum stock levels whilst providing a frictionless customer experience.
The problem for most computer vision solutions is dealing with the fact that supermarkets are highly dynamic places. From constant throughput to frequent packaging updates, the changing nature of a grocery store and its contents presents a complex and challenging environment for traditional computer vision to operate effectively.
In order to consistently perform in a perpetually dynamic environment, the Computer Vision model needs access to gargantuan levels of data at speed. Synthetic Computer Vision thrives in such a setting. It does so by using a digital 3D recreation of every single SKU on a supermarket store’s shelves. This is done by combining the digital assets of a product with a cutting-edge 3D rendering process to create a SKU’s digital twin.
With digital 3D versions of each SKU available, Synthetic Computer Vision models can process infinite possible variations of the real-world product in infinite possible variations of the real-world supermarket setting. A flexible yet robust solution like this was exactly what UVESCO was looking for when they set out to solve their On-Shelf-Availability problem.
The goals of our original proof of concept project with UVESCO were twofold:
To prevent out-of-stocks,
To measure the impact of consistent real-time shelf monitoring using Synthetic Computer Vision.
The resulting data also helped UVESCO to identify which products sold out most frequently and at what times, enabling the retailer to prepare for consumer demand changes. This allowed UVESCO to guarantee that front-of-store inventory levels were optimal at all times and, most importantly, increase overall sales for the store.
Using a similar approach, Neurolabs helped UVESCO monitor distributor performance to determine and address the root cause of stock shortages.
The project was providing insights within weeks. The first thing Neurolabs did was focus on one specific product category in the store so that we could rapidly prove the benefits of real-time shelf monitoring whilst also reducing the risk of initial investment for UVESCO. By providing results quickly, UVESCO could green-light the project for full-scale rollout without any unnecessary delays. They could also capitalise on the benefits immediately.
With no need for image gathering, data labelling, or coding, Neurolabs can rollout object recognition technology at a rapid pace. The CV monitoring system takes less than a week to set up, and because the AI software uses synthetic data to train the model, retailers can start using the technology immediately after installation.
The ability to operate at a distance was critical at the height of the COVID-19 pandemic. Without the need to collect significant amounts of real data in person, the project could get up-to-speed with limited store visits from Neurolabs’ employees. The timing of the project gave UVESCO an insight into on-shelf product availability when they needed it most. With real-time shelf monitoring in place, the supermarket was much better equipped to respond to the challenge of fluctuating customer demand.
Over four months, Neurolabs ZIA helped UVESCO gain invaluable insights into consumer shopping and distributor shelf fulfilment behaviour. Some of the insights gained were:
Average shelf capacity
Percentage of the time shelves are full
Percentage of the time distributors deliver on time
Estimates of missed sales opportunities
Top products with availability inefficiencies
For Chris Burleigh, Head of Digital Transformation and Innovation at GRUPO UVESCO, the collaboration was a success:
“Thanks to the computer vision technology ZIA that Neurolabs provided, we were able to monitor the store shelves closely. This increased the visibility and awareness of what was happening in our shops in real-time.”
Not only can UVESCO react to inventory gaps in real time but the solution allows them to be more proactive with their inventory analysis, allowing them to stay one step ahead at all times.
“We would call this project a success. For the first time, we’ve been able to quantify the extent of inventory inefficiencies and learn lessons that will help us improve our processes. It’s a great first step and we are looking forward to extending the analysis to other stores and regions.”
Proof of Concept to Rollout
Following the pilot’s success, UVESCO green lighted a larger roll out and, working with solution provider Xabet, Neurolabs expanded the technology’s deployment to multiple stores. This corresponds to the post-COVID trends of increasing demand from retailers for automated solutions such as CV technology to tackle On-Shelf Availability (OSA) and Out Of Stock (OSS) challenges. Those that are fastest to adapt to this trend will undoubtedly see the greatest reward over the long term.
Neurolabs’ object recognition software, ZIA, can maximise efficiency not only for individual stores but for entire retail networks. The CV technology can be set up in warehouses to monitor stock levels, while in-store monitoring can alert businesses when additional supplies are needed. By automating the inventory management process, retail businesses can expect to see improved results from their supply chains and ultimately, greater profits.
At Neurolabs, we are revolutionising in-store retail performance with our advanced image recognition technology, ZIA. Our cutting-edge technology enables retailers, field marketing agencies and CPG brands to optimise store execution, enhance the customer experience, and boost revenue as we build the most comprehensive 3D asset library for product recognition in the CPG industry.