• John Jones

    Nice introduction to the problem Benny. One suggestion that helps improve the data quality is to calculate User Hour of Week. In the US we have 6-7 time zones and log files record Server Hour of Week. Many (IP address, time zone) databases are available. More subtle is a second problem of conditional dependencies. For example sex and age have a big impact on hour of week conversion rates. And lastly, grouping hours of the week into segments is a good idea. Consider grouping the segments so that your conversion rate changes are greatest across segments while keeping the number of segments small.