Case Study

How the world’s most innovative FMCG company crowdsourced Machine Learning models to understand consumer preference

Hindustan Unilever Limited

www.humarashop.com is an e-commerce initiative of Hindustan Unilever Limited (HUL), which helps small retailers have an online presence by creating a different page for each store and carrying out the delivery for them.

HUL’s aim is to become the largest e-commerce player in the grocery/FMCG segment. The company believes that the key to this lies in understanding the changes in consumer behavior and preferences and adapting accordingly.

What’s HUL's story?

HUL is India’s largest, fast-moving consumer goods company with a heritage of over 80 years in India. On any given day, two billion people use Unilever products to look good, feel good, and get more out of life.

The company is a subsidiary of Unilever with sales in over 190 countries and an annual sales turnover of around €52.7 billion. HUL is one of the most innovative companies in the world with more than 35 brands spanning 20 distinct categories.

Goal

Understand consumer preferences in small retail stores in neighborhoods by capturing sales data through the point of sales system and leverage it with innovative Machine Learning (ML) and analytical models.

Challenges

Need for a sophisticated innovation platform that would allow them to engage and collaborate externally to generate ideas and prototypes
Access to a large and active ML community

Solution: HackerEarth Sprint

HackerEarth Sprint is an innovation management software solution which has an in-built ML platform. It enabled HUL to conduct ML challenges where the company provided actual sales data from India’s top 6 Point of Sales (POS) systems. This allowed users to build ML models and manage submissions.
With a global community of over one million developers and 500,000 data scientists, IoT specialists, and Big Data engineers, HackerEarth’s community was quintessential to HUL’s open-innovation model.

Machine Learning innovation campaign

HUL provided the sales data sets from the top 6 POS systems. The company invited data scientists and Machine Learning enthusiasts to submit their ideas and build applications.

The Machine Learning innovation campaign was conducted in three phases:

1.Idea submission

HUL selected Machine Learning and Analytics as the theme of innovation. The participants were invited to submit their ideas on any of the given themes. This ideation phase lasted for 40 days. A total of 130+ ideas submitted from 2004+ teams.

2.Application submission

In this phase, the best ideas were shortlisted. Sprint’s proprietary algorithm and analytics-aided decision making allowed them to segregate transformative and incremental ideas and eliminate the rest.

3.Offline presentation of applications

The 12 shortlisted teams were invited to present their applications.

Winning ideas

Application 1

Functionality: Accurate prediction of data by extrapolation using ML algorithms

Theme: Analytics

The central idea focused on the following points:

Filtering data
Achieving more accuracy
Predicting the next week’s data and letting the store manager know about the products for which they may run out of stock.

The team was successful in filtering the data with prominent accuracy. Accurate predictions were made by extrapolation using Machine Learning algorithms.

Application 2

Functionality: Auto-scalable and low-cost analytics solution for small retail stores

Theme: Analytics

The solution included the following:

Code that captures data from a camera and bar-scanner with a RaPi module
Code that uploads the data to an S3 cloud bucket
Algorithms to convert the data in the camera into analytics-ready data
Database schema that gives a structured view and flow of the captured data, which is received from multiple sources
Tableau and Spotfire to read the data from the database and create visualisation of the overall performance of small retail stores

Application 3

Functionality: ML application for prediction of product sales and demand

Theme: Machine Learning