CUSTOMER ANALYTCS ENGINE
WHY
-
Since sales reps make 1:1 visits to hospitals to explain & promote Pharmaceutical products, hence this data science engine enables personalized recommendations on which customers should be visited more and which channels, contents and frequency would be optimal to increase engagement
WHAT
Machine learning predictive model utilising Pyspark
WHERE
J&J
(2021)
SUMMARY
-
Coded machine learning model using PySpark, creating over 150 features across over 30 data sources as explained below. ​Modelling steps summarised here:
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​
​

PROCESS & OUTPUTS
​​
-
​We aim to optimise sales, hence a chart (see image below) is generated for each customer, which measures the relative impact of different factors against sales:​
-
​Internal factors such as customer characteristics - age, specialty, years of experience, past response to marketing material etc. ​
-
External factors such as competitor drug prices, market share of drugs in the market, COVID-19 cases over time, population size over time etc.​
-
Actionable factors such as historical number of face-to-face visits, webinars, seminars, email marketing, virtual calls in a month
-
​
-
-
-
-
-
-
-
-
-
We focus on the influence of actionable factors (such as the number of face-to-face visits, number of marketing emails in a month, type of email content) on sales
-
​Too many interactions result in customer fatigue, hence we use machine learning to automatically find the optimal amount for each customer​ ​
-
-
These outputs are condensed into a sales rep : customer summary table to validate with the commercial team​​​​
-
Business constraints are added to the machine learning engine as rules - e.g. a sales rep has a minimum & maximum number of meetings in a month, some customers are not accessible due to location etc.
-
-
Micro-segments are created based on top 5 important characteristics, so marketing team can utilise customer segmentation for future campaigns​
-
Final recommendation for actionable factors are condensed & sent to each sales rep to carry out. Sales results are measured over time
​
​
(Utilized Light GBM algorithm for tuning, Kedro python framework for building modular data pipelines, AWS cloud for data storage and agile scrum methodology)

