Measuring Customer Health with Allison Metcalfe (LiveRamp)
Last week on the Bowery Capital Startup Sales Podcast, we hosted Allison Metcalfe, VP of Customer Success at LiveRamp, to discuss using cohort analyses for customer health. Allison shared her experience using cohort analyses, amongst other metrics and exercises, to properly track the health of her customers. At LiveRamp, and in past roles at companies like Demandbase and Jigsaw (nka Data.com), she has taken an analytical approach to ensuring a happy user base, and for gauging things like churn risk or upsell readiness. Of course, customer success is both a science and an art. Spending time with your customers live, as Allison explains in our episode, is absolutely critical to ensure that they are achieving maximum value from the solution. As a business pricing on a usage basis, it has more potential variability in its bookings than traditional recurring revenue SaaS companies, LiveRamp must pay particular attention to the status of its existing clients in order to maintain an accurate picture of the future. Having worked for many years at software companies with both business models, Allison’s point-of-view is particular insightful. Take a listen to the podcast if you haven’t yet had a chance. In this post, I’ll expand on that podcast by explaining: what cohort analyses are, how you might use them for customer health, and how to get started.
What is a cohort analysis and what is it used for?
Cohort analyses are used to slice up time series data so that we can extract more than just the overall trend. Time series data are collected at regular (usually equally-spaced) intervals over a period of time. “Revenue by month for 2016,” for instance. Revenue by month, however, doesn’t tell you anything more than the overall trend. If you want to know why revenue is increasing or decreasing, you could take this analysis a level deeper by adding a second dimension: let’s say geography. So while the x-axis remains time by month, the y-axis of your table is now “Revenue, North America,” “Revenue, Asia,” “Revenue, Europe,” and so on. This could tell you that although North America revenues are stronger than ever, a big drop in Europe is pulling down the overall figure. But what if we’re seeing a decline across the board? Perhaps geography is not the culprit.
Let’s say we suspect a change in product made in February has led to a steeper user learning curve, causing new customers thereafter to take longer to start buying (this is an eCommerce-type example). Such a hypothesis would be hard to verify using our revenue analysis current table because two time dimensions are in play: (1) revenue by calendar month, and (2) revenue by time-since-sign-up (measured discretely in months). This is where cohort analyses shine. A “cohort” of customers, is a group that shares any time-based attribute, such as first purchase month or ticket submitted month, but here we’re using sign-up month. Note that we’re adding a second time dimension to our analysis. Re-imagine our revenue table with the x-axis staying revenue by calendar month, but the y-axis now changed to sign-up month. Here, each row represents a cohort. The second row (February sign-ups cohort) would only show revenue starting in the second column (February revenues). Revenue would only begin for third row in the March column (since this is our third cohort, March sign-ups). This “waterfall” of data now shows you how time-since-sign-up affects your overall revenue trend. You could now test your hypothesis: if correct, you’d expect, ARPU (avg. revenue per user) growth in the first few months post-sign-up to be lower for February & later cohorts than for earlier cohorts.
How would you use a cohort analysis for customer health?
We’ll create a fictitious example to suss out the range of uses of cohort analyses for customer health in partner: MeowCorp is selling all-you-can-eat subscriptions to cat ring tones for $10 per month. We’ll smooth over a few things to make it simple: billings & revenue recognition are instant and there are no plans other than monthly. In 2016, MeowCorp experienced healthy monthly revenue growth. In Q1 2017, however, that growth was stunted due to an unprecedented amount of customers cancelling their subscriptions. As the head of Customer Success at MeowCorp, you are tasked with determining why these customers churned and fixing the problem. Certainly you could look at sign-up month as in the example above. But at MeowCorp, a flat-monthly-fee subscription business, revenue is not variable. So how do we get a sense for indicators of potential churners? There are many other datasets you may want to examine across cohorts to elucidate customer health issues: NPS scores, referral leads, webinar attendance, content consumption, product usage, support tickets submissions, customer success rep proactive outreach, profiles completed, integrations added, etc. MeowCorp aside, quantifiable every measure that you might include in a Customer Health Index (more on CHIs in this post I wrote for Square1) could be valuable to examine using cohort analyses.
Given there are many such resources online already today, I skipped creating an interactive Google Sheet example of a cohort analysis and opted for a written description. If the easily-Google-able resources aren’t working for you or you’re still confused, however, I’d be happy to create one. Just drop me a line on Twitter or LinkedIn. Also, make sure to check out our podcast on this topic with Allison Metcalfe, VP of Customer Success at LiveRamp, if you haven’t already; it should give you more color and inspiration for getting started using cohort analyses for customer health. Thanks and I hope this is helpful!