Don’t look at Average Time to Conversion

At Custora, we have a lot of clients who run a free to paid business model. These customers are interested in improving their free to paid conversions, and we help them with that by providing accurate metrics, and providing e-mail tools. In this post, we’ll talk about how to run these analysis yourself, and common pitfalls to watch out for.

In retail businesses this is especially common in flash sales and group buying sites, such as Groupon or Everlane, where retailers require customers to give their email address before the customers can see the deals. It is also common in Software as a Service businesses, such as Dropbox where there is a free usage tier.

Businesses like these want to get these members into paying customers. This is a critically important step in the customer lifecycle. More and more, companies have been using data to figure out how their marketing efforts are doing, and to figure out what motivates members to become paying customers.

However this particular business challenge can be a bit tricky to optimize for some businesses because the two metrics that most people think of first are misleading. At Custora, many of our clients ask to see the average time to conversion and the average conversion rate. However, these metrics are actually misleading.

If a marketer wanted to increase the conversion rate, he might plot conversion rate as a function of sign up date. He would end up with a graph like this.


This looks bad, conversion rates have been dropping from pretty high when we first started to nearly zero now.

The marketer would likely be rather disappointed with this graph, seeing conversion rates plummet. With trepidation, he plots the time to conversion as a function of join rate, and sees a graph that looks like this.

He is delighted by what he sees. The time to conversion has been dropping rapidly. He concludes that even though people are not converting as much, the ones who do convert are doing so faster.

Or are they?

Well let’s do a thought experiment. Think about the oldest customers. They signed up months ago, and have a higher conversion rate, since they’ve been around longer. They also have a higher average conversion time because the guys that take a long time to conevrt bring down the average conversion time. The new cohorts have a lower average conversion time. For example the cohort that is two months old has a maximum time to converison of 60 days. But a cohort that is 12 months old has a maximum time to conversion of 365 days. So older cohorts have higher time to conversions.

So what do we do about this? We need to pick another metric to focus on, one that does not depend on the age of the cohort. There are two good approaches to do this. One is rather than summarizing the data down to single point, to plot how the conversion rates progresses after sign up.

For this we will group the customers by month of acquisition and plot their conversion rates over time. This creates a graph that looks something like this.

The other is to reduce it into a single statistic that is age independent. One metric that we like to use is conversion rate at 30 days since sign-up since it captures much of the conversion rate, and does not take too long to measure. This creates a graph that looks like so:

You can use other time frames as well, conversion rate on day of sign-up, conversion rate 90 days after sign-up. The longer time frame, the more conversions you capture, but the longer you have to wait before you can see how your marketing is doing.

In conclusion, it is important to measure how your free to paid conversions are performing, and it is important to pick the right measure to do this.

If you are interested in improving your free to paid conversion, you can check out Custora’s lifecycle marketing platform.

  • Andrew Z

    Using the method in the first part of the article, you would have a similar problem looking at the retention of customers, and your solution is like survival analysis (also called time to event analysis): cases are usually measured in days relative to a common event where 100% of the group is in one state (not converted) and eventually some, but often not all, members of the group change to a second state (converted). Survival analysis “whether and when” the cases (customers) change state (convert), or you can simplify it to just “whether” by looking at fixed periods such as 30 days or 90 days (as you suggested).

  • http://twitter.com/reinpk Peter Reinhardt

    Ah there are so many pitfalls in data analysis. Thanks for sharing. I think you have a typo here: “They also have a higher average conversion time because the guys that take a long time to conevrt bring **down the average conversion time**.” I think you mean something like “They also have a higher average conversion time because some of the people that have been around for a long time _finally_ convert and pull UP the average conversion time.”