Breaking Assumptions: Building a Customer Retention Model
Project Type: Quick Wins
Customer retention was decreasing year-over-year for a SaaS company. While our client had assumptions about their customers, they knew their subscription-based revenue model could benefit from a customer retention evaluation.
To align with their 2020 planning, our client engaged AMEND to validate their hypotheses for customer retention and to develop a model to predict customer retention.
The final model allowed us to take data from our client's CRM and output a customer's likelihood to leave, as well as a monthly breakdown of revenue impacts from customer service changes.
Overview
-
Company Overview
-
-
- Industry: Software & Technology
- Privately-held
- $290 Million Revenue
-
-
-
Assessing the Landscape
- Customers spanning various industries, sizes, and geography
- Utilized a homegrown Customer Health Scoring System that incorporated subjective factors to determine at-risk customers.
- One of the country’s largest payroll service software providers
-
Identifying Opportunity
- Drive enterprise value by increasing recurring revenue
- Redesign and improve the Customer Health Scoring Tool
- Incorporation of the tool into new sales strategies
Tools & Programs:
Results
(Within six weeks)
$4,600,000
Est. Gross Loss Reduction
100%
Analyzed "Risks and Impact" Retention Relationship
Strategy
Developed Strategic Roadmap for 2020 Customer Retention
The Solution
Single-Variable Hypothesis Testing
Our client developed theories about why their customers were churning. We then validated the list of assumptions using single-variable testing.
Our client understands its customers at a deeper level and can now shift retention efforts to focus on impactful areas within their control.
Example of hypothesis tested: “Do clients that complete training prior to their first use of our product churn less?”
Classification and Regression Analysis
Math: Linear Regression, Logistic Regression, Classification & Regression Tree
We expanded upon the hypothesis testing by incorporating 30+ variables captured in our client’s data to create a machine learning model. This model helped our client understand the behavior of customers in their current portfolio.
Additionally, this analysis highlighted key attributes about customers that could be indicative of a higher likelihood of leaving the client. We were able to apply this output to historical financial data to create projections for 2020 Annual Recurring Revenue.
Customer Health Monitoring Tool
To readily populate our model's analysis, AMEND developed a proof-of-concept dashboard for customer service and sales representatives. The dashboard can monitor the individual customer portfolios to better address customers who are “at-risk” of leaving.
By giving our client's team real-time data to work with, they can proactively plan sales forecasting in the future, as well as strategize their customer retention approach.
The AMEND Assessment Process
Problem Statement and Initial Design
Assess 20+ hypothesis, collect customer data from Salesforce (financial, sales, customer information), data cleaning and transformaton.
Data Modeling
Hypothesis testing and model used linear regression, logistic regression, and classification trees. All of our data wrangling, cleaning, and modeling was done in R, a statistical programming language.
Model Interpretation and Recommendations
-
- After each hypothesis test and model iteration, we focused heavily on analyzing the results to create an actionable interpretation. We put a lot of effort into model interpretation and the language used to easily present insights that the rest of the business can use.
