Audience Estimator Error Threshold Calculation | Community
Skip to main content
NickMannion
Community Advisor
Community Advisor
November 20, 2024
Solved

Audience Estimator Error Threshold Calculation

  • November 20, 2024
  • 2 replies
  • 1019 views

I am wondering if anyone knows exactly how the error threshold in the audience estimator is calculated? I understand that it reflects the variance between your audience estimate sample and the overall profile population. How would that be calculated?

 

Thanks,

Nick

Best answer by NickMannion

Hi All, I wanted to provide an update on this. After speaking with some adobe consultants, I was able to determine that the error threshold in the audience estimator is calculated using a binomial confidence interval calculation (see formula below). Essentially, this formula calculates (at a 95% confidence level) the margin of error of your estimator. P represents the proportion of interest which is the ratio of how many profiles in your sample population qualified for your audience/your total sample population. Z is 1.96 which is the Z score for confidence intervals at a 95% confidence level. N is your actual entire profile size. That is how you calculate the audience estimator error threshold

 

Thanks,

Nick

 

2 replies

kautuk_sahni
Community Manager
Community Manager
December 2, 2024

@Tof_Jossic @ReenaJohn @Prateek-Garg @Hitaprem @Ankit_Chaudhary Request you to please look at this question and share your thoughts.

Kautuk Sahni
NickMannion
Community Advisor
Community Advisor
December 11, 2024

Hi, following up on this. Still wondering how this is calculated. Is it a reflection of the variance in profiles between your sample population and the actual population in your profile store?

kautuk_sahni
Community Manager
Community Manager
December 12, 2024

@_manoj_kumar_ @nnakirikanti @atulchavan @indra-2 @hitaprem @brekrut @ccg1706 @abhinavbalooni @kumar29917170hcyp @gigicotruta @akshaykapil23 Kindly take a moment to review this question and share your valuable insights. Your expertise would be greatly appreciated!

Kautuk Sahni
brekrut
Adobe Employee
Adobe Employee
December 12, 2024

Hello @nickmannion 

 

The sample size of the scan depends on the overall number of entities in your Profile store. These sample sizes are represented in the following table:

Entities in Profile store Sample size
Less than 1 millionFull data set
1 to 20 million1 million
Over 20 million5% of total

 

source https://experienceleague.adobe.com/en/docs/experience-platform/segmentation/api/previews-and-estimates

NickMannion
Community Advisor
Community Advisor
December 12, 2024

Hi thanks for the response,

 

Yes, so from my understanding, I know that the sample data is a reflection of 5% of our total profile store and that data is only refreshed when there is a 5% increase or decrease in the number of profiles in our profile store. What I want to understand more is how that error threshold in the estimated profiles is calculated. You know how if you click on the i next to the estimated profiles, it will show you the estimate's error threshold? I don't see any documentation on how that threshold is calculated. We've seen a few instances recently where the actual profile size is way outside of estimated error threshold so I want to understand how that is calculated