Visualising Customer Call Patterns

I came across an article which demonstrated an interesting technique to visualise Twitter account activity across multiple time-scales. The author plotted individual tweets on a scatter plot, using time since previous tweet (x-axis value) and time until next tweet (y-axis value).

Given that a twitter account may have no activity over many weeks and bursts activity over a few seconds (in which case the account is probably a bot), the author used a log-log scale for both axes.

Having come across a sample database of a bank's customer call centre activity, I decided to apply this same visualisation technique to all customer calls made to the bank during the year of 1999. 

Each data point on the following scatter plot (with heat map filter) corresponds to one customer's call.

The horizontal position (the x-axis) of the data point is the time since the customer's previous call; the vertical position (the y-axis) of the data point is the time until their next call. Red regions represent a higher volume of calls with the same time since/until previous/next calls.

The results show that callers will tend to call again in a time-frame which reflects the length of time between their last two calls. In shorter, hourly, time-frames, I suspect that bank opening hours would have some effect, however for time-frames over one week, it looks like if callers call again it will most often be within 1-3 weeks.

I also plotted the same information but this time coloured the data points based on the time of day when the call was made.

Without a heat-map filter, the raw data points show the 'bands' of time between calls. I suspect the evening calls taking place around 10 hours after calls (between 4 hours and 1 day) could be instances of three calls occurring during the business hours of one day.