Tableau to Power BI Migration in Manufacturing: Plant Displays, Shifts and OEE
September 8, 2026
On a plant floor, part of the estate behaves like equipment: mounted on a wall display, refreshing without anyone touching it, watched by people who never sign in. Check those dashboards first. Power BI's automatic page refresh carries requirements a Tableau wall display never had, and the login-free terminal turns out to be a licensing question rather than a conversion one.
The rest of a manufacturing migration is the process described on the migration guide. What is specific is where the reports live, how the calculations are built, and how differently each plant built them.
Which manufacturing dashboards are actually devices
| Plant-floor scenario | Power BI equivalent | What to check |
|---|---|---|
| Wall display refreshing itself all shift | Automatic page refresh | It requires a DirectQuery source; Import mode does not support it at all |
| Sub-minute refresh on a line-side screen | Fixed-interval page refresh | Minimum 30 minutes in shared capacity; as low as 1 second on a dedicated capacity, subject to a capacity admin minimum that defaults to 5 minutes, as of August 2026 |
| Rotating between views on one screen | Report pages, with no presentation rotation of their own | Tableau stories have no direct counterpart; rotation is usually handled by the display client |
| Shared terminal with no personal login | An identity and licensing decision, not a report setting | Viewers need licenses unless the content sits on an F64 or larger capacity; embedding is a separate route with its own work |
| Shift supervisor on a tablet on the floor | Power BI mobile with a phone layout | A report designed for a wall screen is not a phone report; layouts are authored separately |
| Threshold alerting on a running metric | Alerts and subscriptions | They evaluate on refresh, so an alert is never more timely than the model behind it |
What plant-floor refresh costs you
Automatic page refresh sends queries continuously, and the load lands twice: on the capacity, and on whatever the historian or warehouse is. Microsoft builds in two safeguards worth knowing before promising a number to operations. Those queries run at a lower priority than interactive ones, and a new refresh query is not sent until the previous one returns, so a one-second interval against a four-second query effectively refreshes every four seconds.
There is also an honest check to run per screen. Many Tableau displays described as real time were rendering an extract refreshed every so often, not a live query. The equivalent of that is a scheduled refresh, capped at 8 per day on Power BI Pro and 48 with Premium Per User or a capacity, as of August 2026. Hold each screen's actual expectation against those two options before conversion, and the awkward conversation on the shop floor never happens.
Where OEE logic breaks
Manufacturing calculations are unusually dense for their visual simplicity. A single tile showing OEE can rest on a chain of ratios, shift boundaries and event ordering that Tableau expressed compactly and DAX will not.
| Logic or source | Usual home in Tableau | What it needs in Power BI |
|---|---|---|
| OEE as availability times performance times quality | Ratios of aggregates inside LOD expressions | Measures evaluated at the right grain over a star schema; a ratio of sums is not a sum of ratios |
| Downtime duration between state changes | Table calculations over ordered rows | Explicit DAX over an event table, with the ordering stated rather than implied |
| Yield and scrap by shift | Shift boundaries buried in calculated fields | A shift dimension joined to the fact table, so every measure inherits the same definition |
| Rollups across plants | Blends across per-plant extracts | One model with real relationships; blend semantics do not survive the move unchanged |
| Historian and OT tag data | Direct connections or per-plant extracts | Read through a warehouse or time-series layer, at a grain a report can query |
| Plants on restricted networks | Server or Bridge inside the plant network | An on-premises data gateway, sized and sited as on the Tableau Server page |
The first three rows are where converted numbers most often disagree with the originals, and the mechanics are on LOD expressions to DAX and table calculations to DAX. Validate OEE at line and shift grain rather than at the plant total, because offsetting errors hide neatly in a daily figure.
Why plant estates diverge and what to do about it
Every site grew its own version of the same report, because every site had its own analyst, its own line layout and its own idea of when a stoppage counts. A migration is the natural moment to decide whether to standardize or to preserve that, and the decision has to be made before conversion. Made afterwards, it becomes a redesign in the middle of a delivery schedule, and the schedule stops being predictable.
A rule that survives contact with plant managers: standardize the model and the metric definitions, preserve local layout. Definitions are what a group rollup depends on, and layout is what an operator recognizes at a glance. Running that decision across sites is coordination work rather than conversion work, which enterprise migration strategy covers.
What Antares does in a manufacturing migration
Antares is a BI migration tool, and in a multi-plant estate the analysis matters before anything is converted. The free Analyzer inventories every site's content in one pass: which reports repeat across plants, where LOD expressions and table calculations concentrate, which sources each plant reaches, and how complex each dashboard is to convert. That is the input to the standardize-or-preserve decision above. It is deterministic, reads metadata only rather than business data, runs inside your own environment, and changes nothing in the source.
The Converter turns each source Tableau dashboard into a Power BI report page for a flat $200, no per-user licensing and no time limit, with LLM costs going to your own provider at typically under $20 per dashboard and private endpoints supported. It is deterministic-first with guardrailed, validated AI steps, and flags calculations it cannot map confidently instead of guessing, which is the behavior you want on OEE logic. It converts the dashboard's semantic model but is not a data migration tool: historians, gateways and time-series layers stay your work. Run the free Analyzer across two plants and calibrate the rest from the difference.
Related reading: retail migrations, LOD expressions to DAX and the Converter. Primary sources: Microsoft on automatic page refresh, DirectQuery and star schema design.
← Back to the Tableau to Power BI Migration guide
Related Migration Resources
Frequently asked questions
Can a Power BI report refresh itself on a plant-floor screen?
Yes, through automatic page refresh, with conditions. The report has to use a DirectQuery source, because Import mode does not support it. In a shared-capacity workspace the minimum interval is 30 minutes; on a dedicated capacity it goes as low as 1 second, subject to a capacity admin minimum that defaults to 5 minutes, as of August 2026.
Do we need a Power BI license for a shared shop-floor terminal?
This is a licensing and identity question rather than a conversion one, and it is worth settling early. Viewers need licenses unless the content sits on a Fabric F64 or larger capacity, where free-licensed users can open shared content. Shared accounts, kiosk devices and embedding each carry their own conditions, so involve whoever owns identity.
Where should historian and OT tag data connect in Power BI?
Through a warehouse or time-series layer, at a grain a report can query, rather than directly from the historian. Most Tableau estates already do this, and the migration inherits it. Where a dashboard reads raw tags directly, treat the rework as data-platform work with its own owners rather than as part of the report conversion.
Will OEE numbers match after conversion?
They should, but that is exactly the calculation to validate deliberately. OEE is a chain of ratios, and a ratio of sums is not a sum of ratios, so grain errors survive at the plant total while showing up per line and per shift. Validate at line and shift grain, where offsetting errors cannot hide, before anyone signs off the summary.