- Published
How to Build a Simple Customer Rating
In marketing you often need to rank and group leads or customers by some attribute. That might be average check, number of comments on ad posts, number of orders, and so on. Ranking itself is easy — take a metric and sort. But what if you need to form groups so each group gets its own campaign? For example, group customers by “number of orders” over the last quarter. That is where a rating system comes in.
What a rating is
You can read the definition of “rating” on Wikipedia, but for convenience here is our simple definition:
A rating is a numeric measure of how significant a studied metric is, with a limited finite set of values that is smaller than the set of the metric itself.
Why does a rating need a limited finite set? If the rating were unbounded, sorting by it would just be a plain sort of the metric with no grouping.
Rating calculation example
As an example, take customers’ order counts over the last quarter. Suppose we had 10 customers and their order counts look like this:
Customer Order count
1 12 2 32 3 11 4 55 5 44 6 32 7 22 8 14 9 5 10 40
We want to split customers into 3 groups and assign each group a rating from 1 to 3 (group 3 has the highest metrics, group 1 the lowest). There are many ways to calculate a rating — for example based on probability distributions of a random variable — but those are complex and need both the right skills and software. We will use the simplest approach that is still quite effective.
Since our rating has only 3 groups, we split the sample into groups. The full customer sample is 100%, and we want 3 groups, so the split looks like this:
Since our rating has only 3 groups, we split the sample into groups. The full customer sample is 100%, and we want 3 groups, so the split looks like this:
Rating % of customers 1 0%-25% 2 25%-50% 3 75%-100%
Next, sort customers by order count. In our example we sort ascending, and the table becomes:
Row
number Customer Order count
1 9 5
2 3 11
3 1 12
4 8 14
5 7 22
6 2 32
7 6 32
8 10 40
9 5 44
10 4 55Once we have the sorted list, we can calculate the rating.
To get the rating we need the boundaries of the groups defined earlier. Boundaries are calculated like this:
0%-25% = 10 * 25 / 100 = 2.5 ~ 3 (row number)
25%-50% = 10 * 50 / 100 = 5 (row number)50%-75% = 10 * 75 / 100 = 7.5 ~ 8 (row number)
Here 10 is the total number of observations (customers). After we have the boundary row numbers, we can assign ratings. For example, boundary 3 means customers 9, 3, and 1 fall in the first rating band and are the outsiders (first three customers). Ratings for the rest are assigned the same way. In the end we can build a table with customer ratings:
Row
number Customer Order count Rating
1 9 5 1
2 3 11 1
3 1 12 1
4 8 14 2
5 7 22 2
6 2 32 3
7 6 32 3
8 10 40 3
9 5 44 3
10 4 55 3Takeaways
We walked through a simple way to calculate a customer rating by order count; the method works for any number of customers and any metric. The number of groups (we used 3) can vary by case.
Sometimes three groups are enough; sometimes you need many more. When there are few customers, you can calculate a rating by hand or in a spreadsheet — but when there are many customers, many metrics, and different time windows, you can use services that recalculate ratings online. That role often falls to CRM systems, though not every CRM has it out of the box. In datalizeCRM customer ratings are implemented for various metrics. In our ratings we use more advanced statistical modeling methods tuned to each case. Moreover, for all clients on an annual subscription we add ratings for any metrics at no extra charge.
Sometimes three groups are enough; sometimes you need many more. When there are few customers, you can calculate a rating by hand or in a spreadsheet — but when there are many customers, many metrics, and different time windows, you can use services that recalculate ratings online. That role often falls to CRM systems, though not every CRM has it out of the box. In datalizeCRM customer ratings are implemented for various metrics. In our ratings we use more advanced statistical modeling methods tuned to each case. Moreover, for all clients on an annual subscription we add ratings for any metrics at no extra charge.
×
![]()