Following is my proposition for analyzing the ordering patterns.
For each customer, create a histogram of orders per unit time. Perform a neighborhood averaging filter, which is a common algorithm in image processing. Basically, a new data set is created, where a given element is calculated by averaging the value of elements in that and neighboring positions in the first set. For example, the 5th entry in the new set is calculated by averaging the 4th, 5th, and 6th entries in the original set. Then take the peak value, normalize it by dividing by this customer's overall average, and this gives a quality factor for how distributed their purchases are.
Consider 3 customers, who each bought 4 items in 8 weeks. Person A bought one item every other week. Person B bought one item every week for the first four weeks. C bought 4 items the first week. The processed set for A is [.5 .5 .5 .5 .5 .5 .5 .5], person B is [1.0 1.0 1.0 .66 .33 0 0 0] and person C is [4 2.66 0 0 0 0 0 0]. The highest value for A is .5 , B is 1.0, and C is 4.0. From these numbers we can tell that A was the most evenly distributed, C had the sharpest spike, and B was in between.
In addition to my idea for the ordering pattern, I would like to mention I have extensive experience with excel, and I believe I could accomplish all of this in 1-2 days. I quoted 7 so you may review my results, and let me know any changes you would like done.
Thank you,
Ronjan
Thank you,