About the information we collect

Metabase uses Google Analytics and Snowplow to collect anonymous usage information from the installed servers that enable this feature. Below is a representative list of the events we have instrumented, as well as the information we collect about the user performing the action and the instance being used.

We also will phone home some anonymized metrics from the metabase application server on a nightly basis. Note that we don’t collect any usernames, any emails, the server IP, database details of any kind or any personally identifiable information in this process.

While this list of anonymous information we collect might seem long, it’s useful to compare this to other alternatives. With a typical SaaS platform, not only will this information be collected, but it will also be accompanied by information about your data, how often it is accessed, the specific queries that you use, specific numbers of records all tied to your company and current plan.

We collect this information to improve your experience and the quality of Metabase. We use the information you voluntarily share to understand how our users are actually using our product, what kinds of features to prioritize, and how many items to target in our design process. For example, without knowing what the distribution of the number of accounts on each instance in our install base is, we can’t know whether the tools we provide can’t scale until someone complains. And even then, we only hear complaints and not the people who are quietly happily using us. We’re striving to create the best product possible.

If you prefer not to provide us with this anonymous usage data, you can always go to your instance’s admin section and toggle off the option for Anonymous Tracking.

If you’re in the process of setting up your Metabase, an admin can also toggle off tracking during the Usage Data Preferences onboarding step. We collect a few anonymous events before that point, but won’t do so anymore if you choose to opt out.

In the list below, we spell out exactly why we collect each bit of information.

Example questions we want to answer

  • Is our query interface working?
    • Are users stopping halfway through a question?
    • Are users using filters?
    • Are users using groupings?
    • How often are users using bare rows vs other aggregation options?
    • Are people clicking on column headings to sort or manually adding a sort clause?
  • How often are users writing SQL instead of using the query interface?
    • Are these queries written by a select group of analysts or is the entire company SQL literate?
  • Are people using dashboards as a starting point for queries?
  • How many clicks are there on dashboard cards?
  • How many of these clicks result in modified queries that are executed?
  • How often are questions saved?
  • How often are saved questions added to dashboards?

What we will do with the answers to these questions

  • Prioritize improvements in the query interface vs the SQL interface
  • Optimize the product for common usage patterns
  • Stay on top of browser incompatibilities
  • Optimize our dashboards for either passive consumption or as a starting point for further exploration depending on how they are being used

While we will closely follow reported issues and feature requests, we aim to make as many of our users happy and provide them with improvements in features that matter to them. Allowing us to collect information about your instance gives your users a vote in future improvements in a direct way.

The data we collect

NOTE: We never capture any specific details in any of our tracking methodology such as user details, table names, field names, etc. Collected data is limited to the types of actions users are taking with the product.

Google Analytics Events

Category Action Why we collect this
Links and Page Views General website tracking of what pages are most used This provides better understanding of what parts of the application are liked and used by customers so we know what’s popular and potentially what needs more improvement.
Dashboards When the dashboard dropdown is used, when dashboards are created and updated, what types of edits occur such as adding/removing cards and repositioning. We use this information to understand how dashboards are being used and what types of activities users most commonly do on their dashboards.
Pulses When pulses are created and updated, what types of pulses are created, and how many cards typically go in a pulse. This is used to have a sense for how teams are structuring their push based communication. When and where is information most often sent and how much information allows Metabase to continue improving features around push based data interactions.
Query Builder When questions are saved and viewed along with what types of choices are made such as chart types and query clauses used. Helps the Metabase team understand the basic patterns around how users are accessing their data.
SQL Query When a SQL query is saved or run. This mostly just gives us a sense for when users are bypassing the GUI query interface. We never capture the actual SQL written.
Admin Settings We capture some very basic stats about when settings are updated and if there are ever errors. We also capture non-intrusive settings such as the chosen timezone. We use this information to make sure that users aren’t having problems managing their Metabase instance and it provides us some sense for the most common configuration choices so we can optimize for those cases.
Databases We simply capture when databases are created or removed and what types of databases are being used This helps Metabase ensure that we spend the most time and attention on the types of databases that are most popular to users.
Table Metadata The saving and updates on tables, fields, segments, and metrics are all counted, along with a few other details such as what types of special metadata choices are made. We use this data to help ensure that Metabase provides an appropriate set of options for users to describe their data and also gives us a sense for how much time users spend marking up their schemas.

Snowplow Events

Category Action Why we collect this
Set up General clickthrough tracking of what steps are completed vs. abandoned This provides us with a better understanding of how we can improve the on-ramp for new Metabase users.
Database configuration Whether your database connection attempt is successful vs. unsuccessful, which database you connect with. We don’t collect any of your database credentials here. Helps us understand which databases are most popular and performing as expected so we can invest in improving how Metabase works with them.
Dashboard When you create a new dashboard or add a new question to a dashboard. We can use this to determine if Metabase is adding value to your organization, and therefore if we’re doing our jobs correctly.
Account When a new user is created or a new instance is set up. This powers basic reporting to help us understand how your Metabase team is growing.

Server-side Analytics

Metric An example of why we collect this
Number of Users/Admins and whether SSO is enabled To understand which auth methods are being used, and whether to prioritize features that scale with number of users.
Number of user groups To understand how complicated a permissions model most of our users have, and to make sure that we don’t over-simplify our designs.
Number of Dashboards Whether we need to provide ways to organize dashboards.
Number of Cards per Dashboards Do we need to provide more structure to make long dashboards easier to parse?
Number of Dashboards per Card Are our users only creating a card to put it in a dashboard or are they used in many places?
Types of Databases Which database driver bugs to prioritize
Number of pulses with attachments Are people using attachments?
Number of alerts Are people using alerts? Do they typically have a few or does each user have them?
Number of Collections Do we need to add additional organization tools?
Number of Databases Are users using a single DB or many? How large should the icons for a database in the databrowser be?
Number of Schema Are users actively using namespaces on redshift? Do we actually need to design for 100s of schemas or is that just a small percentage of our users?
Number of Tables What kind of data models are people using? Do we need table search?
Number of Fields Can we pre-fetch all the fields in our metadata api to improve performance for most usres, or should we fetch them per table to scale more efficiently?
Number of Segments Are people using segments widely? If so, should we surface them up higher in the UI?
Number of Metrics Are metrics common? If not, should we remove the Metrics option in the New Question Flow
Number of Queries Run How many queries do our most active instances run per day? Do we need to improve caching?
Number of Query Errors Do we need to change how we display errors in the logs? Are they being spammed?
Query Latencies What percentage of our user base runs queries that allow for iterative querying (less than 1 second)
Timezone We have a bug in a certain timezone, how many users are in that timezone?
Language How many non-English speaking users do we have? How fast should we be pushing internationalization?
OS and JVM Version Can we deprecate Java 7 yet?

Note this is meant to be representative. The actual code that is running to generate this can be audited at https://github.com/metabase/metabase/blob/master/src/metabase/analytics/.

Data privacy

Check out our page on data privacy and security.

Read docs for other versions of Metabase.

Thanks for your feedback!

See something that needs fixing? Propose a change.