I’m thrilled that my team was able to increase our monthly conversion rate in March by 40 percent.

There were a lot of factors, but here’s one part of how I think we did it.

There are always hundreds of ideas on how to improve your product. Here are 3 questions we use to brainstorm ideas and decide what to actually code.

Questions

What part of your model do you want to target?

For example, if you had a site with features x, y, and z, and people who use feature z a lot pay you, but only 20% of your customers use feature z, then you might want to try to increase the usage of that feature.

Prereq: An empirically based guess equation that describes why users pay. E.g. you might see with a decision tree or regression that it looks like a high usage of feature x leads to people paying you.

What is the single improvement that will have the largest change on that part of the model and on paid conversions?

For example, maybe feature z is hidden beneath a few other pages. If you instead introduce people to feature z as soon as they signup, maybe that will increase usage and conversions.

This change can really be anything, though, so it requires, I argue, a good amount of thought. Google has found that changing whitespace gives them 10+% improvements. So, the single improvement can be anything from an additional feature to a text change.

What is the minimum viable way to test whether you’re right?

If the improvement that you’ve thought of involves a lot of code, think of a good proxy for it, and do that instead. Don’t spend a month building anything, just build something so simple it feels like a mock, then see if your customers are interested in it.

After you make this change, you have to be able to know if it worked or not. Simply watching conversion rates is probably not going to be enough. For example, if you am going to introduce feature z to users at the very beginning, then look specifically at the usage of feature z. (Rails aside: gruff graphs plus rake)

Even better, a/b test it and only introduce half your users to the new change.

Within a week or so, it will hopefully be clear whether or not your change worked. Even if it didn’t, you now have a better idea of what works/doesn’t, which will lead to better guesses in the future.