How to migrate from Tableau to Metabase

Rank Tableau views by use, read their logic from the workbook XML and the Metadata API, rebuild them in Metabase on live data, check the figures and cut over.

Start with one workloadThe steps

Written for
The engineer who will do the work
Or run by Pilae
At a fixed price, then operated

In short

  • Metabase has no Tableau importer: the workbook XML and the Metadata API give the specification, and each dashboard still in use is rebuilt as Metabase questions.
  • Usage from Tableau admin views or Admin Insights sets the order of the rebuild, and views nobody opened in ninety days go with the licence renewal.
  • Row-level security and SAML sign-on are paid Metabase features; in the open-source edition, audiences that see different rows get their own read-only connection.
  • Every rebuilt dashboard is checked against Tableau over the same closed period, and Tableau stays readable until month-end closes on Metabase.
On this page (14)

Moving from Tableau to Metabase is an export of the logic followed by a rebuild. Metabase cannot open a Tableau workbook, so nothing converts: each dashboard people still use is built again as Metabase questions on a Metabase dashboard. Most of the work is in the queries, not the charts. A calculated field, an extract filter or a level of detail expression carries a decision about what a number means, and the rebuild has to reproduce that decision before anyone trusts the new figure. This guide covers the order we work in: usage first, then the logic, the rebuild, the checks against Tableau and the cutover.

Why teams move from Tableau to Metabase

Tableau is a strong analytics tool, and analysts who know it well will miss parts of it. Organisations move anyway, usually for three reasons.

  • Licences by role. Tableau licenses Creators and Explorers per user, and Viewers per user or, on Tableau Cloud, through a capacity-based licence. Anyone who builds or explores is a seat. The open-source edition of Metabase has no per-user fee at any level, so opening dashboards to the whole organisation, or letting a department build its own questions, is a question of permissions.
  • Live figures instead of copies. An extract is a copy of the data refreshed on a schedule, so the figure on screen is as fresh as the last refresh. Metabase queries the database when a dashboard opens, through a read-only account, and keeps results only if you turn caching on.
  • Reports that stay next to the data. Dashboards read revenue, salaries and customer lists. Metabase runs beside the databases it reads, on your premises or in a region you choose, and reaches them over a private network.

What you give up is Tableau’s free-form visual exploration, its range of chart types and its pixel-precise layouts, along with Tableau Prep and offline analysis in Tableau Desktop. Apache Superset comes closer on chart types for analysts who miss them. In the open-source edition of Metabase you also go without row-level security and SAML sign-on, which are paid features. We list what has no equivalent during the inventory, before you commit.

Step 1: rank workbooks and views by use

The inventory starts from what people open, not from what is published. Where the usage data sits depends on how you run Tableau.

  • Tableau Server. The administrative views, reached from the Server Status or Site Status page, include Traffic to Views, with daily view counts, the most viewed views and the people who open them most, and Stale Content, which lists content not used within a threshold you set, with its last used date. The event history behind these views is kept for a configurable period, 183 days by default.
  • Tableau Cloud. The Admin Insights project holds data sources a site administrator can analyse. TS Events records who opened which view and when, and Site Content lists what is published. They cover 90 days by default.
  • Either, by script. The REST API’s Query Views for Site method returns a total view count for each view when called with includeUsageStatistics=true. A total says nothing about recent use, so it is for finding views nobody has ever opened; the last ninety days come from the admin data.

Put every view in one of three groups. It is rebuilt if people other than its author opened it in the last ninety days. It is merged if several workbooks show the same figures cut different ways: in Metabase these usually become one dashboard with filters. It is retired if nobody opened it, and goes with the licence renewal. Check the subscriptions before retiring anything: people who read a view in their inbox may rarely open it. Views over metrics, logs or anything measured over time go to Grafana instead, which is built for them.

Step 2: export the workbooks and their logic

For every workbook you keep, take the file, the logic and the figures.

  1. Download the workbook. At the top of a view in Tableau Cloud or Tableau Server, click Download, then Tableau Workbook. Which formats you are offered depends on the permissions the owner and site administrator have granted. In bulk, the REST API’s Download Workbook method returns the .twb or .twbx file; add includeExtract=False to leave a large extract out. A .twbx is a zip file: the .twb inside it is the workbook, next to any extracts and local files.
  2. Read the logic from the XML. A .twb is an XML document. It holds the connections and any custom SQL, the calculated fields with their formulas, the parameters, the filters, and every sheet and dashboard. Tableau notes that the content of a workbook file is stored in plain text, filter values included, so keep the downloads where the data itself may be kept.
  3. Query the Metadata API across the site. The Tableau Metadata API is a GraphQL endpoint at /api/metadata/graphql. For each calculated field it returns the formula, the sheets that use it and the tables upstream of it, which turns a pile of workbooks into one list of definitions. On Tableau Server an administrator enables it with tsm maintenance metadata-services enable; on Tableau Cloud it requires Data Management, or an edition that includes it.
  4. Save the figures. For each kept view, pick a closed period, such as a month that has already been reported on, and download its numbers with Download, then Crosstab. These files are the reference for step 6.
  5. List what is not in the file: extract refresh schedules, user filters and other row-level rules, subscriptions and data-driven alerts, and every link to a view from an intranet, a wiki or another application.

Step 3: connect Metabase read-only and model the data

Before rebuilding anything, give Metabase the data Tableau reads, and nothing more.

  • One read-only account per source, limited to the schemas people need and reached over the private network. Heavy dashboards point at a replica, not at the database that takes orders.
  • Tidy the metadata. Metabase syncs the tables and columns of each database. In the table metadata settings, give them readable names and descriptions, set semantic types such as currency or category, and hide the columns nobody should pick. Much of the renaming and formatting a Tableau data source carried ends up here.
  • Turn each published data source into a model. A Metabase model is a curated starting table, built with the query builder or in SQL, with its own column names, descriptions and types. Joins, renamed fields and row-level calculated fields go into it once, and every question built on it inherits them.
  • Define shared numbers as metrics. A metric is a saved aggregation, such as revenue or active customers, that the query builder offers next to Count and Sum. It gives a team one definition of a number instead of one per workbook.

Models or transforms

Metabase describes transforms as the successor to models and says it will gradually phase models out. A transform writes its result as a table in your own database on a schedule, instead of running the query live, so it needs a connection that can write to a target schema. With read-only accounts that is a deliberate exception: one schema set aside for Metabase’s tables, agreed with whoever owns the database.

Step 4: map Tableau concepts to Metabase

The two tools share most ideas under different names. The table is the map for the rebuild; the sections after it cover the four that need care.

Tableau Metabase
Project Collection, with permissions per group
Workbook A dashboard and its questions, saved in a collection
Worksheet Question, in the query builder or in SQL
Dashboard Dashboard, with filters wired to each card
Published data source A database connection, and a model for each shared dataset
Live connection The default: every question queries the database
Extract A live query on a replica, or a summary table in the database
Calculated field, row level Custom column, or a column in a SQL model
Calculated field, aggregate Custom aggregation, or a metric
Table calculation Window function in a SQL question or model
LOD expression Subquery or window function, or a summarised question joined back
Parameter SQL variable, or a dashboard filter
Filter and quick filter Question filter and dashboard filter
Filter action Click behaviour: update a dashboard filter
Go to URL, Go to Sheet action Click behaviour: go to a custom destination
Subscription Dashboard subscription, by email or Slack
Data-driven alert Alert on a question: results, goal line or progress bar
User filter, data policy No equivalent in the open-source edition: see below
Tableau Prep flow SQL views, or the data pipeline you already run

Calculated fields and LOD expressions

Row-level calculated fields become custom columns, written in Metabase’s expression language (case, if, text and date functions), or plain SQL in a model. Aggregate calculations become custom aggregations such as SumIf and CountIf, or metrics when more than one dashboard needs them.

Level of detail expressions need the most care. A FIXED expression computes at the dimensions it names, whatever the view shows; INCLUDE adds dimensions to the view’s level, and EXCLUDE removes them. Metabase has no such syntax, so each one is written out: in SQL, an aggregation in a subquery joined back to the detail, or a window function partitioned by the named dimensions; in the query builder, a summarised saved question joined to the table, which works when both come from the same database. Tableau also computes FIXED expressions before dimension filters, so a filtered view can show a FIXED total that ignores the filter. Reproduce that on purpose or change it on purpose, and tell the owner which.

Extracts

Most extracts disappear: the question reads the source live. Three cases need a decision. An extract filter that kept only part of the data becomes a filter in the model. An extract that aggregated the data to make it small becomes a summary table, or a materialised view, refreshed in the database by your existing pipeline. An extract that existed because the live query was slow becomes a replica, a summary table or cached results. The open-source edition sets one caching policy for the whole instance; caching per question or per dashboard is a paid feature.

Parameters, filters and actions

A Tableau parameter becomes a dashboard filter, or a variable in a SQL question written as {{variable}}. Prefer field filter variables, which map to a column and give people a proper date picker or dropdown, and wire them to the dashboard filters. Filter actions become click behaviour that updates a dashboard filter, so clicking a bar filters the other cards; Go to URL and Go to Sheet actions become custom destinations.

Row-level security

Tableau restricts rows with user filters, USERNAME() calculations, data policies on virtual connections or the database’s own security. The open-source edition of Metabase has none of these: every group can view data, and which saved questions and dashboards people see is decided by collection permissions. Where audiences must see different rows, give each its own database connection, using an account that can only read its rows through database views or the database’s own row-level security, and keep each audience’s dashboards in a collection only that audience can open. When that becomes unwieldy, the paid edition’s row and column security does the job in one place, under a licence you hold from Metabase.

Step 5: rebuild the dashboards in order of use

Start with the most opened view from step 1 and work down the list. The first few set the conventions everyone follows.

  1. Build the questions on the models from step 3. Use the query builder wherever it can express the logic, because viewers can then drill through the results themselves; use SQL where it cannot.
  2. Assemble the dashboard. Place the questions, add the filters and connect each filter to every card it should change. Set click behaviour for the cross-filtering and links the Tableau version had.
  3. File it by audience. Create collections per team or audience and give each group Curate, View or No access. Map groups from your directory over LDAP, so access follows the directory. Decide per group whether it may create its own queries, and with the query builder only or with SQL as well.
  4. Move the deliveries. Recreate Tableau subscriptions as dashboard subscriptions, by email or Slack, with CSV or XLSX attachments where people want the numbers. Recreate data-driven alerts as alerts on the underlying question. Different filter values per subscription are a paid feature; in the open-source edition, a copy of the dashboard per audience, with its filters set, does the same job.
  5. Write down what does not carry over, such as a pixel-precise layout or a chart type Metabase lacks, and agree the replacement with the owner before moving on.

Step 6: check the figures against Tableau

A dashboard is ready when its owner agrees the numbers, not when it looks right.

  1. Run the same closed period with the same filters, and compare against the crosstab files from step 2. Totals first, then each breakdown.
  2. Explain every difference. The usual causes are an extract that was out of date or filtered, a data source filter or context filter nobody remembered, a FIXED expression that ignored a filter, a distinct count compared with a count of rows, null handling, time zones, and the first month of the fiscal year.
  3. Sign off per dashboard. The owner signs off each one. A dashboard whose figures nobody can reconcile is not cut over until someone can.

Step 7: train people and cut over

Train by what people do. Viewers need filters, subscriptions and where their dashboards live. People who explore need the query builder, starting from models, and their personal collection, where everyone can always save. Analysts need the SQL editor, variables and how models are maintained. An hour each, with their own dashboards on screen, works better than a general tour.

The cutover, in order

  1. Pilot: One team moves first, with its dashboards checked, and what they find is fixed before anyone else moves.
  2. One cycle: Owners compare the figures in both tools through one reporting period, team by team.
  3. Links: Bookmarks, intranet pages and wiki links move to Metabase; embedded views become signed embeds, or public links where the figures may be public.
  4. Deliveries: Tableau subscriptions and alerts are switched off as their Metabase versions start, so nobody receives both.
  5. Close: Tableau stays readable until month-end closes on Metabase, then the licences end at renewal, with the downloaded workbooks kept as the record.

Running Metabase in production

A Metabase that a whole organisation reads needs a little more than the default container.

  • PostgreSQL for the application database. It holds users, questions, dashboards and permissions. The embedded H2 file Metabase starts with is for trials and makes restores fragile.
  • Sign-in from your directory. The open-source edition signs people in with LDAP, with LDAP groups mapped to Metabase groups, or with Google. SAML and JWT single sign-on need a paid Metabase licence.
  • Mail and chat for deliveries. Subscriptions and alerts need an SMTP relay, and a Slack connection if people want them there.
  • The application database is your content. Exporting questions and dashboards as files, which Metabase calls serialization, is a paid feature. In the open-source edition, the backup of the application database is the copy of everything you rebuilt.

Pilae operates Metabase this way: PostgreSQL behind it, read-only connections over your private network, daily encrypted backups stored offsite and, on Enterprise, a restore test every month recorded in the console. Each upgrade runs migrations on the application database, so the Pilae Agent tests it on a copy and applies it in your window once you approve. It runs on your premises or on dedicated machines in Pilae Cloud. For a comparison of the options before you start, see our page on Tableau alternatives.

Let Pilae run the move

Pilae runs Tableau to Metabase migrations at a fixed price: the usage inventory, the export of the logic, the rebuild, the checks against Tableau and the cutover. Then we operate Metabase for you under a contract that, from the Business plan, includes a monthly availability commitment with service credits. See migration services for what the fixed price covers, or tell us what your teams report on and we will scope it.

Questions

Can Metabase import Tableau workbooks?

No. Metabase has no importer for .twb or .twbx files, so every dashboard still in use is rebuilt. The workbook is still worth downloading: a .twb file is XML that records the connections, the calculated fields with their formulas, the parameters, the filters and every sheet. It is the specification each Metabase dashboard is rebuilt from.

What replaces Tableau extracts?

Live queries, in most cases. Metabase reads the database when a dashboard opens, so an extract that only existed to copy the data becomes a query against the source or a read replica. Where an extract existed because the live query was too slow, the fix belongs in the database, as a summary table or materialised view your pipeline refreshes. Metabase can also cache the results of slow queries.

How do LOD expressions translate?

Metabase has no level of detail syntax. A FIXED expression becomes an aggregation computed at its own level and joined back to the detail: a subquery or a window function in SQL, or in the query builder a summarised saved question joined to the table. INCLUDE and EXCLUDE become a second level of aggregation in SQL. Each one is checked against the Tableau figure before the dashboard is signed off.

Does the open-source edition of Metabase have row-level security?

No. Row and column security, connection impersonation and SAML or JWT single sign-on are in the paid Pro and Enterprise editions. Without them, audiences that must see different rows get their own database connection, whose account reads only their rows, and collections decide who sees which dashboards. If you need the paid features, we deploy the paid build under a licence you hold from Metabase and say so in the proposal.

Is Metabase open source?

The open-source edition is AGPL-3.0 and has no per-user fee. Pro and Enterprise add features such as SAML single sign-on, row-level permissions and SCIM under the Metabase Commercial License, which you license from Metabase if you need them.

How long does a Tableau to Metabase migration take?

It depends on how many views people actually open, not on how many workbooks are published. The usage data settles that in the inventory. Pilae quotes the rebuild, the checks against Tableau and the cutover as one fixed price once the inventory is done.

Let us run the move for you

An inventory, a parallel run and a cutover plan, at a fixed price. Then we operate the app.

Start with one workload

  • Go deeperSaaS migrationA fixed-scope migration off Google Workspace, Microsoft 365, Zapier, Airtable and ChatGPT to open-source apps, with a parallel run and a cutover plan.
  • CheckManaged operationsPatching, daily encrypted backups, monitoring and incident response for your open-source apps, under an SLA with an availability commitment of up to 99.9%.