# Tableau to Databricks AI/BI Migration: The Practical Guide for 2026

> What a Tableau dashboard becomes in Databricks AI/BI: a full mapping table, what happens to extracts and access, the cost shape, and where Genie Code fits.

- Canonical: https://getantares.io/tableau-to-databricks-ai-bi-migration/
- Published: 2026-09-07
- Updated: 2026-09-07

---

A Tableau to Databricks AI/BI migration is three layers of work: data logic into SQL datasets and Unity Catalog metric views, dashboards rebuilt on the AI/BI canvas, and access rebuilt in Unity Catalog. Automation covers the first two; the third is a design decision. This is not about [connecting Tableau to Databricks](https://docs.databricks.com/aws/en/partners/bi/tableau) as a data source, which Databricks documents separately.

Names first: search results still carry the old ones. AI/BI dashboards were Lakeview dashboards until Databricks [renamed them Dashboards](https://docs.databricks.com/aws/en/sql/release-notes/2024) on April 18, 2024. The older Databricks SQL dashboards, now called legacy dashboards, [reached end of life](https://docs.databricks.com/aws/en/ai-bi/release-notes/2026) on March 5, 2026, and Genie Spaces became Genie Agents on July 9, 2026 (release notes, as of September 2026).

## What a Tableau dashboard becomes in Databricks AI/BI

One line per concept; rows with real conversion work link to their owner page. Two Databricks words show up before their section: a **dataset** is the query a dashboard reads, and a **metric view** is a measure definition kept in Unity Catalog rather than inside any one dashboard.

| Tableau | Databricks AI/BI | Note |
| --- | --- | --- |
| Workbook (`.twb`, `.twbx`) | Not a unit in AI/BI | Converts one at a time; worksheets ride along |
| Dashboard | AI/BI dashboard | The unit of work and pricing |
| Published data source | Metric view in Unity Catalog | Shared across dashboards |
| Custom SQL | Dataset SQL | Read-only: no `CREATE` or `INSERT` |
| Extract (`.hyper`) | Result cache plus a schedule, or a materialized view | No file exists |
| Live connection | Dataset SQL on a SQL warehouse | Serverless, pro or classic |
| Calculated field | Dataset SQL column, custom calculation, or metric-view measure | [Calculated fields](/tableau-calculated-fields-to-databricks-sql/) |
| LOD expression | Fixed or Coarser LOD, or dataset SQL | INCLUDE takes modeling: [LOD expressions](/tableau-lod-expressions-to-databricks-sql/) |
| Table calculation | `AGGREGATE OVER`, or a SQL window function | Where it runs changes the total: [table calcs](/tableau-table-calculations-to-databricks/) |
| Quick filter, parameter, filter action | Field filter, dashboard parameter, cross-filtering | [Parameters and actions](/tableau-parameters-to-databricks-ai-bi-parameters/) |
| User filter | Unity Catalog row filter, plus the publish mode | The mode decides if it applies: [RLS](/tableau-row-level-security-to-unity-catalog/) |
| Story | No documented equivalent | Dashboards have pages; the docs describe no story-style sequence |

## Why the logic moves before the pictures

Tableau lets a workbook carry its own semantics: an LOD expression can live inside the dashboard that needs it, which is sensible when the dashboard is the product. Databricks AI/BI assumes the opposite. A dashboard reads [datasets](https://docs.databricks.com/aws/en/dashboards/manage/data-modeling/datasets), dataset queries are read-only SQL, and the arithmetic exists before the chart does.

We recommend three layers: base views in Unity Catalog that clean and join source tables, [metric views](https://docs.databricks.com/aws/en/uc-semantics/metric-views/) holding the measures and their dimensions, and dashboard datasets on top. The payoff is that a disputed number has one definition, in SQL, that the whole team can read.

## What happens to extracts and live connections

There is no extract file in Databricks. A Tableau [extract](https://help.tableau.com/current/pro/desktop/en-us/extracting_data.htm) is a subset saved separately in `.hyper` format on its own refresh schedule. That job splits in two here, neither half an extract.

The first half is the dashboard [result cache](https://docs.databricks.com/aws/en/dashboards/caching): 24 hours, best effort, and new data does not invalidate it, so only a schedule refreshes it reliably, the nearest relative of your extract refresh. The second half is [materialized views](https://docs.databricks.com/aws/en/ldp/concepts/materialized-views), precomputed results held in Unity Catalog and refreshed on an interval. Nothing is downloaded, so neither is an extract.

A live connection becomes dataset SQL on a [SQL warehouse](https://docs.databricks.com/aws/en/compute/sql-warehouse/warehouse-types), serverless, pro or classic. Two rendering limits belong in the plan: most charts render up to 15,000 rows and tables up to 100,000 ([dashboard limits](https://docs.databricks.com/aws/en/dashboards/limits), as of September 2026).

## Should you migrate one to one first, or rationalize first

Inventory first: which dashboards were opened last quarter, who owns them, which three are one dashboard with different filters. Then convert like to like whatever survives, keeping redesign separate, per dashboard, after the numbers tie out.

The reason is arithmetic. A dashboard whose total no longer matches has one suspect if you only converted it, two if you redesigned it the same week. The structural budget belongs here too, at 15 pages, 100 datasets and 100 widgets per page as of September 2026, where a workbook of forty worksheets becomes a question about how many dashboards it should be.

## What the cost shape looks like

Tableau bills per seat. Databricks bills the compute under the dashboard, and words it plainly on its [business intelligence page](https://www.databricks.com/product/business-intelligence): "There are no additional license fees to use AI/BI. Standard Databricks Lakehouse $DBU rates apply." Genie is the exception: Databricks paused billing for Genie One and Genie Agents on July 15, 2026 and usage by users is free through January 31, 2027, while Genie Code, which runs `/importBI`, stays pay-as-you-go ([release notes](https://docs.databricks.com/aws/en/ai-bi/release-notes/2026)). The current terms, with the allowances, are on [the comparison page](/databricks-ai-bi-vs-tableau/).

Viewer count therefore stops being a line item and becomes a capacity question. Every dataset query runs on the SQL warehouse, but results served from the 24-hour [dashboard cache](https://docs.databricks.com/aws/en/dashboards/caching) do not start it, and a schedule warms that cache so the first viewer of the day does not pay for it. Publish mode decides how far one query stretches: Share data permissions gives every viewer a shared cache, while Individual data permissions gives each their own, so each person's first load runs a query. Serverless [SQL warehouses](https://docs.databricks.com/aws/en/compute/sql-warehouse/warehouse-types) start in seconds and autoscale with the queue.

## Databricks' own import, and where a migration tool fits

Databricks ships an importer. Genie Code's `/importBI` command [accepts](https://docs.databricks.com/aws/en/dashboards/manage/import-bi) a `.twb`, `.twbx`, `.tds`, `.tdsx` or `.pbit` file and builds an AI/BI dashboard connected to local metric views, generally available since August 20, 2026. Genie Code is metered: excluded from the Genie free promotion, it keeps a pay-as-you-go allowance of 150 free DBUs per user a month ([release notes](https://docs.databricks.com/aws/en/ai-bi/release-notes/2026), as of September 2026). The docs carry no limitations section and no list of which Tableau constructs convert, so a person checking finds whatever a workbook lost, which is the work a tool built for estates is shaped around.

## What Antares does on this route

Everything above holds whichever tool you use. This section does not. Antares is a BI migration tool, [listed on the Databricks Marketplace](https://marketplace.databricks.com/details/90d24b77-11c2-4a72-be14-0a1b3026fae9/Tier-One-Analytics_Antares-BI-Migration-Automation), and Tableau to Databricks AI/BI is shipped. The [**Analyzer**](/analyzer/) is free, deterministic and metadata-only: it reads the Tableau environment, not the business data in it, and returns the inventory sequencing depends on.

The **Converter** is deterministic-first with guardrailed, validated AI steps, at a flat $200 per source dashboard. LLM costs sit on top, typically under $20 per dashboard, billed by your own provider, private endpoints included. No per-user licensing, no time limits, and it runs as SaaS or inside your environment.

One Tableau dashboard becomes one Databricks AI/BI dashboard, worksheets riding along. It takes calculated fields to SQL and metric-view measures, LOD expressions to windowed aggregates, custom SQL to native Databricks SQL, and row-level security to Unity Catalog row filters. It is not a data migration tool: it converts dashboards and the semantic logic inside them, not data into Delta tables, the lakehouse, or Unity Catalog itself. [Run the free Analyzer](https://try.getantares.io) to see what the estate holds.

## If your source is Power BI

The shape is the same, the vocabulary shifts. A report page becomes an AI/BI dashboard, DAX measures become SQL and metric-view measures, RLS roles become Unity Catalog rules read through the publish mode, and Databricks' importer takes `.pbit` files. The Antares Power BI to AI/BI route is on the roadmap this year, not available today; the [Converter page](/converter/) takes signups.

The rest of the cluster: [LOD expressions](/tableau-lod-expressions-to-databricks-sql/), [calculated fields](/tableau-calculated-fields-to-databricks-sql/), [table calculations](/tableau-table-calculations-to-databricks/), [parameters and actions](/tableau-parameters-to-databricks-ai-bi-parameters/), [row-level security](/tableau-row-level-security-to-unity-catalog/), and [AI/BI next to Tableau](/databricks-ai-bi-vs-tableau/).

Primary sources for this page: Databricks on [datasets](https://docs.databricks.com/aws/en/dashboards/manage/data-modeling/datasets), [metric views](https://docs.databricks.com/aws/en/uc-semantics/metric-views/), [caching](https://docs.databricks.com/aws/en/dashboards/caching), [limits](https://docs.databricks.com/aws/en/dashboards/limits) and [importing BI files](https://docs.databricks.com/aws/en/dashboards/manage/import-bi), plus Tableau on [extracts](https://help.tableau.com/current/pro/desktop/en-us/extracting_data.htm).

## Related resources

- [Databricks AI/BI vs Tableau](/databricks-ai-bi-vs-tableau/)
- [Tableau LOD Expressions in Databricks SQL](/tableau-lod-expressions-to-databricks-sql/)
- [Tableau Calculated Fields to Databricks SQL](/tableau-calculated-fields-to-databricks-sql/)
- [Tableau Table Calculations in Databricks AI/BI](/tableau-table-calculations-to-databricks/)
- [Tableau Parameters in Databricks AI/BI](/tableau-parameters-to-databricks-ai-bi-parameters/)
- [Tableau Row-Level Security to Unity Catalog](/tableau-row-level-security-to-unity-catalog/)

## FAQ

### How long does a Tableau to Databricks AI/BI migration take?

No per-estate number would survive contact with your estate, so here is the shape instead. Inventory takes days once you can read usage and ownership. Conversion is per dashboard and predictable. Validation dominates the calendar, because every converted number has to tie out at two grains and two filter states before anyone signs it off.

### Should we migrate one to one first or rationalize first?

Inventory first: which dashboards were opened last quarter, who owns them, which three are the same dashboard with different filters. Convert like to like whatever survives that cut, then redesign separately, per dashboard, once the numbers tie out. A total that stops matching then has one suspect instead of two.

### Do we have to move our data into Databricks first?

The tables the dashboards read have to be in Databricks and registered in Unity Catalog, because AI/BI datasets query them there. That is a data platform project and it comes first. Antares converts dashboards and their semantic logic; it does not move data into Delta tables or build the lakehouse.

### Does Databricks AI/BI require Unity Catalog?

Databricks states on its business intelligence page, as of September 2026, that AI/BI "requires your data to be managed in Unity Catalog." Unity Catalog is also where row filters and column masks live, so the access rules that used to be Tableau user filters get rebuilt there rather than inside the dashboard.

### What happens to Tableau extracts?

Nothing is downloaded. The extract's job splits between the dashboard result cache, which holds results up to 24 hours on a best-effort basis and is refreshed reliably only by a schedule, and materialized views in Unity Catalog for aggregations that are expensive to recompute. Materialized views are server side, so calling them extracts misleads people.

### Is Power BI to Databricks AI/BI available from Antares?

Not yet. Tableau to Databricks AI/BI is shipped; the Power BI route to AI/BI is on the roadmap for this year, and the Converter page carries a notification signup for when it ships. Databricks' own Genie Code importer already accepts Power BI template files, which is the native option meanwhile.