Font Readability matters for Conversion Rates -- and other Results

With enough time, you can literally test everything and see if it makes a difference. I recently concluded a massive multivariate test at media.io that tested:

  • Three different fonts.
  • Five different background colours.
  • Three different logos.
  • Three different line styles.
  • Two different button styles.

Overall, over 140 different variations were tested on half the traffic of media.io.

Some of the test results include:

  • Chosing Arial over Verdana and Tahoma increased conversion rates noticeably. Personally, I would rate the font readability of Arial lower than Verdana (especially at 12 px) but it seems that users prefer Arial.
  • Background colours and logos affected conversion rates a lot, too, whereas…
  • …link styles didn’t.
  • The impact of button styles in this design didn't make a lot of difference either.

In response to the multivariate test, I have redesigned the media.io website slightly – and made it even faster. Below this post, you’ll find the usual “before” and “after” screenshots. I can’t comment yet on what the net gain in conversion rate of this redesign was but we’re looking at a number of around 5–10 %.

Reliability Updates

On 2010-08-17, I implemented a subtle change with massive implications for media.io. I did some deep analysis on crash dumps and configured the application so that the seemingly random crashes should be a thing of the past.

Server Migration

I’m also working on moving media.io to a new server that features more up-to-date software. Many people are trying to convert audio that the old software I’m using now can’t convert (this includes floating-point WAV or Windows Media v9).

A new server with more up-to-date software is now 90 % set up. In the next couple of days, I will redirect half the traffic of media.io to the new server. When you convert audio online at media.io, you will soon be randomly routed to one of the servers.

     
Click here to download:
Font_Readability_matters_for_C.zip (84 KB)

media.io update #2

I haven't done anything big to media.io in the last two days – I’ve been busy setting up two SSDs.

One thing I'm doing right now is tracing the memory usage of media.io. What I found surprising is how many programs print awful amounts of debug information. If you simply read all lines using something like BufferedReader, a lot of temporary Strings will be generated.

One thing I do to keep memory usage down is to ignore stderr. Also, I only read into a byte array from stderr to reduce object allocations.

It still takes a lot of tweaks to build something that will scale well to many users. :-)