Business Context: A Leading Fashion brand uses direct sales agents to sell products directly to the customer. Agents host product collection in their local areas . fashion brand had reported a decline in the revenues for past couple of years along with decline in sales force. Agents being hired were continuously under-performing.
Solution: Identify Sales force likely to perform at hiring stage. As part of hiring process assign performance score and potential revenue to each applicant. This helps in identifying potential high performers in next one year as they continue to expand sales force going forward. Agent were classified to be a high performer if he exceeded threshold sales targets in a year.
Pattern recognition were build on all available data at hiring stage for agents which included external data* with more than 1000 attributes and sub- attributes at household level, namely, household demographics, life style attributes, and financial attributes. Pattern recognition used combination of business understanding, response rate classification, categorization and Weight of Evidence- Information Value as variable selection techniques post data sanity had been verified. This was required as final model needed to have a manageable and meaningful set of attributes that would represent high performers.
Different Machine Learning & Statistical Models were built in champion challenger environment. Final selection of model was based on its accuracy in classification and one that represented most actionable attributes. Overall model had classification accuracy of 87.5% using Support Vector Machine. Below is the corresponding lift chart.
If agents were hired randomly, out of 100 agents only 15% would be among top performers as the overall response rate of agent was 15% but with the help of the prediction model, out of the same 100 agents, 56% would be among the top performers.
Some key attributes representative of top performers were built around
- Educational Qualification.
- Involvement in Partying and outdoor sports activities
- Particular communities
- Household market value between certain bands.
- Financial Attributes: Economic Stability indicator, Household Income, Premium credit card user, Home owner or renter, Home market value etc.
- Household Demographics: Country of origin, Race, Gender, Marital status, generation in household, educational background etc.
- Lifestyle Attributes: Dwelling type of agent, Membership clubs, Magazines read by agent, interest in casino, Golf player, interest in arts and antiques, outdoor grouping etc.
- Companies made significant changes in the hiring process i.e. agents were hired basis scores, resulting in increased Top performers.
- Revenue per agent was increased substantially.
- Agents who were dormant but segmented as top performers were re-mobilized.
- Helped company to make strategies to boost the performance of under-performing agents and secondly driving the top performers to excellence .i.e. more focused sales efforts.
This demonstrates clearly how we can use Machine Learning to solve business problems with high degree of accuracy.