B2B cross-sell with predictive scoring
Context & Objectives
A B2B services company needed to develop more analytical agility and work more efficiently to drive commercial outcomes.
The company wished to contact its customers who might be interested in subscribing to additional products and services. However, the challenge rested in identifying cross-selling opportunities for existing clients to target effectively during campaigns. And, knowing where to invest time and effort is crucial for profitability. With data in multiple sources and systems, they could not bring all these elements together to extract valuable commercial insights.
The marketing team used to create the contact lists themselves for the cross-selling campaigns. So they spent a lot of man-days every year analyzing and gathering a list of potential customer opportunities.
Thus, our client set sights to focus on cross-selling and recommendations to improve efficiency, profitability, and customer performance. They called on Agilytic to rapidly deliver a solution to help them focus on cases with a higher chance of cross-selling while recommending the best tactics to the commercial team.
Approach
We worked with the data, strategy, and marketing team to ensure our efforts would have a significant impact. Before starting, we held a workshop with data sources owners to identify relevant data.
First, we gathered data to predict customers’ behavior and identify ideal customer profiles. We worked with four data sources, a CRM, backorder processing tool, ticketing tool, and mailing automation tool, to describe the customers' behavior and their buying habits, and ease the process of scoring customers.
After this data collection and analytical data building phase, we scored the customers. We performed data extraction from different systems and data analysis (Sales seasonality, data quality), feature engineering (aggregations, merge from different sources, cleaning, target definition, and creation), modeling, prediction, and finally, model explanations.
In the end, we created three models. Each model provided a cross-selling score for a different product and on a different time horizon, to illustrate:
Will the customer buy (product 1) in the next 12 months?
Will the customer buy (product 2) in the next three months?
Will the customer buy (product 3) in the next 12 months?
We closely collaborated with our client to validate the result during this process. Together we reviewed the working assumptions to ensure the final quality of the model and our client’s adoption. We tested the model to assess its reliability over time. On top of the fully implementable algorithm, our client has received detailed explanations on the influencing factors.
We delivered the following:
Cross-selling scores per customer (.csv)
Documented code on the VM where the models where created (.py, .ipynb, .md)
Models description and recommendations (.ppt)
Results
In less than 26 days of work, we developed a model that...
extracts data from different sources,
analyzes data,
models,
shows predictions,
and offers explanations
By using the cross-selling models, the client’s marketing team will save a lot of time with a more analytical approach. Our client can now rely on the model’s predicted scores to target customers during their cross-selling campaigns.
