Power BI for Practitioners

Two things decide whether Power BI will work for you before you build anything: the authoring tool is free, and it runs only on Windows. Both matter more for a practitioner in Dhaka or Bhopal than for someone with an IT-issued corporate laptop. This module gets you installed and oriented, and sets out the trade-offs.

By the end you can

• Install Power BI Desktop for free and explain what it does that the cloud and mobile apps do not
• Pick a realistic route to author on a Mac or Linux machine, with the costs named
• Navigate the five regions of the interface and the six ribbon tabs
• Describe exactly what a .pbix file bundles, and why that makes your analysis portable

Three things people call “Power BI”

The single word covers three separate products. Confusing them is the first thing that trips people up, especially around what costs money.

Three things people call 'Power BI'

Power BI Desktop

The authoring application. You connect to data, clean it, build the model, write calculations, and design reports here. Free, Windows-only, and where Modules 1 to 7 live.

Power BI Service

The cloud at app.powerbi.com. You publish finished reports here to share them. Sharing is where the paid licence starts. Covered in Module 8.

Power BI Mobile

Apps for viewing published reports on a phone or tablet. A consumption surface, not an authoring one.

Free-tool reality

Power BI Desktop is free to download and use, with no time limit and no feature lock on the modelling engine. You can connect to over a hundred sources, build a full star schema, write any DAX, design unlimited report pages, and export to PDF or PowerPoint without paying. The cost appears only when you publish to the cloud to share live, which needs a Pro licence at 14 US dollars per user per month as of 2026 (Microsoft Power BI pricing). Everything taught in Modules 1 to 7 costs nothing.

The Windows question, answered straight

Power BI Desktop is a Windows-only application. Microsoft has not shipped a macOS or Linux build and has signalled no near-term plan to. If you are on a Mac or Linux machine you have three realistic routes, in rough order of friction.

Coach callout · Varna

Sort out the Mac question before you train a whole team, not after. For mixed teams the usual answer is a few people authoring on Windows or in a VM, and everyone else viewing in the browser. I have watched a rollout stall because half the team could not open the tool and no one had planned for it. If you are buying a machine mainly for data work, a Windows laptop avoids the problem.

Install it

Hands-on lab · Get Power BI Desktop running

On Windows this is about ten minutes. Do it now if you can; the rest of the course assumes you can follow along.

1. Go to powerbi.microsoft.com/desktop, or search Power BI Desktop in the Microsoft Store. The Store version self-updates, which is the easier path.
2. Run the installer. It needs 64-bit Windows 10 or 11. There is no 32-bit build any more.
3. Open it and sign in. A work or school email unlocks publishing later; a personal address still lets you build and save locally. No paid account is needed to start.
4. Close the splash screen. You are looking at a blank report canvas. This is where everything happens.

The five regions of the screen

The interface looks busy. It is five regions, and you use all of them in every project.

The five regions of the screen

Power BI Desktop — Report viewinterface rebuilt for teaching

HomeInsertModelingViewOptimizeHelpGet dataRefreshNew visual

Blank report canvas

Data

NFHS_district State Indicator Value

Visualizations

The three small icons on the far left switch Report, Data and Model views. Fields pane lists every loaded table and column; Visualizations pane sits beside it; the ribbon runs along the top.

The six ribbon tabs, briefly, so the names stop being mysterious:

• Home holds Get Data and Refresh, the two buttons you press most, plus Transform Data and Publish.
• Insert adds visuals, text boxes, shapes and buttons to a page.
• Modeling is where you write measures and calculated columns, manage relationships, and create roles for security.
• View controls page layout, gridlines, themes and the mobile layout.
• Optimize helps when a slow report needs tuning. You will ignore it for a while.
• Help is the documentation and the community.

Your first five minutes, narrated

Open Desktop. Click Home > Get Data > Text/CSV and pick any small spreadsheet you have. A preview window appears. Click Transform Data (not Load) and the Power Query Editor opens; you will live here in Module 3. For now, click Close & Apply on the Home ribbon. The data loads, and its table appears in the Fields pane on the right. Click the Data view icon on the left to see the actual rows, then the Report view icon to return. Drag a text column onto the blank canvas and Power BI guesses a visual. That loop is the whole workflow: get data, clean, load, look, build.

What a .pbix file holds

When you save, you get one file with a .pbix extension. It can bundle three things together: the queries that fetch and clean your data, the data model with its calculations, and the report pages with their visuals.

What one .pbix carries

That portability is the point. You can email a .pbix and the recipient has your entire analysis, queries and all, provided they also run Windows. A small NFHS extract might produce a file of a few megabytes; a large model with imported data can run to hundreds.

Coach callout · Varna

Treat the .pbix as a working file, not an archive. The spreadsheet habit is one giant file that does everything; don't carry it over. One file per question, or per dashboard, keeps refreshes fast and makes it clear to whoever comes next what each file is for. A 60 MB file with eleven hidden report pages is just inherited confusion. Name files for the question they answer, not the date you last touched them.

Where module 1 goes wrong

Assuming a Mac can run Desktop natively

It cannot. Plan the VM or browser route before committing a team, or you train people who cannot open the tool.

Signing up for a paid trial to ‘unlock’ building

Building is already free. The trial only matters for publishing and sharing, which is Module 8.

One enormous .pbix for everything

Split by question. Large all-purpose files refresh slowly and become unmaintainable.

Try it yourself

Plan a small rollout. Your team is six people: two on Windows, four on Macs. Three need to build dashboards, three only need to view them. Sketch the cheapest workable setup.

Show a worked answer

The three builders need full authoring. Put the two Windows users on native Desktop. One Mac builder runs Windows in a VM (Parallels on Apple Silicon, or VMware Fusion free). The three view-only people use the Service in a browser, which is free to view shared content, though sharing itself will need at least one Pro licence (Module 8). Net new cost so far: one VM licence, plus Pro seats decided later by how you share. Nobody is blocked from starting today.

Check yourself · 5 questions

A colleague on a MacBook cannot install Power BI Desktop. The most accurate response is:

A. She must buy a Windows laptop, there is no other wayB. Desktop is Windows-only, but she can author in a Windows VM or view reports in any browser via the ServiceC. It runs natively on macOS once updatedD. Power BI does not work on Macs in any form

B is correct. No Mac build of Desktop exists. The Service runs in any browser for viewing; a VM gives full authoring on Apple Silicon.

Which component do you use to clean data and build a report?

A. Power BI ServiceB. Power BI MobileC. Power BI DesktopD. Power BI Report Server

C is correct. Desktop authors. Service shares in the cloud, Mobile views.

What does the free Desktop licence let you do at no cost?

A. Publish and share with the whole teamB. Build models, write any DAX, design reports, and export to PDF or PowerPoint locallyC. Schedule eight cloud refreshes a dayD. Give every colleague a viewer account

B is correct. Building and local export are free. Publishing, sharing and scheduled cloud refresh need Pro or higher.

A .pbix can bundle which of these together?

A. Only report pagesB. Only the dataC. The queries, the model with calculations, and the report pagesD. A cloud link and nothing else

C is correct. One file holds queries, model and report, which is why it is portable.

You want to inspect the actual rows of a loaded table, not a chart. Which view?

A. Report viewB. Data viewC. Model viewD. Service view

B is correct. Data view (the middle left-edge icon) shows the rows. Report is for visuals, Model for relationships.

Reflection prompts

1. What machines does your team actually use? If a third are Macs, what is your honest plan for them?
2. Name two reports you re-paste by hand in Excel every month. Those are your first Power BI candidates.
3. How many people need to view a finished dashboard? That number drives the licence cost in Module 8.

Resources

Power BI Desktop downloadOfficial Microsoft download for the free authoring tool.Microsoft Learn: Get startedFree official learning path, readable without an account.Power BI pricingThe current Pro and capacity pricing, so you can check the figures yourself.

Getting the data in is the first step, and a few small choices here save hours later. This module covers how raw data comes into Power BI through Get Data. We use a National Family Health Survey extract, because survey data is what most South Asian practitioners work with.

By the end you can

• Connect to Excel, CSV and web sources through Get Data
• Set the encoding and delimiter so Indian-language data reads correctly
• Choose Transform Data over Load, and say why in terms of reproducibility
• Pull an NFHS-5 indicator table in and recognise what Power BI guessed about it

Get Data is one button, then a Navigator

Everything starts at Home > Get Data. Power BI ships with well over a hundred connectors; you will use a handful. After you pick a source you meet the Navigator, which previews what is inside before you commit.

Power BI Desktop — Navigatorinterface rebuilt for teaching

Get DataNavigatorSearchRefresh

NFHS5_extract.xlsx

Sheet1 (raw)

tbl_districts ✓

Notes

DistrictIndicatorValue

ArariaStunting42.1

PatnaStunting31.4

The Navigator after choosing an Excel workbook. Prefer a named table (the grid icon) over a raw sheet, which drags in stray title rows and notes.

Two paths from every source: Load and Transform Data

Load versus Transform Data

After you point at a file you get two buttons. Load drops the data into the model as-is. Transform Data opens the Power Query Editor first, so you clean before loading. The instinct is to Load and get to the charts. Almost always, click Transform Data. Survey exports are never clean enough to use raw, and the cleaning steps you record become a repeatable recipe. This is the reproducibility principle that any empirical researcher already knows: the value is not only the cleaned dataset, it is the documented, re-runnable path from raw to clean.

Coach callout · Vandana Soni

Clean in Transform Data before you load, not after. I have watched people lose an afternoon because they clicked Load, built half a dashboard, then found a column had come in as text and every total was wrong. Going back means unpicking finished visuals. Power Query is also where you write down how the data was prepared, so that six months later you, an auditor, or a sceptical funder can see exactly what was done to the numbers. Treat it as your methods section.

CSV: the encoding trap

The CSV encoding trap

Comma-separated files are the common currency of open-data portals like data.gov.in. Two settings in the preview decide whether the file reads correctly.

• File Origin / encoding. Set this to 65001: Unicode (UTF-8). If district names in Hindi, Bangla or Tamil arrive as garbled boxes, a wrong encoding is almost always why.
• Delimiter. Usually a comma, but Indian government files sometimes use tabs or semicolons. The preview shows instantly if columns are not splitting.

Excel and Web

For a workbook, prefer a named table (small grid icon in the Navigator) over a worksheet. A sheet drags in title rows, merged cells and blank columns you then strip out. For the web, Get Data > Web takes a URL and finds the tables on a page or behind an API. The World Bank exposes indicators through a documented API; you can also point at a CSV on a public portal so a refresh re-pulls the current file instead of you downloading it by hand each quarter. When a source supports it, Power Query pushes filters back to the source (query folding), so you pull only the rows you need rather than everything.

Loading an NFHS-5 extract, narrated

Prepare a CSV with columns State, District, Indicator, Value, Round. Home > Get Data > Text/CSV, choose the file. In the preview, set File Origin to 65001: Unicode (UTF-8) and confirm the comma is splitting columns. Click Transform Data. The Power Query Editor opens with an Applied Steps list already holding Source and Changed Type: Power BI has guessed your types. Those guesses are exactly what Module 3 teaches you to check, because a Value column guessed as text will silently refuse to sum.

Real versus sample data

On the data. Lab tables in this course are illustrative samples for teaching the mechanics; you swap in real figures. For reference, genuine NFHS-5 national values (IIPS and ICF, 2019-21) include stunting among under-fives at about 35.5 percent, anaemia among women aged 15 to 49 at about 57 percent, and institutional births at 88.6 percent. Verify any district figure against the official factsheet before it reaches a report.

Coach callout · Vandana Soni

The moment a file contains anything that identifies a person, a name, a phone number, a household location, you have left the world of harmless aggregates. The door is the right place to decide this, not after it is buried in the model. If district aggregates answer your question, do not load the individual records at all. Module 8 covers the law; the habit starts here.

Where module 2 goes wrong

Clicking Load to save time

You skip the recorded recipe and often discover a type error after building visuals. Transform first.

Leaving encoding on the default

Non-Latin scripts garble. Set UTF-8 (65001) at the preview.

Connecting to a busy worksheet instead of a named table

You inherit title rows and blank columns. Pick the table object.

Loading personal records when aggregates would do

Minimise at the door. What you never load cannot leak.

Try it yourself

Spot the broken load. A colleague's NFHS dashboard shows every district's stunting rate as blank, and the state names display as square boxes. Two settings were wrong at import. Name them and the fix.

Show a worked answer

First, the Value column was loaded as text, not a decimal number, so it will not aggregate and the measures return blank. Fix it in Power Query by setting the column type to Decimal Number (Module 3). Second, the file was read in the wrong encoding, garbling the Devanagari or Bangla names; re-import with File Origin set to UTF-8 (65001). Both are door-stage decisions, which is why this module insists on Transform Data before Load.

Check yourself · 5 questions

Load and Transform Data both appear after selecting a CSV. Which is the better default and why?

A. Load, it is faster and you fix issues in the charts laterB. Transform Data, cleaning first records a repeatable recipe and avoids rebuilding visualsC. Either, they are identicalD. Load, Transform deletes data

B is correct. Transform Data opens Power Query so you clean first; the recorded steps re-run on next year's file, and you avoid discovering a type error after building visuals.

Indian-language district names show as garbled boxes. The likely fix is to set:

A. The delimiter to semicolonB. The file encoding to UTF-8 (65001)C. The decimal separatorD. The data type to text

B is correct. Garbled non-Latin script is almost always an encoding mismatch. UTF-8 reads Devanagari, Bangla and Tamil correctly.

Why prefer a named table over a worksheet when connecting to Excel?

A. Named tables are encryptedB. A worksheet drags in title rows, merged cells and blanks you must stripC. Worksheets cannot loadD. Named tables refresh faster on the free tier

B is correct. A named table is a clean rectangular object; a sheet often carries notes and gaps that become cleanup.

The advantage of Get Data > Web pointing at a public CSV URL is:

A. It is the only way to read CSVsB. A refresh re-pulls the current file automaticallyC. Web data needs no cleaningD. It bypasses the Pro licence for sharing

B is correct. A web source can refresh, so the latest published file flows in without a manual download each cycle.

Where do the cleaning steps you apply before loading get recorded?

A. In page notesB. In a separate Excel logC. In the Applied Steps list in Power Query, re-run on every refreshD. Nowhere, they are one-time

C is correct. Power Query keeps an ordered Applied Steps list: both an audit trail and a recipe that re-runs automatically.

Reflection prompts

1. Which routine source could be a Web connection that refreshes itself rather than a monthly manual download?
2. Open one survey export. How many of its columns would a dashboard actually use? The gap is your cleaning workload.
3. Does any file you plan to load contain personal data? Write down now where it will and will not be allowed to go.

Resources

NFHS-5 district factsheetsOfficial factsheets for India, states and 707 districts.data.gov.inIndia's open government data portal, a large source of loadable CSVs.Microsoft Learn: connect to dataOfficial guide to the Get Data connectors.

Power Query is where most of the work in a Power BI project happens: cleaning the data. It records your cleaning as a sequence of steps, so a messy survey export becomes a tidy table and every change you made is written down and re-runnable. If you care how a number was produced, this is the part to get right.

By the end you can

• Read the Applied Steps list as an audit trail you can edit and reorder
• Apply the core transforms: promote headers, set types, filter, trim, replace, split
• Unpivot a wide survey factsheet into the long shape charts expect
• Merge tables by a key and append rounds into one table, and tell the two apart

The Applied Steps list is an audit trail

Applied Steps as an audit trail

When you click Transform Data, the Power Query Editor opens. On the right is the Applied Steps panel. Every action adds a line; click any step to see the table as it stood at that point, edit it, reorder, or delete. Underneath, each step is a line of the M language, visible in the Advanced Editor. For an empirical researcher this is a reproducible data-preparation script that you build by clicking and that documents itself.

Power BI Desktop — Power Query Editorinterface rebuilt for teaching

HomeTransformAdd ColumnViewFilterReplaceUnpivot

A¹² DistrictA¹² Indicator1.2 Value

ArariaStunting42.1

ArariaAnaemia61.2

PatnaStunting31.4

Query Settings

Applied Steps

Source

Promoted Headers

Changed Type

Unpivoted Columns

The Power Query Editor: data preview on the left, the editable Applied Steps audit trail on the right. Click any step to time-travel to that point in the cleaning.

The cleaning moves you use weekly

Unpivot: the move that fixes survey data

Unpivot: wide to long

This is the single most useful transformation for development work, and the one people miss. Survey factsheets come wide: one row per district, a separate column for every indicator. That layout is unusable for charting because Power BI wants one row per observation. The principle has a name: Hadley Wickham’s tidy data (Journal of Statistical Software, 59(10), 2014), where each variable is a column and each observation a row. Unpivot turns wide into tidy.

Unpivot, and the M it records

Select the district column, right-click, Unpivot Other Columns. The indicator names collapse into a single Attribute column (rename it Indicator) and the numbers into a single Value column. Now one slicer can switch the whole dashboard between indicators. Underneath, Power Query has written a single line of M, Table.UnpivotOtherColumns, that re-runs on every refresh:

// what Power Query records when you unpivot
let
  Source = Csv.Document(File.Contents("aser.csv"), [Encoding=65001]),
  Headers = Table.PromoteHeaders(Source),
  Typed = Table.TransformColumnTypes(Headers, {{"District", type text}}),
  Long = Table.UnpivotOtherColumns(Typed, {{"District"}}, "Indicator", "Value")
in
  Long

The recorded query. You did not write this; clicking generated it. But you can read and edit it, which is the reproducibility payoff.

Coach callout · Varna

Use Unpivot Other Columns, not Unpivot Columns. The difference bites when next year's file adds an indicator. If you unpivoted the specific columns you saw this year, the new one is silently ignored. If you unpivoted everything except the district key, the new indicator flows in automatically. Build for the data you will get, not only the data in front of you. It is the same discipline as not hard-coding the number of columns in a script.

Merge and Append: joining tables

Merge versus Append

Two operations combine tables, and people confuse them.

• Merge is a join. It brings columns from one query into another by matching a key, the way you add a region label to each district by matching district codes. Power Query supports the join types you already know from relational data: left outer, inner, full outer, and the anti-joins for finding non-matches.
• Append is a union. It stacks rows. Use it to combine the NFHS-4 file and the NFHS-5 file into one long table with a Round column, so you can chart change across rounds.

Hands-on lab · Clean a messy ASER-style education extract

ASER, the Annual Status of Education Report run by Pratham, produces exactly this kind of real, slightly messy data: a title row, mixed types, a wide layout. Use an ASER-style sample to practise the full sequence.

1. Home > Remove Rows > Remove Top Rows to drop any title rows above the real header.
2. Home > Use First Row as Headers to promote the header.
3. Click each column's type icon: Value columns to Decimal Number, District to Text. Watch the Changed Type step appear.
4. Select the District column, right-click > Unpivot Other Columns. Rename Attribute to Indicator.
5. Transform > Format > Trim, then Clean, on District and Indicator to kill stray spaces.
6. Home > Replace Values to turn NA into null so it counts as missing, not text.
7. Rename the query to aser_long. Home > Close & Apply to load the cleaned table.

About ASER

ASER reports reading and arithmetic levels for rural children, gathered by Pratham volunteers across hundreds of districts. The wide-to-long reshape you just did is what you would need before charting, say, the share of Standard 3 children who can read a Standard 2 text, district by district.

Coach callout · Varna

Applied Steps is your methods section. When a funder asks how you cleaned a dataset, the honest answer is usually a shrug and a half-remembered set of Excel edits. Here it is a list anyone can read and re-run. Name your queries and your steps so the next person, or you in a year, can follow the logic without a phone call. Reproducibility is not a luxury in our sector; it is what separates a defensible number from a rumour.

Where module 3 goes wrong

Leaving a numeric column typed as text

It will not sum or average; measures return blank. Set the type explicitly.

Unpivoting named columns instead of Other Columns

New indicators next year get dropped silently. Unpivot everything but the key.

Forgetting to Trim keys

‘Bihar ’ and ‘Bihar’ split into two categories and relationships fail to match later.

Confusing Merge with Append

Merge adds columns by a key; Append stacks rows. Combining two survey rounds is an Append.

Try it yourself

Reshape and reason. You have an NFHS file with one row per district and twelve indicator columns, plus a separate small file mapping each district to its NITI Aayog aspirational-district status. Describe the two Power Query operations that get you to a single long table you can slice by aspirational status.

Show a worked answer

First, Unpivot Other Columns on the district key in the NFHS file, collapsing the twelve indicator columns into Indicator and Value, giving one row per district-indicator pair. Second, Merge that long table with the aspirational-district lookup on the district key (a left outer join keeps every NFHS row and adds the status column). You now have District, Indicator, Value and Aspirational-status in one table, so a single slicer can compare aspirational against other districts across any indicator. Note the order: unpivot first keeps the merge simple, one key, one added column.

Check yourself · 5 questions

A factsheet has one row per district and a column per indicator. To chart it you should:

A. Build one chart per columnB. Unpivot the indicator columns to get one row per district-indicator pairC. Delete all but one indicator columnD. Load as-is, Power BI reshapes automatically

B is correct. Unpivoting produces the tidy long shape, one observation per row, that charts and slicers expect.

Why Unpivot Other Columns rather than Unpivot Columns?

A. It runs fasterB. A new indicator column next year flows in automatically instead of being ignoredC. It is the only one for textD. It keeps the wide layout

B is correct. Unpivoting everything except the key future-proofs the query against added columns.

'Bihar' and 'Bihar ' split your totals into two states. The fix is:

A. Change the type to numberB. Apply Trim to remove the trailing spaceC. Merge the queriesD. Delete a row

B is correct. A trailing space makes the strings unequal. Trim removes leading and trailing spaces.

To combine the NFHS-4 and NFHS-5 files into one table with a Round column, use:

A. MergeB. AppendC. UnpivotD. Split column

B is correct. Append stacks rows (a union). Merge joins columns by a key.

Beyond cleaning, Applied Steps is most valuable as:

A. Password storageB. A transparent, re-runnable record of every change to the dataC. The chart designerD. The publish button

B is correct. It is an audit trail and a recipe in one, re-running on every refresh: the heart of reproducible preparation.

Reflection prompts

1. Take one survey file. Is it wide or long? If wide, how many indicator columns collapse into one Indicator field after unpivoting?
2. Where do you currently make undocumented manual edits in Excel? Each belongs in Applied Steps instead.
3. If a funder asked you to show exactly how you cleaned a dataset, could you today? Power Query is how that becomes yes.

Resources

Microsoft Learn: Power QueryOfficial documentation, including unpivot, merge and append.Wickham, Tidy Data (2014)The paper behind long-format data and why unpivot matters.ASER reports (Pratham)Annual Status of Education Report data and methodology.

A model is how your tables relate to each other. When it is set up well, charts, filters and calculations all behave. When it is one giant flat table instead, you end up fighting the tool. This module covers how to structure the data — ideas that predate Power BI and apply well beyond it.

By the end you can

• Explain why a single flat table breaks slicing, counting and trust
• Build a star schema with a fact table and dimension tables, and name the grain
• Create relationships with the right cardinality and cross-filter direction
• Build and mark a dedicated date table for reliable time calculations

Why one flat table fails

Flat swamp versus star

The spreadsheet habit is one wide sheet: district, state, region, scheme, year, indicator, value, all on every row. It works until you slice. Region names repeat thousands of times, so a single typo invents a phantom region. Distinct counts get awkward. Filters behave unpredictably. This is not a Power BI quirk; it is the redundancy problem that Edgar Codd’s relational model (1970) and database normalisation were designed to solve. The Power BI answer is a particular, deliberately denormalised shape: the star schema.

Fact and dimension tables

A star schema

The star schema comes from Ralph Kimball’s dimensional modelling (The Data Warehouse Toolkit). It splits data into two kinds of table.

Fact table

The events or measurements, one row per observation. The numbers you aggregate live here: the indicator value, beneficiaries reached, budget spent. It holds keys that point out to the dimensions. Its grain, the meaning of one row, is the first thing to pin down.

Dimension tables

The descriptive context, each item listed once. A District table with one row per district and its state and region. A Date table. A Scheme table. You slice and group by these.

The shape is a star: the fact table in the centre, dimensions around it, each joined by a key. A District dimension of 707 rows joins to a fact table of hundreds of thousands through a district code.

Relationships and cardinality

One-to-many, single direction

In Model view you drag a key from one table to the matching key in another to make a relationship. Power BI infers the cardinality, usually one-to-many: one row in the District dimension to many rows in the fact table. The cross-filter direction is normally single, so filtering District filters the facts, the natural flow. Resist setting everything to bidirectional; it creates ambiguous filter paths and slows the model. Turn it on only for a specific need you can name.

Power BI Desktop — Model viewinterface rebuilt for teaching

HomeModelingViewManage relationshipsMark as date table

fact_nfhs

district_id

indicator_id

value

dim_district

district_id

state

region

dim_date

date

round

year

1

∗

Model view. Drag district_id from the dimension to the fact to form a one-to-many relationship. The 1 and asterisk on the line show cardinality.

The date table you should always make

Time calculations, year-on-year change, comparisons across survey rounds, work reliably only against a dedicated Date table with one continuous row per period, marked as a date table under Modeling. Relying on a date column buried in the fact table will eventually fail. Build the Date table once, mark it, relate it, and the time intelligence in Module 5 simply works.

Turning the long NFHS table into a star

Start from the aser_long or NFHS long table from Module 3. In Power Query, reference it to create a District dimension: keep only District and State, then Home > Remove Duplicates so each district appears once. Close & Apply. In Model view, drag District from the fact onto District in the new dimension; confirm the line reads one-to-many, dimension to fact, single direction. Back in Report view, put State on a slicer and the value on a bar chart. The slicer now filters cleanly because it lives in a dimension, not repeated down a flat table.

Coach callout · Vandana Soni

The hardest habit to break is the love of the single master sheet; it feels safe to see everything in one place. But the model isn't for reading data; it is the set of relationships the engine uses to answer questions. When I coach teams I ask them to say each table's job out loud: ‘this one is facts, these are the things I slice by.’ If a table is doing both jobs, it usually needs splitting. Once the star clicks, people stop fighting their filters and start trusting them.

Coach callout · Vandana Soni

Pin down the grain before you build anything: what does one row of your fact table mean? One district-indicator-round? One household visit? One disbursement? If you cannot say it in a sentence, the model is not ready, and no amount of clever DAX later will rescue an unclear grain. This is the discipline Kimball insists on, and it is the cheapest hour you will spend on a dashboard.

Where module 4 goes wrong

One flat table for everything

Repeated context invites typos and breaks distinct counts. Split into fact and dimensions.

Bidirectional cross-filter everywhere

Ambiguous filter paths and slow models. Keep single direction unless a named need forces otherwise.

No dedicated date table

Time intelligence misbehaves. Build one, mark it, relate it.

Relationship will not form

Usually a type mismatch or un-trimmed key. Fix the type and Trim in Power Query first.

Try it yourself

Find the grain. A programme tracks monthly disbursements to self-help groups across districts, and wants a dashboard of total disbursed by district and by month, plus a count of distinct groups paid. Describe the fact table grain and the dimensions you would build.

Show a worked answer

The natural grain is one row per disbursement (group, date, amount), which lets you sum amounts and count distinct groups cleanly. The fact table holds group_id, date, district_id and amount. Build three dimensions: a District table (district_id, state, region), a Date table (one row per day, marked as date table, carrying month and year), and a Group table (group_id, formation year, block). Total disbursed is SUM over the fact; distinct groups paid is DISTINCTCOUNT(group_id); by-month and by-district come free from the dimensions. A flat monthly summary table would block the distinct-group count, which is why grain comes first.

Check yourself · 5 questions

The core problem with one wide flat table is:

A. It cannot hold over 1000 rowsB. Repeated context creates typo-driven phantoms and breaks clean slicing and distinct countsC. Power BI refuses flat tablesD. Flat tables cannot be charted

B is correct. A flat table repeats context on every row; one typo invents a category and distinct counts get awkward. A star fixes this.

In a star schema the numbers you aggregate live in:

A. The dimension tablesB. The fact tableC. The Date table onlyD. The slicers

B is correct. The fact table holds the measurements, one row per observation. Dimensions hold descriptive attributes.

A District dimension has 707 rows; the fact has 300,000. The relationship is:

A. Many-to-manyB. One-to-oneC. One-to-many, District to factD. None

C is correct. One district relates to many fact rows: the standard one-to-many star relationship.

Bidirectional cross-filtering is discouraged by default because:

A. It is paidB. It creates ambiguity and slows the model; use single direction unless neededC. It deletes relationshipsD. It is cloud-only

B is correct. Bidirectional filtering adds ambiguous paths and cost. Keep single direction and enable bidirectional only for a named requirement.

A relationship will not form between two district-code columns. The likely cause is:

A. The tables are too largeB. A type mismatch or stray spaces in the key, fixable in Power QueryC. You need ProD. Power BI cannot relate on codes

B is correct. Relationships need matching types and clean keys. Fix the type and Trim the key in Power Query.

Reflection prompts

1. Sketch your data as a star: what is the fact, what are the things you slice by? If you cannot tell which is which, that is the work.
2. Do you have a real date dimension, or a date stuck in the fact table? Plan to build a proper one.
3. Which label in your spreadsheet repeats on thousands of rows? That is a dimension waiting to be pulled out.

Resources

Star schema guidance (Microsoft)The official reference on dimensional design for Power BI.Model relationshipsCardinality and cross-filter direction explained.Create a date tableHow to build and mark a dedicated date table.

DAX is the formula language that turns loaded data into answers. It looks like Excel formulas but works differently. The thing to understand first is the difference between a calculated column and a measure; most of the rest builds on it. Underneath sits filter context, the subject of Alberto Ferrari and Marco Russo's reference, The Definitive Guide to DAX.

By the end you can

• Decide correctly between a calculated column and a measure
• Write aggregate measures with SUM, AVERAGE, COUNTROWS and DISTINCTCOUNT
• Use DIVIDE for safe division and CALCULATE to change filter context
• Build a change-over-rounds measure using variables and a date table

Calculated column versus measure

Calculated column versus measure

Both are written in DAX; they behave completely differently.

Calculated column

Computed row by row when data refreshes, then stored in the table. It sees one row at a time (row context). Use sparingly: it costs memory and is usually wrong for aggregates. Good for a flag that classifies each district.

Measure

Computed on demand when a visual asks, against whatever filters are active (filter context). Nothing is stored. This is what you want for totals, averages, rates and shares. Almost every number on a dashboard should be a measure.

The test

The deciding question: does the number change when you click a slicer? A district's name does not change when you filter, so it can be a column. A stunting rate recomputes for whatever districts are selected, so it must be a measure.

The handful of functions that do most of the work

• SUM, AVERAGE, MIN, MAX: aggregates over a column.
• COUNTROWS, DISTINCTCOUNT: count rows, count distinct values (how many districts reported).
• DIVIDE: safe division. DIVIDE(a, b) returns blank instead of an error when b is zero. Use it instead of the slash, always.
• CALCULATE: the most important function in DAX. It evaluates a measure while changing the filter context. This is how you build any comparison.
• FILTER: returns a filtered table, used inside CALCULATE for conditions a simple equals cannot express.

Your first measure

A rate is a sum over a sum. Write it as a measure so it recomputes for any selection.

Power BI Desktop — measure editorinterface rebuilt for teaching

HomeModelingViewNew measureFormat: %

fx  Stunting Rate =

DIVIDE( SUM(fact[children_stunted]), SUM(fact[children_measured]) )

suggestion: DIVIDE(numerator, denominator [, alternateResult])

The measure editor with the formula bar and IntelliSense. Measures live in the model, in the Fields pane, not inside a single visual.

Because it is a measure, this returns the right rate whether the visual shows one district, one state, or the whole country.

CALCULATE: changing the question

CALCULATE changes the question

CALCULATE takes a measure and a set of filters and evaluates the measure as if those filters applied. It is how comparisons get built.

CALCULATE re-evaluates the existing Stunting Rate but forces the state filter to Bihar, whatever the visual otherwise shows.

Time intelligence needs that date table

Comparisons across rounds or years rely on the dedicated date or round table from Module 4. With it in place, change computes cleanly. Variables (VAR) make multi-step DAX legible, which matters when someone else, or future you, reads it.

VAR captures the current rate and the NFHS-4 rate, then RETURN gives the difference. Readable DAX is maintainable DAX.

Why DIVIDE, always

Why DIVIDE, always

Coach callout · Varna

Name measures the way you would name a variable in a methods section, so a stranger knows what they compute. Stunting Rate is good; Measure 1 is how dashboards become unmaintainable. And reach for DIVIDE every time you divide. I have seen a published district dashboard show a blank where a rate should be, because one district had zero children measured and a plain division threw an error that propagated across the visual. DIVIDE returns a clean blank and the rest survives. Small habit, real consequence.

Reading filter context off a real visual

Put Stunting Rate in a card and it shows the national rate, because no district filter is active, so the filter context is ‘everything’. Drop the same measure into a bar chart with District on the axis, and each bar shows that district's rate, because each bar's filter context is that one district. You wrote one measure; the visual decided the filter context. That is the whole of DAX in a sentence: a measure is an instruction, and the filter context at the moment of evaluation decides what it returns. Add a State slicer and click Bihar, and both the card and the bars recompute, because the slicer narrowed the context. Nothing was recalculated and stored; it was all computed on demand.

Hands-on lab · Build the core measures for the NFHS dashboard

Build the measures the dashboard will reuse. Measures live in the model, so build once and use everywhere.

1. In Report view, right-click your fact table in the Fields pane > New measure.
2. Write the Stunting Rate measure using DIVIDE. Press enter; it appears with a calculator icon.
3. Set its format to Percentage, one decimal, from the Measure tools ribbon.
4. Create Districts Reporting = DISTINCTCOUNT( fact[district] ) for coverage.
5. Create the CALCULATE-based single-state measure and confirm filter context behaves as expected.
6. Drop Stunting Rate into a card and a bar chart by district; the card shows the overall rate, the bars the per-district rates, from one measure.

Coach callout · Varna

Do not paste raw numbers into measures. If you find yourself typing a national average into a formula, stop. Hard-coded numbers go stale the moment the data updates and make the dashboard lie without anyone noticing. Compute from the data every time. The point of this whole tool is that next quarter's refresh updates every number on its own; a typed constant opts straight out of that promise.

Where module 5 goes wrong

Using a calculated column for an aggregate

Columns are fixed at refresh and waste memory. Rates, totals and shares are measures.

Dividing with a plain slash

A zero denominator errors and can break the visual. Use DIVIDE.

Vague measure names

‘Measure 1’ is unmaintainable. Name for what it computes.

Time calculations without a date table

They misbehave. Mark a date table (Module 4) first.

Try it yourself

Measure or column?. For each, decide measure or calculated column: (a) the percentage of children stunted, (b) a label marking each district 'aspirational' or 'other', (c) the number of distinct districts reporting, (d) the change in stunting since NFHS-4.

Show a worked answer

(a) Measure: a rate that recomputes for any selection. (b) Calculated column: a per-row classification that does not change with filters, so it can be stored and used as a slicer. (c) Measure: DISTINCTCOUNT over the current filter context. (d) Measure: a comparison across rounds via CALCULATE and a date table. The single column is (b), precisely because it is the one value that does not depend on what the user clicks.

Check yourself · 5 questions

A number should recompute whenever a user clicks a slicer. Build it as a:

A. Calculated columnB. MeasureC. Static text boxD. New table

B is correct. Measures respond to filter context. Columns are fixed at refresh and stored per row.

Why DIVIDE(a, b) instead of a / b?

A. It is fasterB. It returns blank instead of an error when the denominator is zeroC. It rounds automaticallyD. It is required for measures

B is correct. DIVIDE handles division by zero, returning blank. A plain slash errors and can break a whole visual.

CALCULATE does what?

A. Deletes filters permanentlyB. Evaluates a measure while modifying the filter contextC. Creates a new tableD. Counts rows

B is correct. CALCULATE computes a measure under a changed set of filters: the engine of comparisons and conditional aggregates.

Which counts distinct reporting districts correctly?

A. SUM( fact[district] )B. COUNTROWS( fact )C. DISTINCTCOUNT( fact[district] )D. AVERAGE( fact[district] )

C is correct. DISTINCTCOUNT counts unique values. COUNTROWS over-counts when a district has several indicator rows.

Change between survey rounds depends on:

A. A bidirectional relationshipB. A dedicated date or period table related to the fact tableC. A Pro licenceD. Calculated columns only

B is correct. Reliable time comparisons need a proper date or period dimension, as built in Module 4.

Reflection prompts

1. List the five numbers your dashboard must show. For each: measure (changes with selection) or column (fixed)? Most are measures.
2. Where in your spreadsheets is a number typed into a formula? That hard-code should be computed.
3. Would a colleague opening your model understand each measure from its name alone?

Resources

DAX function referenceMicrosoft's complete free reference.DAX Guide (SQLBI)Ferrari and Russo's free reference and pattern library.Columns versus measuresThe distinction this module turns on.

A chart makes a claim about the data, and the wrong chart or careless formatting makes a claim you did not intend. This module covers both building visuals and choosing them responsibly, which your Data Visualisation course goes into at length. Chart choice is not just taste: Cleveland and McGill (1984) measured how accurately people read each kind of encoding, and length on a common scale beats angle and area.

By the end you can

• Choose a chart from the question, using the perceptual accuracy hierarchy
• Build bars, lines, maps, cards and slicers and wire them to measures
• Avoid the count-versus-rate map trap and the truncated axis
• Make visuals accessible with sort order, labels and alt text

Match the chart to the question

Match the chart to the question

Cleveland and McGill's hierarchy explains why bars beat pies: people judge length on a common baseline far more accurately than angle. Translate that into a working guide.

Building a visual, mechanically

Power BI Desktop — building the reportinterface rebuilt for teaching

HomeInsertFormatViewNew visualSlicer

Stunting rate by district (sorted)

35.5%National rate

Data

Stunting Rate District State Round

Visualizations

X-axis

District

Y-axis

Stunting Rate

Drag District to the Axis well and Stunting Rate to Values. The card and map run off the same measures.

Hands-on lab · Build the core visuals

You have the measures from Module 5. Put them on a page.

1. Click a chart type in the Visualizations pane, or drag a field onto the canvas and accept the suggestion.
2. Drag District to Axis and Stunting Rate to Values for a bar chart.
3. Use the visual's More options > Sort axis > by Stunting Rate, descending, so the worst districts read first.
4. Add a Card and drop the overall Stunting Rate for the headline.
5. Add a Slicer with State; clicking a state now filters every visual.
6. Add a Map, put district or state in Location and the measure in colour or bubble size; check the geocoding finds the right places.

The ethics are not optional

The truncated axis

Power BI makes several misleading choices easy, and a couple are on by default. These are the same hazards your DataViz course names; here is how they show up in this tool.

Truncated axes

A bar chart whose y-axis does not start at zero exaggerates differences. Force bars to start at zero. For a line showing change, a non-zero start can be defensible if you say so.

Counts versus rates on maps

A choropleth coloured by the number of stunted children mostly maps population: big districts look bad for being big. Map the rate, unless the count is genuinely the point.

Pie charts past three slices

People compare angles poorly. Beyond three categories a sorted bar communicates better, exactly as Cleveland and McGill predict.

Colour and access

Do not encode meaning in red-green alone; about one man in twelve cannot separate them. Add labels or patterns, and write alt text so screen readers convey the visual.

Count map versus rate map

Coach callout · Vandana Soni

The chart that most often misleads in our sector is the district map coloured by a raw count. A funder sees a dark patch over a populous district and concludes the problem is concentrated there, when that district simply has more people. Always ask: is this colour showing where the rate is worst, or where the most people live? Normalise to a rate and the map tells the truth. This is not a styling preference; it is the difference between a map that informs a decision and one that misdirects money.

Coach callout · Vandana Soni

Write the one-sentence takeaway before you format anything. If the sentence is ‘stunting is worst in these five districts’, then those five bars should be the darkest thing on the page and everything else should recede. If everything on the page is emphasised, nothing stands out. Knaflic calls this drawing attention; in practice it means deciding what the reader should notice first, then making the design obey.

Where module 6 goes wrong

Truncated bar axes

Small real gaps look huge. Start bars at zero.

Count maps for a rate question

They map population. Colour by the rate.

Red-green as the only signal

Colour-blind readers lose the message. Add labels or patterns and alt text.

A pie with eight slices

Unreadable angles. Use a sorted bar.

Try it yourself

Diagnose a misleading dashboard. A district health dashboard shows a map shaded by total anaemia cases, a bar chart of six blocks with the y-axis starting at 40 percent, and a red-green legend with no labels. Name what each could mislead a reader into believing, and the fix.

Show a worked answer

The count map implies anaemia is concentrated in the most populous districts; a reader may steer resources there even if their rate is average. Fix: colour by anaemia prevalence (a rate). The truncated bar axis at 40 percent exaggerates small differences between blocks, suggesting a crisis gap where there may be a few points; fix: start the axis at zero. The red-green legend without labels is invisible to colour-blind readers and ambiguous to everyone; fix: add direct labels or a pattern, and write alt text. None of these is dishonesty by intent, which is exactly why they are dangerous: the defaults made the claims, not the analyst.

Check yourself · 5 questions

To compare 28 states on one indicator, the clearest visual is usually a:

A. Pie chartB. Sorted bar chartC. Line chartD. Scatter plot

B is correct. Length is read more accurately than angle or area (Cleveland and McGill). A sorted bar gives an instant ranking; a 28-slice pie is unreadable.

A choropleth coloured by the raw number of stunted children mostly shows:

A. Where the rate is worstB. Where the most people live, since counts track populationC. Nothing usefulD. The same as a rate map

B is correct. Counts scale with population. Map the rate to show where the problem is proportionally worst.

Why force a bar chart's value axis to start at zero?

A. It looks tidierB. A truncated axis visually exaggerates differences, misleading the readerC. Power BI requires itD. It loads faster

B is correct. Bar length encodes magnitude; a non-zero start makes small differences look large.

Encoding meaning only in red versus green fails because:

A. Those colours are copyrightedB. About one man in twelve cannot reliably distinguish themC. Power BI cannot render greenD. It slows the report

B is correct. Red-green colour vision deficiency is common. Pair colour with labels or patterns and add alt text.

When give a matrix or table rather than a chart?

A. NeverB. When the reader needs exact values, not a visual shapeC. Only on the free tierD. Only for time series

B is correct. A table is right when precise numbers matter; a chart when a pattern or ranking is the point.

Reflection prompts

1. Compare a dashboard you admire with one you find confusing. What chart choices separate them?
2. Do any of your maps colour by count rather than rate? What decision might that have nudged wrongly?
3. Pick a visual you make often. Could a colour-blind colleague read it without the colour? If not, what fixes it?

Resources

ImpactMojo Data VisualisationThe companion flagship on charting craft and ethics.Power BI accessibilityMicrosoft's guidance, including alt text.Visual types in Power BIThe built-in visuals and their uses.

A handful of visuals is not yet a report. A good report is a single page a busy district officer can take in quickly and then explore further. Stephen Few defines a dashboard as the most important information needed to meet an objective, arranged on one screen so it can be monitored at a glance. If the reader has to scroll through pages, you have built a report, not a dashboard.

By the end you can

• Lay out a single-page dashboard with a clear visual hierarchy
• Use cross-filtering, drill-through and bookmarks so one page does the work of ten
• Apply Shneiderman's overview-zoom-filter-detail pattern to interactivity
• Set a report theme so a whole organisation's dashboards stay consistent

One page, well laid out

A one-page dashboard layout

The default canvas is a corporate 16:9 rectangle. Treat it as a designed page.

• Top strip for headline cards: the two or three numbers that frame everything, national rate, districts reporting, change since the last round.
• A clear primary visual, usually the sorted bar or the map, given the most space.
• Slicers grouped together, top or left, so the reader knows where the controls are.
• Alignment and spacing: View > Gridlines and Snap to grid. Misaligned visuals read as careless and cost you credibility you never see being spent.
• A title and a source line: every report states what it shows and where the data came from, with the survey round and year. In our sector an unsourced number is worthless.

Interactivity that earns the page back

Interactivity that earns the page back

Ben Shneiderman's information-seeking mantra is the design rule: overview first, zoom and filter, then details on demand. Power BI gives you the pieces.

Cross-filtering

Click a bar and every other visual filters to it, no setup. Click Bihar and the map, cards and tables narrow to Bihar. Tune it per visual via Format > Edit interactions.

Drill-through

Right-click a district and jump to a detail page built for one district, carrying the filter. The summary stays clean; detail lives one click away (details on demand).

Bookmarks and buttons

Capture a page state, filters and visibility, and bind it to a button. Build a Health / Education / Nutrition toggle that swaps focus without leaving the page.

Report-page tooltips

Design a mini-page and set it as a tooltip, so hovering a district shows a small custom card instead of a bare number.

Assembling the page, narrated

View > turn on Gridlines and Snap to grid. Drop three Cards across the top for your headline measures and align their tops with Format > Align. Place the sorted bar and the map below, giving the bar more width. Group the slicers in one corner; add a Text box title and a one-line source. Create a second page as a district detail view, select the bar, and set up Drill through on the District field so a right-click jumps there carrying the filter. Add two Bookmarks, one for health indicators and one for education, and wire each to a button via Insert > Buttons with Action set to Bookmark. Now test as a reader: click a state, drill into a district, use the toggle. Whatever surprises you is a design bug; fix it before anyone else sees it.

Coach callout · Varna

Design for the person with ninety seconds and a specific question, not for yourself who knows where everything is. My test: hand the report to someone who has never seen it and say nothing. Watch where their eyes go and where they stall. If they cannot find the headline number in three seconds, or do not realise the bars are clickable, the design failed regardless of how clever the model underneath is. A dashboard nobody can navigate is a private spreadsheet with extra steps.

Consistency for free

Power BI ships with theme JSON. View > Themes loads a small file that sets palette and fonts once, so every visual is consistent and on-brand. Build one theme for your organisation and reuse it: the same instinct as the design tokens behind a good website.

Coach callout · Varna

Resist the urge to add one more chart. Every visual you add taxes the reader's attention and the page's load time. The strongest M&E dashboards I have seen show four or five things well and send everything else to a drill-through page. If a stakeholder asks for ‘just one more number’, ask which existing one it replaces. Restraint is a design decision, and usually the right one.

Where module 7 goes wrong

Misaligned, ungridded visuals

They read as careless. Turn on Snap to grid and align.

No source line

An unsourced number cannot be trusted. State the round and year on every page.

One crowded page instead of summary plus drill-through

Send detail to a drill-through page and keep the summary clean.

Inconsistent colours across reports

Build one theme JSON and reuse it organisation-wide.

Try it yourself

Redesign brief. A team's dashboard is a single page with nine visuals, no titles, slicers scattered in three corners, and the key chart bottom-right in the smallest tile. Give three concrete changes, in priority order, and the interactivity feature that would let them cut visuals without losing detail.

Show a worked answer

First, establish hierarchy: move the key chart to the top-left and enlarge it; demote or remove the least-used visuals. Second, group the slicers in one corner so the controls are findable. Third, add a title and source line so the page is self-explanatory and trustworthy. To cut visuals without losing detail, use drill-through: keep a clean four-or-five-visual summary and move the detailed breakdowns to a per-item detail page reached by right-click, which is Shneiderman's details on demand. The aim is fewer things, better placed, with depth one click away.

Check yourself · 5 questions

Clicking a bar and watching the other visuals narrow to it is:

A. Drill-throughB. Cross-filteringC. BookmarkingD. A slicer

B is correct. Cross-filtering is on by default. Drill-through is the right-click jump to a detail page.

For a clean summary plus a per-district detail view one click away, use:

A. A second fileB. Drill-through to a detail page carrying the filterC. More slicersD. A pie chart

B is correct. Drill-through sends the reader to a detail page for the selected item, keeping the summary uncluttered.

Bookmarks bound to buttons let you:

A. Publish freeB. Capture a page state and toggle views without leaving the pageC. Write DAX fasterD. Encrypt the report

B is correct. A bookmark stores filters and visibility; a button triggers it, building toggles like Health / Education.

Why a source line with round and year on every page?

A. Power BI requires itB. An unsourced number cannot be trusted or checkedC. It shrinks the fileD. It enables drill-through

B is correct. A figure without source and date is not usable in development work; stating the round and year lets a reader verify.

A report theme (theme JSON) is used to:

A. Store dataB. Set palette and fonts once so visuals stay consistent and on-brandC. Count rowsD. Share the report

B is correct. A theme applies colours and typography across the report: the Power BI equivalent of design tokens.

Reflection prompts

1. Hand a report you made to a colleague with no explanation. Where do they stall? That is your redesign list.
2. Which report would benefit from a clean summary plus a drill-through detail page rather than one crowded page?
3. Does your organisation have a consistent look across dashboards? A single theme JSON fixes it in an afternoon.

Resources

Drill-throughSetting up drill-through pages.Bookmarks and buttonsBuilding interactive toggles.Report themesCreating and applying a theme JSON.

This is where ‘free’ needs an asterisk. Building a report costs nothing; sharing it, beyond a few channels, either costs money or carries risk. This module sets out what you can do for free, where the paid wall sits, and the privacy obligations on anyone who publishes data about real people in India.

By the end you can

• List the genuinely free ways to share and what each is good for
• Locate the Pro wall and price live sharing per person
• Judge when Publish to web is safe, and when it is a serious risk
• Apply the DPDP Act 2023's practical floor to what you publish

The free ways to share

The free/paid wall

Export to PDF / PowerPoint

File > Export. Free from Desktop. The PDF is static; the PowerPoint drops each page as an image. Fine for a board pack or an email.

Hand over the .pbix

Send the file. The recipient opens it in their free Desktop and interacts fully. The catch: they need Windows and the data travels with the file.

Print or screenshot

Crude but free. For a one-off figure in a report, sometimes the honest answer.

Free-tool reality

The wall, plainly. To publish a report to the Service so colleagues view it live in a browser, the publisher needs a Pro licence, 14 US dollars per user per month in 2026. For others to view it, each also needs Pro, unless the report sits on a large dedicated capacity (Fabric F64 or above) that only big organisations buy. A five-person team that all wants live access is five Pro licences, around 70 US dollars a month. The free path is Desktop plus PDF, PowerPoint or .pbix handoff. That is a real limit, not a trick.

Publish to web: powerful and dangerous

Publish to web is fully public

There is a free-feeling option called Publish to web. It produces a public link and embed code, no licence needed to view. Read this twice: Publish to web makes the report and its data visible to anyone on the internet, with no authentication. Search engines can index it. It is built for public, non-sensitive content like an open data story. It is never for personal data, unpublished figures, or anything a community has not consented to share.

Power BI — Publish to web (public)interface rebuilt for teaching

PublishEmbed codePublicWarning

Anyone on the internet can see this report and its data

Do not publish confidential or personal data. This report will be discoverable and indexable.

https://app.powerbi.com/view?r=eyJrIjoi... (public link)

The Publish to web dialog carries its own warning. Take it literally: this is a printing press, not a private channel.

Watch out

People have leaked salary tables, patient lists and unreleased survey data this way, believing an obscure link was private. It is not. If the data should not be on a billboard, it must not go through Publish to web.

Row-level security, and where it applies

Where row-level security actually bites

Row-level security (RLS) lets you define roles so a Bihar officer sees only Bihar rows. You set it up in Desktop under Modeling > Manage roles, writing a filter per role. The honest catch: RLS is only enforced once the report is published to the Service with Pro, Premium or capacity licensing. In a .pbix handed to someone, RLS protects nothing, because they hold the whole file. RLS is a sharing-tier control, not a substitute for keeping sensitive data out of the file.

Power BI Desktop — Manage roles (RLS)interface rebuilt for teaching

ModelingSecurityManage rolesView as

Roles

Bihar_officer

Odisha_officer

Table filter (DAX)

[state] = "Bihar"

enforced only when published with a paid licence

Manage roles in Desktop: a DAX filter per role. It does real work only on the published, licensed Service.

The legal floor: the DPDP Act, 2023

India's Digital Personal Data Protection Act, 2023 governs the processing of digital personal data. For a dashboard builder, the practical implications are concrete and sit upstream of any sharing decision.

• Personal data needs a lawful basis. Beneficiary names, phone numbers, household identifiers, and health details about identifiable individuals are personal data. District aggregates usually are not, which is why district-level dashboards are the safer default.
• Purpose and minimisation. Hold only what the dashboard genuinely needs. If district aggregates answer the question, do not load the individual records at all.
• Children's data carries heightened obligations under the Act. Education and child-health datasets deserve particular care.
• Sharing is processing. Publishing a dashboard that exposes personal data, especially via Publish to web, is a processing act you must be able to justify.

From the lawyer's chair

This is general information, not legal advice for your situation, and the DPDP Act's rules continue to be operationalised through subordinate rules. The safe engineering default that keeps you clear of most of it: build on aggregates, keep identifiable records off any shared or published artifact, and document why each personal field, if any, is present.

Coach callout · Vandana Soni

Sharing is really a governance question, not a software one. Before anyone asks how to publish, ask what is in the report, who should see it, and what happens if the link gets out. Most dashboards in our sector should not contain a single identifiable person; work at district or block level and the privacy problem mostly goes away. When a programme genuinely needs individual-level views, slow down, involve someone who knows the DPDP Act, and treat RLS and access control seriously.

Coach callout · Vandana Soni

Cost the sharing honestly at the start of a project, not the end. If five people need live browser access, that is roughly 70 US dollars a month, every month, forever. Sometimes a scheduled PDF emailed round is the right answer and saves a budget line that a small organisation cannot spare. The tool's defaults nudge you toward the paid cloud; your job is to ask whether the question really needs it.

Where module 8 goes wrong

Assuming live sharing is free

Publishing for live viewing needs Pro per person. Free routes are PDF, PowerPoint, .pbix.

Using Publish to web for sensitive data

It is fully public and indexable. Open content only.

Trusting RLS in a handed-over .pbix

RLS is enforced only on the licensed Service. The file holds everything.

Loading personal records you do not need

Build on aggregates; minimise to stay clear of most DPDP obligations.

Try it yourself

Choose a sharing route. A district health programme wants twelve block officers to each see only their own block's live dashboard, refreshed weekly, and the data includes facility-level counts but no patient identifiers. What do you recommend, and what does it cost and require?

Show a worked answer

Live, per-officer filtered views mean publishing to the Service with row-level security: define a role per block (or a dynamic role keyed to the officer's login) in Desktop under Manage roles, then publish. RLS is enforced only on the licensed Service, so each of the twelve officers who needs live access needs a Pro licence, roughly 12 times 14 = 168 US dollars a month, unless the organisation has Fabric capacity. Weekly refresh is a scheduled refresh in the Service, within Pro limits. Because the data is facility-level counts with no identifiers, the DPDP exposure is low, but confirm no field re-identifies individuals. If the budget cannot carry twelve Pro seats, the fallback is twelve filtered PDFs emailed weekly, which is free but static and loses the live interactivity.

Check yourself · 5 questions

A genuinely free way to share a finished report is:

A. Publishing to the Service for live viewingB. Exporting to PDF or PowerPoint, or handing over the .pbixC. Giving everyone a Pro accountD. Buying Fabric F64

B is correct. Export and .pbix handoff are free from Desktop. Live browser sharing needs Pro.

Five people want to view a live report in their browser. Roughly the 2026 cost?

A. NothingB. About 70 US dollars a month, five Pro licences at 14C. A one-time 14 dollarsD. Only the publisher pays

B is correct. Each viewer of a non-capacity report needs Pro. Five at 14 is about 70 a month.

The critical risk of Publish to web is:

A. It is slowB. It makes the report and data public, unauthenticated, and search-indexableC. It costs extraD. It is Mac-only

B is correct. Publish to web is fully public: open, non-sensitive content only.

RLS defined in Desktop is actually enforced:

A. In the emailed .pbixB. Only once published to the Service with a paid licenceC. NeverD. Only on Mac

B is correct. RLS is a sharing-tier control; a handed-over .pbix gives the recipient everything. It works on the licensed Service.

The cleanest way to avoid most DPDP obligations in a dashboard is to:

A. Encrypt the .pbixB. Work with district or block aggregates and keep identifiable records out of shared artifactsC. Use Publish to webD. Add slicers

B is correct. Aggregated, non-identifiable data sits largely outside personal-data rules: the safest engineering default.

Reflection prompts

1. Count the people who genuinely need live access to your next dashboard, times 14 US dollars. Budgeted, or is a PDF the honest answer?
2. Does anything you plan to share contain identifiable individuals? What will you aggregate away before it leaves your machine?
3. Write the one-sentence question your capstone answers. If you cannot, it is not ready to build.

Resources

Power BI pricingOfficial Pro, Premium and Fabric pricing.DPDP Act, 2023 (MeitY)India's Digital Personal Data Protection Act and updates.Row-level securityDefining and enforcing RLS.