Power BI Templates: What a .pbit Asks of Your Data

A Power BI template carries no rows of data. That is not a complaint about quality, it is the file format working as designed, and Microsoft says so in its own documentation. What a .pbit carries instead is a description of data you do not have yet: the tables it expects, the exact column names it refers to, the measures written against those names, and the queries it will run once you tell it where to look. Download is the easy part. Satisfying that description is the whole job.

Two disclosures first, because they change how you should read the rest. I build clinic reporting for a living and I sell a Power BI file myself, so I have an obvious interest in you thinking such things are worth money. That is why the last section turns the same test on my own product, and why it does not come out clean. And every number below comes from a file I unzipped on 29 July 2026, not from a product description.

What Microsoft actually puts in the file

Start with the vendor, because on this point the vendor is unusually clear. Microsoft's documentation for report templates lists three things a .pbit carries over from the report it was made from: the report pages, visuals and other visual elements; the data model definition, including the schema, relationships, measures and other model items; and all query definitions, including queries and query parameters.

Then comes the sentence that decides everything else. Template files are smaller than a normal report, the documentation notes, because "templates don't contain any data, just the report definitions themselves."

That sentence is quoted everywhere, including by me until a fact-check caught it. It is worth reading the two sentences Microsoft puts immediately after it, in the same note, because they qualify it: "Report metadata can be saved with data values from your semantic model. For example, if you apply a filter to a visual for the field 'Company' = 'Contoso', the value 'Contoso' is persisted as part of the metadata." Slicer selections persist too. So a template holds no rows, but it is not literally empty, and a template exported from a real model can carry values from that model in its metadata. If you publish one, that is worth a look before you do.

Microsoft also describes what happens when you open a template: a dialog may ask for values for any parameters the original report defined, and then, in its words, "you're prompted for the location of the underlying data associated with the report. The current report creator can then connect to data based on their credentials." Two gates, both of them yours to pass, before anything renders.

None of that is hidden. It is simply not what the pages offering you templates talk about.

What page one offers instead

Here is my method, including the parts that work against me. One automated run, one machine, US result set, no location set, on 29 July 2026. I rendered each result page and searched its rendered text, which means anything stated only in an image, in a video, or one click deeper is invisible to me. Another run on another day would give a different page one. Treat this as one sample, not a survey.

Two bottom-of-page ads sat under a "Sponsored results" heading and I dropped them, which left eight organic results. Reddit's bot protection blocked my fetch, so I could read seven of the eight.

Of those seven, exactly one states that a template holds no data, and it is Microsoft's own documentation. Two use the string ".pbit" at all: Microsoft, and one vendor help centre explaining setup to its existing customers.

My first draft said the remaining five are "the ones actually offering you templates" and that between them the file extension never appears. Both halves were too strong, and the corrections are worth having. One of the five is a 2020 forum thread where a user asks where to find dashboards, so it offers nothing. Two of the others do name a file extension, just not the one I was looking for: they ship .pbix files, which is a different product, and my rendered-text search missed the string because it sits in the page source, not the visible copy. Faulting a page for not writing ".pbit" when it never claimed to sell one is a category error, and I made it.

What survives is narrower and still worth saying. Of the pages that hand you a file, none tells you that a template arrives empty or what it will demand from your data. Two do define the word: one carries a heading reading "What is a Power BI template?" and answers that it is "an interactive visualization tool that transforms complex data into clear, actionable insights". That is a definition of a dashboard, not of a data contract. The gap is not that nobody explains templates. It is that the explanations describe the picture and skip the bill.

I unzipped seven of them

A .pbit is a zip archive, which means you can read one without installing anything. I collected seven free templates that download without a signup, mostly from public repositories, and opened each one to read the model definition inside.

Three hundred and eighty-nine column names, of which 245 have to come from your data. The other 144 the model computes for itself, in calculated columns and calculated tables, and it is worth separating the two: the computed ones are the author's work, and you inherit them for free. The 245 are the bill.

Every measure and every authored relationship is written against those names. If your export calls the field visit_date and the template says Date, the measure that references it does not adapt, it breaks, and it takes the visuals built on it with it. This is the part a screenshot cannot show you and a feature list does not mention.

Almost nothing sits behind those names. Three of the seven ship a handful of rows anyway, baked into the file: FlowViz carries two small lookup tables written out as literal values, one listing age buckets and one listing two sort options, plus two generated series it builds on load. So "a template contains no data" is Microsoft's phrasing and it is right about rows from your business; it is not right that the file is empty.

Notice the last row. That Jira template defines no tables and no measures at all. Its entire model is five parameters and a set of queries that go and get everything at runtime. Downloaded, opened and left alone, it is an empty report with a login form.

Five of the seven stop before they fetch anything

Power BI marks a parameter as required in the file itself, so most of this is countable rather than a matter of opinion. Four of the seven carry required parameters, and they are not cosmetic:

• Jira issues wants a Jira domain, an API path, a filter ID, a field list and a row limit. Five values before the first request goes out.
• FlowViz wants a protocol, an Analytics URL, an organisation and a project, all four pointing into Azure DevOps.
• Story Points, Azure DevOps wants the same four.
• Story Points, Jira wants a Jira URL, a project key, and the name of the custom field your team stores story points in.

A fifth belongs here too, and it is the reason the heading says five. The WordPress template has no parameter marked required, yet its queries reference three values, including the address of your site, and it fetches nothing without them. The flag undercounts, so counting only flagged parameters would have let that file pass as ready to run.

Read the list again and the real requirement becomes obvious. These are not templates for a dashboard, they are templates for a specific service you must already be paying for. FlowViz is 23 tables and 151 measures of careful modelling, and if you do not run Azure DevOps, none of it is reachable. The template is well built and unusable at the same time, depending entirely on facts about you.

Two of them were built from a file on the author's own computer

The two templates in the table with no parameters at all are not the easy ones. They are the ones that cannot refresh.

Both were built from a file sitting on the author's own machine, and the path was saved into the template along with everything else. One expects a CSV pair in a Downloads subfolder named after a Google Drive export; the other expects a workbook at C:\Users\<name>\Downloads\archive (1)\Superstore.xlsx. Both are free hobby projects published by individuals who owe a stranger nothing, and I have shortened the one path that carried a personal username. The repositories are linked in the sources if you want to check.

I want to be careful about how far this generalises. My sample leans on public repositories because those download without a signup, and that over-represents personal projects over commercial ones. I did try five templates from one commercial connector vendor: all five download links returned PublicAccessNotPermitted from their storage account, which is most likely a storage setting rather than a withdrawn product, and it means I could not examine them and will not guess at what is inside. That vendor is not among the eight results above. Two out of seven is a finding about these seven files, not a rate for the category.

Most of the tables were not written by the author

One number in that table needed correcting before I could publish it, and the correction is worth passing on.

My first count of FlowViz said 49 tables. Twenty-six of those turned out to be hidden tables Power BI generated by itself. Microsoft's documentation explains why: when the auto date/time option is on, Power BI Desktop "creates a hidden auto date/time table for each date column" that meets three conditions, namely that the table is in Import storage mode, that the column is a date or date/time, and that it is not on the many side of a relationship. Each one is a calculated table with six calculated columns plus the date. For a new installation, the documentation states, the option defaults to on.

Across the seven files that came to 39 hidden tables carrying 276 columns: seven columns each in 36 of them, and eight in the three from the older WordPress model, which also carry an internal row-number column. I checked one against the documentation and found exactly the seven columns described, named as promised. Those hidden tables also bring their own relationships, which is why the table above counts 42 authored links and not the 76 the file reports.

This matters for a practical reason. Those tables are permanently hidden, they cannot be referenced directly in DAX, and they refresh every time the model does. If a template feels heavier than its report warrants, this is usually where the weight is.

How to read a .pbit before you install it

You can do most of the above yourself, before installing anything, on any operating system, without opening Power BI. The file is a zip archive.

Step 5 is not optional, and I learned it the hard way: one of my seven keeps its entire query layer in DataMashup and nothing in DataModelSchema. That is the very file whose three unflagged values I reported above, and my first version of this recipe could not have found them. DataModelSchema gives you the table names, every column, every measure and every relationship. The queries may be somewhere else.

Two more things worth knowing. The JSON is saved as UTF-16 with no byte-order mark, so a plain text editor may show it with gaps between characters; that is an encoding quirk, not corruption. And while a .pbix unzips just as happily, step 3 fails there, because the model is stored as a compressed binary instead of readable JSON. The template format is the one you can audit, and the pages handing you a template are not the ones pointing that out.

The other things sold under this name

Not every result for this search is a .pbit, and the differences are about cost and access more than convenience.

A .pbix is the ordinary report file, and unlike a template it does contain data, which is why sample galleries ship them. It also cannot be read the way described above.

A template app is a packaged app a partner publishes through AppSource, and it is installed in the Power BI service rather than opened on your desktop. Microsoft's overview describes them as letting "Power BI Pro or Power BI Premium users" get insight from prepackaged dashboards and reports, which frames them as paid-tier content; Desktop, where a .pbit opens, is free. If you are weighing that up, I went through the licensing in detail in Power BI Pro cost: do clinics actually need it?.

A report theme is a JSON file of colours, fonts and visual styling. Microsoft describes it as applying design changes to a report, and from reading the format there is nothing in it that carries data, queries or measures. It will not put a single number on a page.

What to do with all this before you download

The useful question about a template is not whether it is free. It is how much of the contract you can already satisfy.

• If it names a service (Jira, Azure DevOps, Salesforce, your EMR), the only question that matters is whether you are a customer of that service with API access. Everything else about the template is irrelevant until that is true.
• If it names a file, open the model definition and find out whose file. A path under someone's user folder means the template needs rewiring, not configuring.
• If it names neither, you are buying a layout, and you should price it as a layout.
• Count the columns you have to supply, not the ones you see. Twenty-nine columns from four small files is a different afternoon from 116 across 23 tables, and both of those are real numbers from the table above.
• Read the measures. They will not run against your data unless the names line up, so this is not a shortcut. It is the part worth taking: 151 measures show you how someone else solved the problem, and that thinking survives even when the file does not.

Which brings me to the awkward part, and a fact-check made it more awkward than I had written it. What I sell is a Power BI file, and my first draft told you it "has exactly the property this article has been describing" and that I "cannot design that away, because it is what the file format is". Both claims were wrong about my own product. Clinic Vitals ships as a .pbix, not a .pbit, which means two things I should say plainly: it opens with a year of demo data already in it, and you cannot run the five-minute audit above on it, because that is exactly the format the audit does not work on. I am recommending a transparency test that my own file fails.

Three more things it does that this article has criticised in others. It asks you to set a folder path as a parameter before it fetches anything, which is the same dialog I just spent a section on, with the one difference that the folder is yours. Every chart in it is drawn by a free third-party visual you install from AppSource first, so a tenant that blocks those visuals gets blank rectangles. And six of its widgets ship with placeholder numbers, including average wait time and the middle stages of the funnel, because those signals are not in the three exports it asks for; they are placeholders until you supply the data behind them. What I can defend is the contract itself, and after writing this article I moved it out of the download and onto a public page: every column Clinic Vitals expects is now readable before you buy, required flags and all, along with that list of six placeholders and the two non-data things the file asks of you. Three flat exports rather than a live connection to one vendor's API, and a contract you can check against your own export in about five minutes. That is a difference in documentation, not in what the file format can do.

Frequently asked questions

If you want to see what a finished report looks like once the data is in place, Power BI dashboard examples covers what the screenshots leave out, and Power BI for clinics walks through turning three EMR exports into a working model. For the wider question of what the word "template" means outside Power BI, see dashboard template: what you actually get.

Written by

Olha · clinic data analyst

I build the reporting our managers open every morning at a multi-branch medical clinic — and package it so other practices don't have to start from scratch.

This article was published on 29 July 2026 and corrected the same day after three independent fact-checks, which found more than the usual number of errors and all of them mine. The measurement counts were wrong: two required parameters were being counted as tables in one file, so the totals moved from 46 tables and 391 columns to 44 and 389, and one template's parameter count moved from two to four. The relationships column silently included links to the hidden date tables the article elsewhere boasts about excluding, so 76 became 42. The claim that 391 columns "must come from your data" was wrong in my favour, since 144 of them are computed by the model; the honest figure is 245. The heading said four of seven templates stop at a dialog when the article's own next paragraph showed it is five, and the wrong number had reached the FAQ and the structured data. Two absolutes failed: "not one row of data" (three files bake in small lookup tables) and a claim that the pages ranking for this search never explain what a template is (one of them carries a heading doing exactly that, and two name the .pbix extension my rendered-text search could not see). The published recipe for auditing a template could not reproduce my own WordPress finding, because that file keeps its queries in a part the recipe never mentioned. I also quoted Microsoft's "don't contain any data" line without the two sentences that qualify it, and its auto date/time line without the three conditions attached to it. Finally, the section about my own product was wrong about my own product: Clinic Vitals is a .pbix, not a .pbit, so the file I sell cannot be audited by the method I recommend, and three further limitations it shares with the templates criticised here are now named in the text. One claim I could not test at all is what commercial connector vendors ship, because all five download links I tried returned a permissions error. Lucid Vitals is not affiliated with any product named here.