Your Sherlock score is based on two things:

  1. Events;
  2. Event weights (the weights you assign them)

Events represent the main actions users can take in your product.

Events weights represent how important each event is to the engagement of your product. The weight of each event will determine how much that event contributes to your overall engagement score. We allow you to weigh each event with a weight between 1 and 10 (if you weigh an event with a 0, it will not count toward your score). You should give the more important events a higher weight; less important a lower weight.

For example, if you have a personal finance application, creating a new budget might be a very important event and should be weighed a 8-10, but creating a tag for a specific expense would be a more common, less important event, so should be given a 1-3 for weight.

The score for any individual user is calculated by multiplying the number of times he/she did each event (in the time frame specified) by the weight given to that event - and then summing up all those totals. As an example:

The total of these event scores is what we call the Raw Score. In order to give you a more “usable” and easily digested, we normalize everyone’s scores to a number between 1-100.

Normalization of Scores

In order to normalize the scores between 1-100, we use a process called Winsorizing (we wins at the 90th percentile threshold). This means, Sherlock:

  • Calculates all raw scores based on your score configuration;
  • Identifies the 90th percentile raw score value;
  • Sets that 90th percentile score at a normalized value of 100;
  • Scales all other raw scores against that value.

As an example, if you had this set of raw scores:

[475, 89, 101, 7, 3, 21, 2, 149, 223, 1, 13, 9, 37]

The 90th percentile for this set of scores is 208.

Any score above 208 will be given a value of 100 and all other scores will be normalized against this value using a calculation ({raw score}/208)*100. The normalized scores in this case are shown in the table below:

Finally, we apply an exponential function to these normalized scores to ensure that differences in the raw scores are more effectively represented.

This will result in Sherlock being able to score the engagement all your users and accounts on a scale 1 and 100:

If you are wondering about how this applies to accounts scores... 

When we score Accounts, we take a count of all the activity for that account - regardless of number of users - and normalize that against all other accounts with activity.

So an account with 20 engaged users will score higher than an account with 5 engaged users. But an account with 5 engaged users will likely score higher than an account with 1 engaged user and 19 less engaged users.

Did this answer your question?