Analyze and predict physical customers’ patterns in Hypermarkets with high precision using the latest advancement in technologies
The time has come to announce that after more than a year and a half of research and tests we finally have a proof of concept to launch our new product – Sens.ai – with the main focus on Retail.
Sens.ai provides insights and analytics on the customers’ and employees’ behaviour in physical stores with high precision and reasonable pricing. The product is based on the latest technologies and innovation in IoT, Robotics and AI. It is a great example of how math and technology can be combined in order to obtain top class quality solutions with limited data in the niche of hypermarkets.
The project started in the summer of 2018. We were contacted by a leading retail consultant working with one of the largest retail chains in Bulgaria about a very specific task. The scope was to measure and analyse the customers’ behavior after layout change in one of their hypermarkets and to optimize store’s merchandising with a solution of relatively low costs, easy to deploy and providing high accuracy. During the discussions that followed a new idea crystallized about analysing staff performance, similar to the analytics of your favorite soccer team players’ in England premier league.
What we had at the start
At that time we were using a combination of two main technologies – WiFi tracking and monitoring with top video cameras. Through these passive approaches we were able to provide a solution, but at a high price. The costs included items such as purchase of equipment and the subsequent installation and maintenance. The solution was over complicated and what someone would explain as “kill a mosquito with a bazooka”. Further to it, we aimed our efforts into a different direction to find a more cost effective solution.
After extensive research, we decided to change focus from monitoring of customers to monitoring of baskets and shopping carts used during store visits. We needed to design tailored sensors with long battery life and high precision. Additionally, they had to be easily installed and deployed to stores. The only drawback of this approach was that it covered around 60-80 % of all traffic depending on the type of store. However, it included the customers with the highest volume of purchases and thus most of the volumes in the hypermarkets. We assumed we would lose some fast clients traffic, but the consultant and our hypermarket prospect assured they found the approach quite useful.
We started developing the first version of the sensor and at the beginning of 2019 we made the first test and installation into a test store. The results were not encouraging – most of the data was lost and there was much noise in the control environment. We took a decision to totally change the architecture and the whole technical approach.
Further to it, we started experimenting with different sensors (steep learning curve for us). Our efforts were focused on assembling hardware, embedded development, data extraction, data fusion and many other steps in order to achieve better results. In the summer of 2019 we concluded that the whole hardware needed to be redesigned and tested by a professional company, so we started cooperating with Tecosys, a company with broader experience in IoT and embedded solutions.
In November 2019 we had our first prototype ready with all the sensors and technologies we believed were needed, based on all testings we had previously done. The results still were not encouraging and we froze the project temporarily.
During the last two months and the quarantine of COVID 19, we decided to give it a try again and reviewed all the code and data we had stored. Our team came upon several interesting scientific papers and decided to apply its findings.We were back in the game and the prototype was in the hands of ShopUp, Sergi Sergiev played with it for a couple of days in the middle of March at home. Now is the time to thank his wife, for the support, patience and great pictures she took 🙂
After 3 weeks of experiments with relatively good results we made testing in a real store (taking all the necessary measures due to COVID-19). We tested early in the morning and thanks to the store management support were able to make 6 rounds with a basket and shopping cart, both equipped with our new sensors.
Based on the experiments and data we had collected from our walks at the store, we developed several filters based on Bayse mathematics and support of different experts from that sector.
Further to it, we developed a model for localization and 2 additional models for tracking sensors motion in a noisy environment.
On the video we present the first model for localization.
The models significantly improved the performance of sensors. They are able autocorrect if there are big gaps of missing or wrong data and work very well in noisy environment. The accuracy of tracking which we were able to achieve is 20-50 sm for a store with size of 100 x 130 meters.
We plan within the next 2-3 months, together with our partner Tecosys to create the first batch of ready to be deployed sensors with battery and all relevant infrastructure for gathering, processing, analysisng and visualizing the data. The data from sensors and models will be visualized on our BI platform. We are also considering migrating some of our old visualization dashboard onto the opensource platform SuperSet, which is scalable and able to deal with huge amounts of data. We expect that during the time of quarantine it is going to be difficult to find a place to deploy the product, so…
We are looking for
A store to deploy our first stable version of Sens.ai in a real time environment after 1-2 months time. We hope that the quarantine will end by then and we will all be back to normal.
The perfect test Retailer can be a store situated in Sofia, with a lot of shopping carts or baskets such as: Kaufland, Lidl, Metro, Fantastico, Mr.Bricolage, T Market, Practikeror others similar to them.
If you are in that sector and find the solution Sens.ai relevant to your work and KPIs contact us and we can share more.