Case studies

IPSG Systems has used effectively Inferneon™ in various scenarios. A couple of interesting use cases are described here.


Apache Hive™

A fashion apparel brand company with about 300 retail outlets across the country distributed their merchandise consisting of several thousands of SKUs (Stock Keeping Units). A key challenge that they were encountering was in deciding the right mix of merchandise to stock in each of the stores. While the fundamental goal was to maximize revenues and profit margins, there were other considerations to be kept in mind while deciding on a mix: a healthy and robust variety of SKUs to ensure freshness, constraints due to a particular set of SKUs that ought to be assigned to a store due to local demand and also limitations of capacity of the displays. As the process of assortment planning was extremely complex given the large number of SKUs and distribution of stores, the merchandise team at the company resorted to ad-hoc choices of the mix at each of the stores based on vague hunches, previous experience and approximate guesses of demand in each store. While the process was necessary, it was error-prone, time-consuming and hard. Moreover, there was no way of validating that the manually determined assortment plan was actually resulting in enhanced revenues or profit margins.

The solution to the problem needed a scientific approach using vigorous analysis and machine learning using earlier sales data. Using Inferneon™, our team assembled a solution, addressing the planning problem as a constrained optimization problem. The team constructed the solution from ground-up and developed an application for the retail company using pre-deployed algorithms already available in Inferneon™, enabling a completely automated approach. When the retail company implemented the assortment plan suggested by the solution, there was a noticeable 17% enhancements in revenues compared to previous seasons of sale. Moreover, the solution is also being used to provide further valuable insights into the data, enabling the company to plan for periodic replenishments of merchandise and decide on optimal pricing of SKUs.



A freeway development corporation needed to record the license plate numbers of vehicles that were using their toll sections for legal compliance and other security purposes. The toll booth attendants were tasked with recording the license number of vehicles passing through the toll booth; however this was resulting in queuing up of vehicles at the booth, longer waiting times for vehicles and imposing heavy load on the attendants. The company evaluated several automatic vehicle license plate recognition software, but none of them were found to be satisfactory. While most tools were providing up to about 80% accuracy, none had robust capacity to learn continuously as they were all statically modelled and there was no facility for online learning. Moreover, the tools used were highly sensitive to ambient conditions like noisy images captured due to poor lighting, weather conditions and presence of other objects in the image. The company requested the services of IPSG systems to help them with the development of a robust system to automatically identify license plate numbers.

Engineers at IPSG Systems Inferneon™, to develop an application to be deployed at the toll booths. The goal was two-fold: to facilitate online learning- preferably unsupervised and to provide a robust solution that was insensitive to ambient conditions. The solution was developed using several techniques using standard object recognition, edge detection as well as deep convolution networks, initially trained with a small subset of pictures under varying conditions. The algorithms were then modified to learn in an online manner, allowing the model to adapt with more sample numbers. This eventually resulted in an accuracy close to about 99%. Additionally, the solution supported training of a large number of images using Spark and Hadoop, thus enabling a scalable system.