As a Power Query fan I’m naturally interested in Copilot in Dataflows Gen2. But since Power Query is already very easy to use and since I’m pretty experienced at using it, can Copilot really do anything to help me be more productive? Well one thing I need to do a lot for blog posts, demos and testing is to generate tables of sample data. Typing data into the Enter Data source can be quite tedious and writing the M code to generate sample data isn’t straightforward, so I was wondering if Copilot in Dataflows Gen2 could do it for me. It turns out it can do so really well.
After a little bit of experimentation I’ve come up with an example prompt that can be easily modified to return a table of random values with columns of different data types and a given number of rows, and which works fairly reliably:
Return a table consisting of the following five columns:
Country, containing one of 10 random country names;
Product, containing one of 15 random fruit names;
Person, containing one of 20 random first names that are common in the United Kingdom;
Date, containing random dates between January 1st 2019 and December 31st 2024;
Sales, containing random integers between 0 and 500.
The table should have 1000 rows. Set an appropriate data type for each column.
Here’s an example of the M code generated:
let
// Generate lists of random values
Countries = {"United States", "Canada", "United Kingdom", "Germany", "France", "Italy", "Spain", "Australia", "India", "China"},
Products = {"Apple", "Banana", "Cherry", "Date", "Elderberry", "Fig", "Grape", "Honeydew", "Kiwi", "Lemon", "Mango", "Nectarine", "Orange", "Papaya", "Quince"},
People = {"James", "Mary", "John", "Patricia", "Robert", "Jennifer", "Michael", "Linda", "William", "Elizabeth", "David", "Barbara", "Richard", "Susan", "Joseph", "Jessica", "Thomas", "Sarah", "Charles", "Karen"},
StartDate = #date(2019, 1, 1),
EndDate = #date(2024, 12, 31),
DaysBetween = Duration.Days(EndDate - StartDate) + 1,
// Function to generate random dates
RandomDate = () => Date.From(StartDate + #duration(Number.RoundDown(Number.RandomBetween(0, DaysBetween)), 0, 0, 0)),
// Generate table records
Records = List.Transform({1..1000}, each [
Country = Countries{Number.Mod(_, List.Count(Countries))},
Product = Products{Number.Mod(_, List.Count(Products))},
Person = People{Number.Mod(_, List.Count(People))},
Date = RandomDate(),
Sales = Number.RoundDown(Number.RandomBetween(0, 500))
]),
// Convert list of records to table and set data types
Source = Table.FromRecords(Records),
TypedTable = Table.TransformColumnTypes(Source, {
{"Country", type text},
{"Product", type text},
{"Person", type text},
{"Date", type date},
{"Sales", Int32.Type}
})
in
TypedTable
And here’s an example of the output:
Definitely a time saver as far as I’m concerned. Is it totally reliable? No: it occasionally produces code that errors or which doesn’t contain genuinely random values, but it’s good enough and it’s faster to try the prompt once or twice than write the code myself. I know there are other, more sophisticated ways of generating sample data like this in Fabric, for example in Python, but as I said I’m a Power Query person.
If you have a Power BI report with a matrix visual on it it’s quite likely that you’ll want all the levels in the matrix to be fully expanded by default. But did you know that the way you expand all the levels could have performance implications, especially if you’re using DirectQuery mode? Here’s an example.
I have a DirectQuery semantic model built on top of some of the tables from the SQL Server AdventureWorksDW sample database (apologies for the poor naming):
There are four DAX measures defined on it:
Sales Amount = SUM(FactInternetSales[SalesAmount])
Monday Sales = CALCULATE([Sales Amount], 'DimDate'[EnglishDayNameOfWeek]="Monday")
January Sales = CALCULATE([Sales Amount], 'DimDate'[EnglishMonthName]="January")
Class H Sales = CALCULATE([Sales Amount], 'DimProduct'[Class]="H")
I wrote these measures specifically to exacerbate the problem I’m going to show (by reducing the amount of fusion that is possible) but they are pretty normal, reasonable measures that you might find in any semantic model.
Now let’s say you add a matrix visual to a report page, put these four measures onto the columns axis of the matrix, and drop the CalendarYear column (from the DimDate table), the Color column and the Style column (both from the DimProduct table) onto the rows axis of the matrix. At this point it looks like this:
…but what you want to do now is show all the styles and colours too.
One way to do it – not the most efficient way, but some people like me just love to click – is to expand every year and style individually:
It doesn’t take too long to expand everything and after all you only need to do it once, right? But let’s take the DAX query generated for this visual and paste it into DAX Studio with Server Timings turned on and see what we can see:
There are 14 separate Storage Engine queries – which result in 14 separate SQL queries being sent to SQL Server. The first two Storage Engine/SQL queries get a list of which years and styles have been drilled down on and then there are (4 measures) * (3 levels of granularity) = 12 other Storage Engine queries to get the data shown in the visual. The overall duration of 230ms here is very low but in the real world the SQL queries could be a lot slower, making the DAX query very slow.
The default limits on the number of SQL queries that a DAX query can run in parallel have a big impact on overall performance here as you can see; even though you can increase those limits you may then hit the maximum number of connections that can be opened up to a DirectQuery source, and even though you can increase that limit too if you’re running on a capacity, there are hard limits here. If Power BI needs to open new connections to the data source in order to run these SQL queries, that can also slow things down too because there can sometimes be a noticeable wait when connections are opened. Reducing the number of Storage Engine queries generated by a DAX query is very important when tuning DirectQuery models; the effect is going to be a lot less noticeable on an Import or Direct Lake semantic model but it could still cause problems.
There’s good news though. If you expand the levels in your matrix in a different (and to be honest, much more convenient) way using the “Expand all down one level in the hierarchy” button on the visual header or the “Expand to next level” option on the right-click menu for the rows like so:
…then you get the same result but with a much more efficient DAX query. Here’s what Server Timings shows for the DAX query generated for the fully expanded matrix now:
This time there are only four Storage Engine/SQL queries, one for each measure, and the overall duration is just 50ms. Even though, as you can see from the screenshot, only three Storage Engine/SQL queries can run in parallel and the fourth has to wait for the first query to finish so it can run, that’s less of an issue given the smaller number of queries. I won’t bother showing the DAX for the two versions of the matrix but it’s clear when you look at them the second one is more efficient because it knows it can expand everything on rows rather than just what has been clicked. Of course this type of optimisation is only possible if you are fully expanding your matrix though.
Over the last few years one topic I have blogged about several times is how to link the detailed data about DAX query execution that can be found in Log Analytics – and now Workspace Monitoring – to the visual in a Power BI report that generated those DAX queries, something that is extremely useful when you’re performance tuning Power BI reports. My first post here from 2021 showed how write the KQL for Log Analytics but the problem has always been how to get the IDs of the visuals in a Power BI report. You can get the IDs from the definition of the Power BI report, as shown here, and Sandeep Pawar has a great post on some other methods here, but all these methods were superseded in the March release of Power BI with the ability to copy the IDs by right clicking on the visual in a Power BI report when editing it (thank you Rui Romano!).
This made me realise that it’s time to revisit my first post on how to get the query details in KQL since the column names in Workspace Monitoring are slightly different from Log Analytics and, indeed, the KQL needed can be simplified from my original version. Here’s a KQL query that you can run in a KQL Queryset connected to your Monitoring Eventhouse:
let VisualId = "InsertVisualIdHere";
SemanticModelLogs
| search VisualId
| project Timestamp, OperationName, OperationDetailName, DurationMs, EventText, OperationId
| order by Timestamp asc
Once you’ve copied the visual’s ID from the report (remember you need to specifically enable this feature and that you need to be in Edit mode) by right-clicking on it and selecting “Copy object name”:
…then you just need to paste the ID into the let statement in the first line of the KQL query and run it:
You can find the documentation for the columns in the SemanticModelLogs table here and the documentation for the events here. The events for a specific query all have the same value in the OperationId column.
Bonus fact: you can now run queries against Workspace Monitoring using Semantic Link Labs, as documented here, which makes it much easier to do other fun stuff with this data. For example, I can imagine there are ways to visualise DAX query and semantic model refresh events in Python that would make them much easier to analyse, but that’s something for a future blog post.
AI is meant to help us automate boring tasks, and what could be more boring than creating documentation for your Power BI semantic models? It’s such a tedious task that most people don’t bother; there’s also an ecosystem of third party tools that do this job for you, and you can also build your own solution for this using DAX DMVs or the new-ish INFO functions (see here for a good example). That got me wondering: can you use Fabric Data Agents to generate documentation for you? And what’s more, why even generate documentation when you can just ask a Data Agent the questions that you’d need to generate documentation to answer?
To test this idea out, I dug up a semantic model I’ve used for blog posts in the past with a simple star schema and a few measures:
Here are the definitions of the measures:
Units = SUM(Sales[SalesUnits])
Value = SUM(Sales[SalesValue])
Apples Value =
CALCULATE([Value], KEEPFILTERS('Product'[ProductName]="Apples"), USERELATIONSHIP('Product'[ProductID], 'Sales'[ProductID]))
Apples Value Forecast = [Apples Value] * 1.1
After publishing this model I created a Data Agent directly connected to it and found I couldn’t access the metadata I needed from it (I didn’t try too hard, but I don’t think it was possible) so I then created a second Import-mode semantic model, used the first as a data source and created a single table bound to the following DAX query using the INFO.CALCDEPENDENCY function:
EVALUATE INFO.CALCDEPENDENCY()
This second semantic model therefore contained a list of all the DAX calculations, their definitions and the dependencies between them, from the first semantic model. Here’s that table:
I then published this second semantic model, built a Data Agent on it, and used it for all my tests. I also added some instructions to the Data Agent – I freely admit I have no idea whatsoever about prompt engineering but after a bit of trial and error the following instructions did the trick:
You are a Power BI semantic model developer.
The CalcDependencies table contains the output of the DAX INFO.CALCDEPENDENCY() function for another Power BI semantic model. It contains information about the dependencies between the columns, tables, measures, calculated columns, calculated tables and relationships in that model.
Each row in the CalcDependencies table represents a dependency between two objects.
The combination of values in the OBJECT_TYPE, TABLE and OBJECT column uniquely identifies the object which is the source of the dependency.
The combination of values in the REFERENCED_OBJECT_TYPE, REFERENCED_TABLE and _REFERENCED_OBJECT columns uniquely identifies the object which is the target of the identity.
You can join the table to itself multiple times to find chains of dependencies between objects.
When I ask about dependencies, please include direct dependencies and also indirect dependencies, for example where one object has a dependency on another object which in turn has a dependency on the object I am asking about.
The EXPRESSION column contains the DAX definition of an object.
A value of CALC_COLUMN in the OBJECT_TYPE table indicates that the object is a calculated column.
First test: create some documentation. The following prompt:
Generate some documentation for the measures listed in the CalcDependencies table. Describe what each calculation does and how it works.
…generated the following output:
Here is the documentation for the measures listed in the CalcDependencies table, describing what each calculation does and how it works:
Measures
1. Units
Definition:SUM(Sales[SalesUnits])
Description: The Units measure calculates the total number of sales units. It does this by summing the values in the SalesUnits column in the Sales table. This measure is useful for tracking the total quantity of items sold.
2. Value
Definition:SUM(Sales[SalesValue])
Description: The Value measure calculates the total sales value. It achieves this by summing up the values in the SalesValue column in the Sales table. This measure is essential for understanding the total revenue generated from sales.
Description: The Apples Value measure calculates the total sales value specifically for the product “Apples.” It works by applying a filter to include only rows where the ProductName in the Product table is “Apples.” Additionally, it uses a specific relationship between the Product and Sales tables defined by the ProductID columns. This measure helps in analyzing the revenue generated from the sale of apples.
4. Apples Value Forecast
Definition:[Apples Value] * 1.1
Description: The Apples Value Forecast measure estimates the future sales value for the product “Apples” based on a forecasted increase. It multiplies the current Apples Value by 1.1 (indicating a 10% increase). This measure is useful for predicting future revenue from apple sales.
These descriptions provide a clear understanding of what each measure does and how it works, helping anyone who looks at the documentation to comprehend the purpose and functionality of each calculation.
The verdict? More than good enough. To be honest, I was expecting the Data Agent to do a good job here though.
Second test: understand the dependencies between calculations. This is an interesting challenge because you can have chained relationships (ie measures that depend on other measures that depend on other measures etc) which makes the DAX harder to write. Also, you can’t easily understand dependencies by just looking at the output of INFO.CALCDEPENDENCIES – most of the third party tools mentioned above solve this by visualising the data with some kind of network diagram.
The following prompt:
Which measures would be affected if I deleted the SalesValue column on the Sales table?
…generated the following output:
If you delete the SalesValue column from the Sales table, the following measures will be affected:
Measure Name: Value
Definition:SUM(Sales[SalesValue])
Since the Value measure directly depends on the SalesValue column, deleting the SalesValue column will cause the Value measure to break. Additionally, any measures that depend on the Value measure, such as Apples Value and Apples Value Forecast, will also be affected indirectly:
Therefore, deleting the SalesValue column will break the Value, Apples Value, and Apples Value Forecast measures.
The verdict? I’m impressed, this is useful.
Final test: can we generate a visualisation of the dependencies? A few months ago one of my colleagues showed me the Mermaid language for defining diagrams like this and the rather cool Markdown & Mermaid custom visual that allows you to display Mermaid diagrams in Power BI. The following prompt:
Generate an expression in the Mermaid diagramming language that shows all the measures in the model and the dependencies between them
…generated the following response:
Here is an expression in the Mermaid diagramming language that shows all the measures in the model and the dependencies between them:
graph TD
Units --> Units
Value --> Value
Apples_Value --> Value
Apples_Value_Forecast --> Apples_Value
Units["Units"]
Value["Value"]
Apples_Value["Apples Value"]
Apples_Value_Forecast["Apples Value Forecast"]
Explanation:
Units is a measure with no dependencies on other measures.
Value is a base measure.
Apples Value depends on the Value measure.
Apples Value Forecast depends on the Apples Value measure, which in turn depends on the Value measure.
This Mermaid diagram visually represents all the measures in the model and the dependencies between them using arrows.
Pasting the expression in the response into the Markdown & Mermaid custom visual gave me the following:
I wasn’t particularly happy with those messy lines linking the Value and Units measures to themselves, but overall it does the job pretty well.
This blog post is already long enough and while these are simple tests, I think I’ve proved my point that Fabric Data Agents can be very useful when it comes to understanding the structure of a semantic model and the calculations in it. Bringing in data from other DAX INFO functions and tuning the AI Instructions would allow you to answer even more complex questions. I can also imagine a solution using Semantic Link Labs that not only dumps the data from DAX INFO functions for all your semantic models into a table in OneLake for consumption by a Data Agent, but also then uses the Data Agent SDK to automatically generate markdown documentation for those models, complete with diagrams. The possibilities are endless!
It’s time for another one of my occasional posts about free books that I’ve been given that you might be interested to check out. The usual disclaimers apply: these aren’t impartial, detailed reviews and the links contain an Amazon UK affiliate code so I get a kickback if you buy a copy.
Chandeep is of course famous on YouTube from the Goodly Power BI channel; I’ve met him at conferences and he’s just as friendly and passionate in real life. That passion shows through in this book. It’s aimed at people who already know Power Query and want to learn M and while it does a great job of that, I think anyone who writes M on a regular basis would also benefit from reading it. It’s packed with practical examples, well-written, everything is clearly explained and it covers more recent additions to the language that older books might not talk about. It’s focused on the M language and doesn’t cover topics like performance tuning but I think that focus is a good thing. Highly recommended for anyone serious about Power Query.
It looks like the Packt marketing machine has gone into overdrive for this title because I see both Greg Lowe and Sandeep Pawar have already published their reviews, and I agree with their sentiments. Power BI (and even more so Fabric) is complicated and so there’s a huge demand for guidance around what all of the components do and how to put them together to create a solution. The team I work on at Microsoft, the Fabric CAT team, has a published guidance documentation here and other people have written books, blog posts and white papers addressing the same problem. This book is certainly a very useful addition to the existing literature. It covers newer topics like Power BI Copilot and some topics that are rarely if ever mentioned elsewhere, such as Power BI’s integration with Purview. As the other reviewers have mentioned, books like this always suffer from changes to the product making them out of date very quickly but that’s unavoidable. Also, being written by a Microsoft employee (and this is something I can relate to), it’s not very opinionated and doesn’t tell you which features of the product are good and which ones should be avoided. All in all, pretty good though.
The series of blog posts I wrote last year on semantic model memory usage, in particular this post on the query memory limit and the “This query uses more memory than the configured limit” error in Power BI, gets a lot of traffic. Since writing that post on the query memory limit I’ve written a few follow-ups on common mistakes that lead to increased query memory usage, such as this one on measures that never return a blank. Today’s post is sort of in that series but it isn’t about a design mistake – it’s just to point out that distinct count measures can be surprisingly memory-hungry.
To illustrate this I built a semantic model consisting of a single table with two columns and 99,999,000 rows, published it and ensured the Large Semantic Model format was enabled:
I created two measures:
Number Of Rows = COUNTROWS('MyTable')
Distinct Customers = DISTINCTCOUNT(MyTable[CustomerID])
Here’s what the model metrics looked like in DAX Studio:
The total model size in memory was 255MB.
I then ran the following DAX query to get the number of rows in the table for each of the 1800 dates in the Date column:
EVALUATE
SUMMARIZECOLUMNS(
MyTable[Date],
"Number Of Rows", [Number Of Rows]
)
The approximatePeakMemConsumptionKB metric for this query was 800325KB – so a lot more than the previous query. In fact even though this model was well under the 1GB size limit for a model not in Premium capacity, the query here used a lot more memory (782MB) than the size of the model itself in memory and it came close to the 1GB limit on the amount of memory a query can consume when the model is not in Premium capacity.
Is there something wrong here? Can the query or model be tuned to reduce memory usage? Not really, no – distinct count queries are almost always more memory intensive than other types of measures. I tested a number of different things such as forcing the use of hash encoding on the CustomerID column, partitioning (Phil Seamark suggested creating one partition for each of the 1800 dates and actually that did reduce memory consumption but it also made the queries extremely slow), changing the ordering of the source data to change how well each column was compressed, calculating the distinct count using the SUMX method, and nothing resulted in lower query memory usage.
What I did find for the model above was that the number of rows returned by the query influenced the memory consumption of the query. So reducing the number of dates returned on rows in my DAX query from 1800 to 366 resulted in approximatePeakMemConsumptionKB going down to 200278KB. So if you’re running into memory errors when running queries with distinct count measures the first thing you should ask yourself is whether you need to show so many distinct counts: I recently ran into this problem with a customer that wanted to plot a line chart of distinct values with dates on the x axis, and we solved the problem by only plotting one day per week for the time period shown on the chart instead of every date. The chart looked almost identical, the DAX query was a lot faster and the memory usage of the DAX query was a lot lower. Distinct count measures combined with table visuals with lots of rows can be dangerous.
The other thing you can do is see if you can remodel your data to turn a distinct count into a count because, as shown above, counts are a lot faster and memory efficient than distinct counts. For example, if you have a fact table containing line items for orders and you need to find the distinct count of order ids, then consider creating a second fact table at the order granularity so you can count the number of rows in it to find the number of distinct orders. This may increase the size of your model but it should certainly reduce your query memory consumption for many queries because you won’t need to do a distinct count.
Since the November 2024 Power BI release blog post announced that queries sent to Snowflake by Power BI include a query tag I’ve had a lot of questions from people who couldn’t see this happening or wanted to know what the query tags contained, so in this blog I thought I would outline the current status.
The query tagging feature for the Power BI Snowflake connector actually didn’t get released in November 2024 and even now, in April 2025, it’s only available for DirectQuery connections and Import mode refreshes that use the V1.0 connector (the V2.0 connector will support query tags soon). Here’s an example of what a query tag looks like for a SQL query generated by Power BI from a DirectQuery semantic model:
At the time of writing only SQL queries sent from the Power BI Service contain query tags, not those sent from Power BI Desktop. Also there is no way to customise the contents and unlike SQL queries sent to SQL Server-related sources there is no information on the report or visual that generated the SQL query. In the future some of these limitations may go away.
Now that Fabric Data Agents (what used to be called AI Skills) can use Power BI semantic models as a data source I’ve been spending some time playing around with them, and while I was doing that I realised something – maybe something obvious, but I think still worth writing about. It’s that there are a lot of amazing things you can do in DAX that rarely get done because of the constraints of exposing semantic models through a Power BI report, and because Data Agents generate DAX queries they unlock that hitherto untapped potential for the first time. Up until now I’ve assumed that natural language querying of data in Power BI was something only relatively low-skilled end users (the kind of people who can’t build their own Power BI reports and who struggle with Excel PivotTables) would benefit from; now I think it’s something that will also benefit highly-skilled Power BI data analysts as well. That’s a somewhat vague statement, I know, so let me explain what I mean with an example.
Consider the following semantic model:
There are two dimension tables, Customer and Product, and a fact table called Sales with one measure defined as follows:
Count Of Sales = COUNTROWS('Sales')
There’s one row in the fact table for each sale of a Product to a Customer. Here’s all the data dumped to a table:
So, very simple indeed. Even so there are some common questions that an analyst might want to ask about this data that aren’t easy to answer without some extra measures or modelling – and if you don’t have the skills or time to do this, you’re in trouble. One example is basket analysis type questions like this: which customers bought Apples and also bought Lemons? You can’t easily answer this question with the model as it is in a Power BI report; what you’d need to do is create a disconnected copy of the Product dimension table so that a user can select Apples on the original Product dimension table and select Lemons on this new dimension, and then you’d need to write some DAX to find the customers who bought Apples and Lemons. All very doable but, like I said, needing changes to the model and strong DAX skills.
I published my semantic model to the Service and created a Data Agent that used that model as a source. I added two instructions to the Data Agent:
Always show results as a table, never as bullet points
You can tell customers have bought a product when the Count of Sales measure is greater than 0
The first instruction I added because I got irritated by the way Data Agent shows the results with bullet points rather than as a table. The second probably wasn’t necessary because in most cases Data Agent knew that the Sales table represented a sale of a Product to a Customer, but I added it after one incorrect response just to make that completely clear.
I then asked the Data Agent the following question:
Show me customers who bought apples and who also bought lemons
And I got the correct response:
In this case it solved the problem in two steps, writing a DAX query to get the customers who bought lemons and writing another DAX query to get the customers who bought apples and finding the intersection itself:
At other times I’ve seen it solve the problem more elegantly in a single query and finding the customers who bought apples and lemons using the DAX Intersect() function.
I then asked a similar question:
For customers who bought apples, which other products did they buy?
And again, I got the correct answer:
In this case it ran five separate DAX queries, one for each customer, which I’m not thrilled about but again at other times it solved the problem in a single DAX query more elegantly.
Next I tried to do some ABC analysis:
Group customers into two categories: one that contains all the customers with just one sale, and one that contains all the customers with more than one sale. Show the total count of sales for both categories but do not show individual customer names.
And again I got the correct answer:
I could go on but this post is long enough already. I did get incorrect answers for some prompts and also there were some cases where the Data Agent asked for more details or a simpler question – but that’s what you’d expect. I was pleasantly surprised at how well it worked, especially since I don’t have any previous experience with using AI for data analysis, crafting prompts or anything like that. No complex configuration was required and I didn’t supply any example DAX queries (in fact Data Agents don’t allow you to provide example queries for semantic models yet) or anything like that. What does this all mean though?
I’m not going to argue that your average end user is going to start doing advanced data analysis with semantic models using Data Agents. The results were impressive and while I think Data Agents (and Copilot for that matter) do a pretty good job with simpler problems, I wouldn’t want anyone to blindly trust the results for more advanced problems like these. However if you’re a data analyst who is already competent with DAX and is aware that they always need to verify the results they get from Data Agent, I think this kind of DAX vibe-coding has a lot of value. Imagine you’re a data analyst and you’re asked that question about which products customers who bought apples also bought. You could search the web, probably find this article by the Italians, get scared, spend a few hours digesting it, create a new semantic model with all the extra tables and measures you need, and then finally get the answer you want. Maybe you could try to write a DAX query from scratch that you can run in DAX Studio or DAX Query View, but that requires more skill because no-one blogs about solving problems like this by writing DAX queries. Or you could ask a Data Agent, check the DAX query it spits out to make sure it does what you want, and get your answer much, much faster and easier. I know which option I’d choose.
To finish, let me answer a few likely questions:
Why are you doing this with Fabric Data Agents and not Power BI Copilot?
At the time of writing Data Agents, the Power BI Copilot that you access via the side pane in a report and Power BI Copilot in DAX Query View all have slightly different capabilities. Power BI Copilot in the side pane (what most people think of as Power BI Copilot) couldn’t answer any of these questions when I asked them but I didn’t expect it to because even though it can now create calculations it can still only answer questions that can be answered as a Power BI visual. Copilot in DAX Query View is actually very closely related to the Data Agent’s natural language-to-DAX functionality (in fact at the moment it can see and use more model metadata than Data Agent) and unsurprisingly it did a lot better but the results were still not as good as Data Agent. Expect these differences to go away over time and everything I say here about Data Agents to be equally applicable to Power BI Copilot.
This isn’t anything new or exciting – I see people posting about using AI for data analysis all the time on LinkedIn, Twitter etc. What’s different?
Fair point. I see this type of content all the time too (for example in the Microsoft data community Brian Julius and my colleague Mim always have interesting things to say on this subject) and I was excited to read the recent announcement about Analyst agent in M365 Copilot. But typically people are talking about taking raw data and analysing it in Python or generating SQL queries. What if your data is already in Power BI? If so then DAX is the natural way of analysing it. More importantly there are many advantages to using AI to analyse data via a semantic model: all the joins are predefined, there’s a lot of other rich metadata to improve results, plus all those handy DAX calculations (and one day DAX UDFs) that you’ve defined. You’re much more likely to get reliable results when using AI on top of a semantic model compared to something that generates Python or SQL because a lot more of the hard work has been done in advance.
Is this going to replace Power BI reports?
No, I don’t think this kind of conversational BI is going to replace Power BI reports, paginated reports, Analyze in Excel or any of the other existing ways of interacting with data in Power BI. I think it will be a new way of analysing data in Power BI. And to restate the point I’ve been trying to make in this post: conversational BI will not only empower low-skilled end users, it will also empower data analysts, who may not feel they are true “data scientists” but who do have strong Power BI and DAX skills, to solve more advanced problems like basket analysis or ABC analysis much more easily.
Although it was announced in this blog post on the Microsoft 365 Insider blog recently, you might have missed the news that Excel Copilot can now generate Power Query queries. There are limitations for now: it can only be used to connect to other Excel files stored in OneDrive or SharePoint and it can’t do any transformations in the queries it creates, but it’s still exciting news nonetheless. Well the kind of news I get excited by at least.
Since the announcement blog post didn’t give many details of how it works let’s see an example of it in action. Let’s say you have an Excel workbook called SprocketsWidgetsSales.xlsx that contains a table of data showing sales of sprockets and widgets – the products your company sells – by country:
Now let’s say you create a new, blank workbook and open the Copilot pane. Entering the prompt:
Search for data on sales of sprockets and widgets
…gives you the data from the first workbook in the response:
At the bottom you can see a citation reference pointing to the workbook containing the source data and clicking that reference opens that workbook in Excel Online, but we don’t want to do that, we want to load the data into the current workbook using Power Query. Clicking on “Show tables to import” shows a preview of all the Excel tables (in this case there’s only one) in the workbook:
Expanding “Show import query” shows the M code for the Power Query query it can generate:
And clicking “Import to new sheet” creates that Power Query query and runs it:
You can see the Power Query query it creates in the Queries & Connections pane and edit it in the Power Query Editor like any other query:
Here’s the output of the query in a table on a new worksheet:
Of course now you have the table of data on your worksheet you can do other things like:
chart this data by country and product
…or ask questions like:
which country had the lowest sales of sprockets?
…and other things that you’d expect Copilot to be able to do. But the key thing is that Copilot is can now generate Power Query queries! I’m looking forward to see how this feature improves in the future.
If you’re working with slow data sources in Power BI/Fabric dataflows then you’re probably aware that validation (for Gen1 dataflows) or publishing (for Gen2 dataflows) them can sometimes take a long time. If you’re working with very slow data sources then you may run into the 10 minute timeout on validation/publishing that is documented here. For a Gen1 dataflow you’ll see the following error message if you try to save your dataflow and validation takes more than 10 minutes:
Failed to analyze issues in the query
For a Gen2 Dataflow, where you can save the Dataflow and publishing takes place in the background, you’ll see the following error in your workspace:
Dataflow publish failed
Apart from tuning your data source and tuning your queries, what can you do about this? Well one of the things that happens when you publish a dataflow is that it works out the columns returned, and the data types of those columns, for all of the queries in the dataflow. It does this by trying to run the queries until they return data by applying a top 0 row filter to them; if you can make that faster then validation/publishing will be faster. Obviously query folding is important here because that top 0 filter should fold, as are more obscure, source-specific settings like this one for ODBC sources. However, there is another trick that you can use if you are happy writing some moderately complicated M code – the trick I blogged about here for making Power Query in Power BI Desktop faster.
Let’s see an example with Dataflows Gen2. Conside the following M code which returns a table with three columns and is deliberately written to take 11 minutes and 1 second to return (see this post for more details on how to create artificially slow Power Query queries).
let
Source = Function.InvokeAfter(
() =>
#table(
type table
[
#"Number Column"=number,
#"Text Column"=text,
#"Date Column"=date
],
{
{1,"Hello",#date(2016,1,1)},
{2,"World",#date(2017,12,12)}
}
)
,
#duration(0, 0, 11, 1)
)
in
Source
As you would expect, trying to publish a Gen1 or Gen2 dataflow that uses this query will fail because it takes more than 10 minutes before it returns any rows. However in this case – as in most cases – you know what columns the query returns so it’s possible to use the Table.View M function to intercept the zero-row filter applied during validation/publishing and return a table with no rows in and the columns that the query above returns. You can do this by adding two extra steps in the M code like so:
The first step added here, called TableTypeToReturn, defines the columns and data types of the table returned by the query; if you use this technique yourself, you will need to alter it so it returns the columns and data types of your query. You can read more about #table and table types here and I have a function that will automatically generate this code from an existing query for you here. The second step, called OverrideZeroRowFilter, looks for situations where a Top N filter is being applied and if N=0 returns a table of the type defined in the previous step with zero rows. For a more detailed explanation see that original blog post.
This new version of the query validates/publishes immediately, although it still takes 11 minutes and 1 second to refresh. Of course if you use this technique and then change your query so that different columns or data types are returned you have to update the extra code every time, which can be fiddly, but if you’re running into a timeout then you don’t have any choice and even if validation/publishing is slow it’s probably worth the extra effort.
