How to Prevent Customer Cancellations

We all know retention is important. I'm gonna tell you the mistake we made, and I think most SaaS companies make as they're going through that. This generation of SaaS companies is way better than we were, right? You guys are much more thoughtful around the metrics that you track, and knowing that churn and retention is important. But here's the mistake. The mistake is you assume by looking at your cancellations, it's like, okay, well, people are not canceling. That means they must be happy with the product. Well, it does not mean that they are happy with the product. It just means for any number of reasons, one of which could be, They're happy with the product, but it could be they're too lazy to cancel, they didn't even know they were paying you, essentially, which happens, by the way, shockingly. So any number of things. So what you have to do, what we did at HubSpot, we built in the first year, is this thing called the Customer Happiness Index. It was a model that we developed internally that looked at all the data available to us, product usage, which sales rep sold it, what day of the week they bought on, which customer service manager, like everything we could possibly have about the customer, all in one database, and we did regression analysis. And the idea behind the customer happiness index was for it to be a predictive model that says, knowing everything we know, how likely is this customer to still be a customer next month? Right? And in fact, not only did we— and we tweaked the model as you would expect, being a bunch of MIT folks, and we made the model relatively accurate. So we could know. It's like, okay, Mr. Smith— we had a team that does this now. Mr. Smith, you don't know this, but you're going to cancel in 4.2 months. Is there anything we can do to kind of help you? By the way, like here we have, it's all included in the product price anyway, why don't you use this and this can help you with that? One of the most brilliant things that we did. So the idea here is don't wait until after they cancel to analyze the data. A lot of the data you need, you already have. And cancellation is a post facto kind of indicator of unhappiness. Try to catch it earlier so you can do something about it.

How do you find that pre-turn? One way is an NPS and surveys, but any insights on how you track, really get to this data?

We do NPS surveys. We haven't had massive amounts of success in terms of finding correlations between NPS in the short run. It's like, okay, well, people will say something, but then the next month they'll cancel in a second.

That's the problem, yes.

So what we like is we like behavioral data. We don't ask people what they think, we watch what they do. They may say they're happy, but you haven't logged into the product in 3 weeks. You can't be happy happy, right? That's the problem. Or you think you're happy, but you have the sales rep that was notorious for having a higher churn rate than average. There's something not quite right. So we had all these factors that went in, And the early part of the customer happiness index, the model was very complicated. Because we threw everything, every piece of data we could find and get our hands readily on, behavioral data, demographic data, all of it in one standard database. And over time, the thing we discovered was it really came down to like 4 things that were the actual predictors of whether they would cancel or not. And so now we've reduced it down to like those are the things, and we'll rerun it probably maybe every year or something.

You can find at your scale, because this is what I think is hard, is unhappy engaged customers, because it's easy, it's quote easy, and we have all the vendors here where you can track lack of usage, right? That's work, right? The really hard ones to find are the prisoners.

Totally.

I hate your product, but it's core to my business, and so you don't see it in usage, and then bam, they move to a competitor in one day, right?

Yep, totally.

So have you been able to find that?

Yeah, so we've tweaked some of that out. So the way we do that is, so one of the issues with, and we have, you all should have, like the kind of customer success teams, right?

Yep.

They're onboarding, and their job is to make a customer successful and help them use the features of the product. One of the things that does though, it sometimes distorts the data. So when they're using something, well, are they using it because they were intrinsically like, okay, they saw some benefit, or is it because we kind of banged on their head until they started doing the thing, right? Like writing a blog article or doing— so you have to somehow separate the things, the data that's kind of pure to some degree versus the thing that was not forced, but like that we kind of tainted the data as a result of our own kind of intrusion, with good intent, no doubt,