Product Update

Mixpanel Product Update: Retiring Predict

  • 9 March 2020
  • 2 replies
  • 349 views

Userlevel 1

Hi everyone,

Brandon, one of Mixpanel’s PMs here.  I’m writing to let you know that over the next several weeks, we will remove the Predict report from Mixpanel.

Why?

We launched Predict to help our customers better target their messages to users, by identifying which users were most and least likely to convert on a goal.  Unfortunately, we’ve found that this is not a key use case for most of our customers.
Given that, we would rather direct our machine learning resources to areas of our product that offer more value.  For example, the technology behind Predict is now helping to power the new Impact Report, which enables product teams to accurately determine how their feature launches changed user behavior, without an A/B test.  Learn more about causal inference in the Impact Report here.

When?

Starting April 6th, you’ll no longer be able to create new Predictions.  However, any existing Predictions will remain, as will Predict Grades on your user profiles.  Messages using Predict Grades as targeting criteria will continue sending.

On April 30th, Predict will be removed from Mixpanel.  Predict Grades on user profiles will no longer be updated, and messages using Predict Grades as targeting criteria will be paused. 

While you will still be able to use Predict Grades to create cohorts and target users, keep in mind that since they will no longer be updated, over time they will likely become inaccurate.  If you’d prefer to remove these grades, delete your Predictions via the instructions here before April 30th.

Please read our FAQ guide here, and comment below if you have any other questions or feedback.

Best,

Brandon


2 replies

Hi Brandon,
Thank you for the post. 
I’m looking for exactly the solution that Predict offers, but it unfortunately looks like I’m a few weeks late. 
We are trying to map user_attributes to a product outcome, and a 1:1 mapping is not sufficient. Ideally I’m looking for a mapping which will tell me that for a % of my user base, what are the min values required among my user_attributes. 

Are you aware of any other user-friendly tools like MixPanel, which can fulfill this type of analysis? 

Thank you,
Adam

Userlevel 1

Hi Adam,

Thanks for the question!  To start, let me make sure I understand what you’re hoping to do - my interpretation is that you’re trying to find what user attributes (or what count within those attributes) leads to users performing a goal?  Let me know if I’m off, and I can re-jigger my suggestions.

But, if that is the case, here’s what I’d do:

  1. Don’t use Predict. 🙂 All Predict does is group users into four groups, those likely, somewhat likely, somewhat unlikely, and unlikely to perform your goal.  Unfortunately, it doesn’t tell you how or why those groups were identified, so you wouldn’t be able to easily attribute the product outcome back to your user attributes.
  2. Start with Insights or Funnels.  In either tool, you can analyze your product outcome (alone in the case of Insights, or as a conversion goal in Funnels), and break out the results by your user attributes.  This will let you see the average value of those user attributes for people who performed the goal, or find values of the user attributes where you start to see a higher correlation with conversion.
  3. Dive deeper with Cohorts.  You might also be interested to see the user attributes for users who didn’t perform the goal, so you can compare them to the attributes of users who did perform the goal.  You can create a cohort of users who did anything but your goal, then use that cohort to filter an Insights report showing users who did Any Event, and repeat the process from step 2.  Then, you can compare your two analyses to find the threshold within the user attribute values where it becomes more likely that a user will perform your product outcome.

Hope this helps! 

Cheers,

Brandon

Reply