Video Analytics with In-store CCTV camera feed for Sports Goods Retailer

When video analytics was initially adopted by retailers, it was done so with the aim of reducing or even avoiding losses from incidents such as shoplifting and employee fraud. It was seen more as a security protocol and nothing more.

However, today video analytics has moved on to being something much more – it has grown to be a tool that can be leveraged for business growth. The video surveillance market is set to hit $ 82 billion USD by 2025. The adoption of video analytics by retailers has increased by 16% year on year for the last 7 years.

Video analytics delivers value to retailers in many forms – in terms of measuring store performance, enhancing the customer experience, increasing engagement with customers and ensuring customer loyalty.

Many retailers face the issue of high footfalls but disproportionate point of sales revenue. The use of video analytics allows retailers to not only make decisions based on footfall data, but also gives them the liberty to drill down further and analyze the following:

• Whether consumers compared a particular product with other brands?
• Who are the returning customers?
• Who exited the store without purchasing anything?
• Whether they looked at an item or spent significant time in a particular area of the store?
• Whether they picked a product?
• Whether they bought multiple products?

Indium’s cognitive analytics capabilities helped one of the largest sporting goods retailers in the world to increase their customer satisfaction levels by more than 15%.

The client had 1500+ stores across more than 45 countries. Being a giant in the space and competing in the fast-moving retail market meant business decisions and market strategies needed to be flexible enough to change quickly as well.

To achieve this, there was a need for real-time store visitor analytics. Footfall data is what most companies use today to achieve this. However, the client wanted to step it up by linking footfall data with POS data. This needed a robust yet simple solution that generated accurate results.

The requirement laid out by the client was to effectively improve performance of the store and increase customer satisfaction. In order to meet the requirements, Indium had to:

• Leverage security cameras across the store to generate a store heat-map.
• Leverage CCTV camera data to understand the variations in footfall by area of the store and time of the day.
• Identify customers using facial recognition to know the number of customers leaving the store without making a physical purchase.
• Build comprehensive dashboards with real-time data refresh for better insight into customer’s behavioral as well as shelf zone analysis.

Seeing this as an opportunity to solve a difficult problem for our client with an out of the box solution, Indium implemented the following solution:

Solution Implementation:

Indium analyzed the video feed data collected from the cameras installed on the shop floor and

built a cognitive analytics model solution that leveraged the data to meet the requirement:

• Firstly, Indium used ImageAI to empower security cameras to count the number of customers who enter the store at a given time period. In addition to this, customized functionality was incorporated which allowed the cameras to count the number of people who performed a particular activity – like walk to the cycling section of the store.
• A neural network model was built for image processing and video analytics. The model was analyzed and optimized in order to ensure maximum accuracy.
• Within the data points that were gathered, outliers were identified as the next step of the solution.
• Pattern recognition was in place around all the shelf zones.
• Even though this was a very complex neural network to build, there could be zero compromise on accuracy.
• Each layer of Neural Network was built by training various classes of the object or person, where 2000+ images were used.
• Annotations were created and tested using sample videos. This particular neural network model which was built would become more and more accurate over time – more the number of videos used, higher the accuracy level over time.

Business Impact:

• An accuracy of more than 80% was achieved in the models specifically built to target customer behavior. This ensured improved customer engagement and better targeting of customers.
• Comprehensive analysis on the conversion rate from visitor statistics to live sales via the POS systems, and analysis of customer interaction in any product section and the product-wise conversion rate, helped improve product placement and cross-selling across product categories.
• The client saw a solid 15% increase in customer satisfaction post implementation of Indium’s solution.
• The easy to use User Interface enabled all stakeholders to get a better insight into customer’s behavioral as well as shelf zone analysis.
• 70% cost savings in short and long-term as the tools used in the project were open source.
• Highly comprehensive dashboards were built with real-time data refresh, tailored to the client’s analytical and business needs.