Gate One, Nvidia and Silicon Valley Bank recently hosted a breakfast seminar to explore the seismic changes AI and video analytics are having on the retail industry.
This is a particularly interesting time for retailers – not just because implementation is still low and early adopters have a huge opportunity to reap rewards, but because the public are waking up to the privacy implications, which could derail efforts if changes aren’t managed carefully. During the seminar, we explored AI and video analytics in retail from every angle.
Charbel Aoun (Nvidia) shared the exciting opportunities provided by existing technology, ranging from loss prevention, to store analytics, to automous shopping. Use of these technologies is largely considered forward thinking – yet many organisations are in fact already using them, in some cases, with rapid pay back of upfront costs in savings or increased revenues.
Jacqueline Gazey (Senior Partner, The Privacy Partnership) explained the obligation for organisations intending to process biometric data to demonstrate necessity and proportionality. The deployment of video analytics, on the other hand, is wide open, provided privacy considerations are factored in from the outset.
Ben Tye (Client Director, Gate One) walked us through the possible barriers to implementation of AI and video analytics. He shared an effective model used by Gate One for scaling innovation, from proof of concept (PoC) through to real business transformation, recognising that significant benefits are only realised when these solutions are deployed at scale.
Alex McCracken (Managing Director, Silicon Valley Bank) hosted a panel including Charbel Aoun, Jacqueline Gazey, Daniel Hulme (Satalia), Bethany Ayers (Peak) and Graham Cooke (Quibit), which discussed the misconceptions of AI in the retail industry, the underlying reasons for resistance to change, and the issues faced when deploying AI for the host of retailers they’ve worked with.
Stephen Beach (Managing Director EMEA, BriefCam) provided a demonstration of BriefCam’s video analytics software. Its versatility in tackling many of the physical store challenges faced by retailers today, from tracking frequent customer paths, to rapidly distilling hours of CCTV footage into minutes to investigate theft, to filtering footage of customers by size and clothing colour to find a missing child, was evident.
There was a tremendous amount of discussion, but here is a recap of some of the key points made.
Many retailers think they are using AI when they are in fact using machine learning. Also, many executives are being put off by use of the term ‘AI’ and its implications. This may be increased by the lack of an industry-wide definition of the term AI. Daniel Hulme described two definitions:
- “Simple” Machine Learning – getting machines to do things humans can do, but better. Also known as automation.
- Goal-directive adaptive behaviour, which is genuine AI and is not yet happening at scale.
- The major stumbling blocks to adoption and deployment are lack of access to proper data, improper tools and putting technology first rather than the business problems.
- The biggest resistance to this change is ignorance, fear and lack of understanding. Despite numerous positive examples, with clear demonstrable ROI, getting proof of concept (PoC) off the ground is proving challenging and there are limited examples of solutions being deployed at scale.
- If you’re considering running a PoC or scale and deploy then you need to consider: the most promising use cases to address, secure business-led sponsorship, software and hardware solutions to select, the necessary Data Protection Impact Assessments and other privacy impacts. When running a PoC you will need to demonstrate value and understanding of the organisational capability needs that must be addressed before scaling.
- Companies don’t have machine learning problems; they have decision-making or optimisation problems. Therefore, they aren’t seeing the value from their data science teams and need optimisation specialists.
- These technology solutions are creating new organisational capability and operating model requirements in the business. This is particularly true around who, where and how decisions are made. In some cases, there’s a disconnect between the data science/analytics teams and business operations.
- GDPR doesn’t need to be a barrier to adoption when it comes to AI. Each use case needs to be assessed on its own merits, as there is no single answer. Companies tend to get into difficulty when they mix and match different data intended for different purposes. This can tip them into GDPR compliance issues and special category data.
- Successful deployments of AI and video analytics tend to be those where transparency with staff, customers and stakeholders is a guiding principle from the outset.
- A plethora of impressive integration solutions addressing different parts of the value chain are now available.
- They key benefits from machine learning in retail include warehousing (which can eradicate even the need for human decision-making), merchandising, reducing stock turnover times and marketing.
- Several retailers are privately reporting ROI within six months with initial PoC trials.
We are, of course, in the early stages of this debate and there will be many more unforeseen challenges and opportunities that arise from the use of AI in retail. We look forward to staying close to, and being part of, future developments in this space.