quizprep - 3:41 pm on May 4, 2009 (gmt 0)

dpd1 and Tonearm hit the nail on the head. The daily fluctuations we've recognized over the past 5 months are unique to anything we've seen in the past 8 years of ecommerce. We're seeing:

- Daily traffic extremely steady and predictable.
- Daily revenue wildly erratic and unpredictable.
- Monthly revenue extremely steady and predictable.

For the past 7 years, the revenue per visitor on each individual day would fluctuate less than 5%. Based on the number of visits, we could estimate with extreme accuracy how much revenue we would make on a given day... like clockwork, the actual sales would be between 98% and 102% of our estimate each and every day.

This has changed. Here's our revenue per visitor data from last week:

Monday - 65% of historical estimate.
Tuesday - 75% of estimate
Wednesday - 145% of estimate.
Thursday - 70% of estimate
Friday - 150% of estimate
Saturday - 96% of estimate
Sunday - 115% of estimate

Total for week - 102% of estimate
Total for month - 101% of estimate

Our data for the past several months show similar fluctuations. 7 years of earlier data show steady streams of days with 95% to 105% of estimate with perhaps 10 or 15 outlier days each year (rather than the current 15+ outlier days we're seeing each month.)

This pattern started around October 2008, right around the time the US financial crisis began.

Our largest store sells widgets that workers must buy in order to do their job. Since October, our average order size has decreased, but order frequency has increased proportionally... they have to buy around 12 a year, but instead of buying 6 at a time, they are buying 4 at a time.

As for the reason for daily fluctuations: we're clueless (and we don't really care.) As Weeks mentioned above "use trend data that comes from over a long period of time over a large sample. And, then, don't worry about day-to-day"