Client
Client is one of the world’s leading and oldest CPG giants, and its products are available in over 190 countries worldwide. It owns over 400 brands that offer a multitude of products, including nutritionally balanced foods and indulgent ice creams as well as sustainable soaps and luxurious shampoos. With a legacy that spans over half a century, the company placed immense trust in Mindtree to make sure that the goods get delivered.
Challenge
At the beginning of the engagement with the client, the salesmen recommendations were being generated through a traditional SQL Platform. This included even the various KPI’s which resulted in 2 crore recommendations - specific to each country - each month. To make this happen, the recommendations were generated using complex logic and through Machine Learning algorithms, and this took up four days per country. Thus, this process had to be optimized.
Mindtree had to make sure that the process time as well as the days required to get the desired output were drastically reduced. Moreover, there was also a need to remove manual intervention of any kind. The tricky bit about this scenario was that, to deliver this, the churn process had to be started 5 days in advance (i.e. by the 25th of every month), with only 25 days of sale data - and this was the only way to deliver new recommendations by the first of the next month. Moreover, a re-churn would then be required on the 3rd of each month - with the complete dataset - in order to ensure accuracy.
Solution
To address some of the fundamental flaws in the extant platform, the entire process had to be shifted to a new, more advanced platform, and this platform was Azure DataBricks. As a result of this, certain features could also be brought into existence - such as the automation of the end-to-end recommendation generation process. Moreover, we decided to use Scala coding for fast data retrieval, consolidations and ML Algorithms, all to serve the primary purpose of decreasing the number of working days/hours required to generate salesman recommendations.
Benefits
As a result of shifting the recommendation engine to an Azure platform, Mindtree was able to deliver the following benefits:
- Reduced churn time for each country - going from 4 man days to 1 day
- Real time (and one time) recommendation generation using 29 days of sale data
- No manual interventions required for recommendation generation, and fewer errors overall
- Simpler refreshing of the models with new data - this was a tedious process and a huge challenge on the SQL platform.
- Enabled advanced deep learning/network algorithms on the new platform - in the earlier platform, this was a major constraint
