# Modeling Revenue in A/B Test

**URL:** <https://discourse.pymc.io/t/modeling-revenue-in-a-b-test/8965>\
**Category:** version agnostic\
**Tags:** modeling\
**Created:** [March 4, 2022, 1:17pm UTC](https://discourse.pymc.io/t/modeling-revenue-in-a-b-test/8965 "2022-03-04T13:17:11Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![EricJ](https://avatars.discourse-cdn.com/v4/letter/e/22d042/32.png) [@EricJ](https://discourse.pymc.io/u/EricJ)\
**Post date:** [March 4, 2022, 1:17pm UTC](https://discourse.pymc.io/t/modeling-revenue-in-a-b-test/8965/1 "2022-03-04T13:17:11Z")

</div>

I have data in the form of:

```auto
Day, Views A, Actions A, Revenues A, Views B, Actions B, Revenues B
1, 100, 5, 21, 98, 6, 30.4
2, 99, 7, 28.8, 105, 6, 29.9
...
...
20 102, 6, 24.5, 97, 8, 40.1 

```

Where:

- Views - Number of views per page.
- Actions - Number of actions made by viewers which generate revenues.
- Revenue - The revenue per day.

All data is given on per day granularity, No access for per user data.  
I am after a model for the parameter `Revenue per View`.  
There are models given per user data (See [Revenue metric for ab testing](https://discourse.pymc.io/t/revenue-metric-for-ab-testing/3800)) but how can one adapt them into aggregated data?

Which model would be reasonable?  
How would you use the actions information in it?

---

<div class="post-metadata">

**Author:** ![MarkW](https://avatars.discourse-cdn.com/v4/letter/m/53a042/32.png) [@MarkW](https://discourse.pymc.io/u/MarkW)\
**Post date:** [March 6, 2022, 7:35am UTC](https://discourse.pymc.io/t/modeling-revenue-in-a-b-test/8965/2 "2022-03-06T07:35:57Z")

</div>

Maybe some model directly on the ratio?

Something like Gaussian with per variation learned mean and shared variance?