In 2016, a viral YouTube video showcased a new convenience store concept. It quickly gained over 16 million views and sparked much online chatter. The store allowed customers to walk in, pick up their items, and leave. No queues, no checkout. The future felt like it was here, and it was. The store was launched to the public in 2018 and now has 43 worldwide.
Behind this innovative concept, e-commerce giant Amazon has employed cutting-edge technologies. This includes computer vision, advanced deep learning algorithms, and sensor fusion to automate the purchase, checkout and payment processes in its Go stores. This revolutionary store model utilised smartphones and geofencing technology to optimise the overall customer experience.
The introduction of such technology marked a significant shift in the way people shop and interact with retail stores. It is akin to the impact of self-checkout systems when they were first introduced. However, the advancements in technology and data-driven solutions have not stopped there. This is where synthetic data comes in, offering new possibilities for enhancing the future of retail and elevating CX to even greater heights.
The advantages of synthetic data in retail
The retail world has a host of data on transactions and product purchases. But this real-world data becomes more complex to accrue when monitoring interactions in-store. This includes analysing how customers interact with the store layout, product placement, spillages, checkout scenarios and spotting theft behaviour.
Moreover, the collection of real-world data brings with it concerns surrounding adherence to data privacy and image licensing regulations. There’s then the coinciding problem of gathering a diverse range of usable, accurate data needed to reliably train AI systems. With fewer scenarios and a more limited dataset to run off, AI is more restricted in its capabilities and more likely to exacerbate bias and discrimination.
Synthetic data generates swathes of realistic data and information without collecting or processing vast amounts of real-world data. This enhances privacy as personal identification information (PII) doesn’t need to be exposed or handed over. Crucially, it allows retailers to reproduce an infinite range of in-store interactions that may not be possible to carry out in public settings.
Transforming retail operations with synthetic data
Real-world reenactments can be expensive, time-consuming and, in some cases, unfeasible due to safety concerns. As mentioned, there can also be limited accurate datasets to use. As such, virtual worlds can dramatically enhance the efficiency and safety of both training for retail operations and real-world operations themselves.
Synthetic data and virtual environments provide an adaptable platform that can be reconfigured in a myriad of ways. These platforms can build virtual worlds that structurally and statistically reflect the real-world version. So, once a scene or shop environment has been set up, multiple corner cases can be easily created.
Synthetic data can be used to complement real-world images by rapidly generating diverse product images needed to train AI models. This gives retailers the capability to randomise floors, wall patterns, in store camera positions and lighting to test the AI network against a variety of conditions. The model can produce checkout counters, baskets, trolleys and similar, and editors can work with retailers to create different scenarios and use cases.
In this environment, it is easy to set up scenes that might be dangerous to create in public. Lighting and outdoor weather, for example, can easily be adjusted to show varying conditions. Shelving and store layout can be reconfigured to test out different store designs. Within this, there are a whole variety of ways to position people and products in the shop.
Enhancing customer experiences in retail
So what does this look like in practice?
There are many use cases of using synthetic data technology in retail. When it comes to tracking items, cameras can be tested for produce recognition, product identification, mis-scanning items, items left at the bottom of the basket and any unscanned item losses. Scenarios can then be placed in the context of the wider store; with heat maps able to track movement of people and products, test spillages, check shelf stocking happens in the correct order, and use person reidentification to enact scenarios such as dealing with a lost child.
It is also possible to model an employee checkout, self checkout or no checkout at all. A retailer might feel inclined to adjust the parameters of the use case to test a range of key detection parameters for their network. This could be using a conveyor belt with moving objects to test stock keeping unit (SKU) identification for AI networks. And this enables retailers to trial shop cameras identifying product barcodes in different positions.
All this helps decision makers to design stores for the optimum personalised customer experience. It helps them market products effectively, ensure a constant stream of sufficient stock, facilitate seamless payment and checkout, better customer services and mitigate for loss, theft and rare corner cases.
It also provides a link between the instore and virtual world. Synthetic training data, joined by real-world data from various online channels, can train AI-powered systems and understand the behaviour, likes and trends of customers. This allows retailers to produce tailored customer recommendations, promotions and a consistent omnichannel experience, be that in store or online.
A world of possibilities
Just as the Amazon Go store marked a new shopping era, synthetic data has potential to elevate the retail journey. By removing a dependency on real-world data – but instead melding with it – synthetic data overcomes many of the issues currently limiting the true potential of AI’s development. Its ability to map out an ever-increasing and diverse range of scenarios could deliver AI systems and store designs that we can’t even comprehend yet. The personalised world of online shopping will increasingly be reflected in the brick and mortar one. With synthetic data, the possibilities for elevating the customer and shopping experience to greater heights are there for the taking.