Solving Churn at Scale

Effective Product Management depends on how well we understand our users—which depends, in turn, on our ability to ask the right questions and use data to get the answers we need along with some degree of intuition.

PMs have to balance data and intuition when making product decisions. The role data plays in decision making is like giving assurance to your decisions so that things are eliminated in whiteboard instead of the sprint board an understanding the significance of the problem we are trying to solve in terms of how much impact it would have if it is launched. It helps to keep our biases away and keep everyone on the same page.

Cohort Analysis and Measuring Value for Hybrid.Chat:

As the founding PM at Smarter.Codes, one of the very first products I had the pleasure of working on was Hybrid.Chat.

The starting state being, it was a product that allowed for users to build out a chatbot using Google Spreadsheets and then embed the widget on their website. The product also offered a Live Chat solution which did not use the Google Spreadsheets route.

After measuring the Current State, I felt intuitively that using Spreadsheets was painful and required buy-in from stakeholders to investigate further through customer interviews. As such, I knew that I needed to present data supporting this hypothesis.

There was no instrumentation in place to collect product data. Marketing data was being collected through Google Analytics and there was Hotjar to record session data. This wasn't enough to do a cohort analysis for understanding value. Partnering with the lead developer, I got access to the event data that the backend systems captured. Using subscription data from Chargebee, I joined the two tables to add in another dimension for monetization.

While it did the job, this exercise raised even more questions in terms of event data. This eventually led to me heading the implementation of Mixpanel and Segment to collect all product and marketing data across multiple touch-points and properties for Hybrid.Chat.

Solving Churn for

We were hired by (a digital signature SaaS), and I played the role of a PM consultant to implement their product analytics stack using Mixpanel and Segment. Through this project, I was able to develop a working relationship with the product leadership and was able to amass enough rank and credibility to carve out a bit of space for me to propose ideas and experiments. I was able to provide further value in two more projects/experiments by leveraging this.

The big problem that they had was user churn. This was back in the day when they had a pay-as-you-go pricing tier. Meaning almost nobody closed their account. They just stopped using the product. So I felt we needed to look at accounts that “go dark”.

After getting the buy-in from stakeholders and a time budget, I went in with a 2 person internal dev team to collect user data from their legacy analytics stack.

The quest was to look for behavioural churn - as there existed a possibility that the users gave up on the product months or years earlier. Based on this, from the dataset, we excluded “one-and-done”s to get to the root cause.

Many users signed up for the product, created a couple of documents and then “went dark”. By matching user data and painstaking manual research, I found that most of them are small/occasional users who don’t really need the product every day.

The focus then became narrowing down the list to new signups with legit businesses that had a bad first experience. This is because it was clear that for this cohort it was an onboarding & activation problem.

We also had “false positives” who stop using the product for a while, then predictably re-start. (e.g seasonal businesses, monthly billers, etc)

So now it was down to tenured, non-seasonal businesses who “go dark”. Unfortunately, that was still hundreds of accounts. Therefore, needing another lens to narrow down the list, I had the idea to look at it from a revenue churn perspective instead of an account churn, as eSignGenie’s revenue was concentrated: ~70% of their revenue came from ~10% of their users.

This left the list to Medium to Large businesses that have been with the product for more than 3 months, and transact regularly. To understand what pain points they experienced with the platform and since the Mixpanel + Segment Analytics stack did not have past data, we spent a month diligently reconstructing each history - logging into every system, user events, customer service, etc. Getting access to those systems was painstaking :) Next, by grouping them into about 10 scenarios, I had a sort of predictive model that was translated into a custom dashboard on Mixpanel. Queries to flag users hitting any of the 10 scenarios fed the dashboard and this was shared with customer service, who could then reach out to the users and help fix those problems.

Since fixing at scale was too hard for a startup with limited resources, I narrowed down the problem to solve at a sub-scale level.

©2023 Dheeraj C

©2023 Dheeraj C

©2023 Dheeraj C