WiseOwl Training - Established 1992 Wise Owl Training

Established May 1992
30 years in business
Wise Owl Training
30 years in business
See 516 reviews for our classroom and online training
Power BI Desktop update for May 2021
Part six of a six-part series of blogs

May's update to Power BI Desktop is a big one, encompassing new ways of viewing models, tables and fields, a new text box, smart narratives (to explain data) and a new anomaly detection feature (to explain exceptions).

  1. Power BI Desktop update for May 2021
  2. A new model view
  3. Standardised table and field lists across Power BI views
  4. New text boxes
  5. Smart Narratives
  6. Automatic detection of anomalies (this blog)

We've been creating our idiosyncratic monthly blogs on Power BI updates since November 2016, and also deliver online and classroom Power BI courses.

Posted by Andy Brown on 19 May 2021

You need a minimum screen resolution of about 700 pixels width to see our blogs. This is because they contain diagrams and tables which would not be viewable easily on a mobile phone or small laptop. Please use a larger tablet, notebook or desktop computer, or change your screen resolution settings.

Automatic detection of anomalies

This is a deceptively large change, hidden away on the Analytics card for a line chart. 

Note that this feature has many limitations, listed in full here (the main ones are that you must have a line chart with a time axis and at least 4 data points, and you can't show forecast, max/min or percentage lines).

What anomalies are and how they work

Suppose you have some time-series data:

Line chart

A line chart showing sales by day across 2020.

You can apply anomalies to highlight questionable data points (any other readers remember the Primeval TV series at this point?):


The anomalies are pretty much where you'd expect them to be.


You can then click on an anomaly to see an explanation of why it might be occurring:

Anomaly explanation

A summary of the anomaly, followed by a list of possible explanation (not shown here).

Full disclosure: I can't get the list of explanations to appear above (instead I always get the message An error occurred while running analysis).  I suspect this is user or data error, as no one else seems to have experienced this.

Starting to show anomalies

To show anomalies, first select a line chart visual with a time axis:

A line chart visual

This visual shows total sales by date for 2020.


On the Analytics card, expand the Anomalies card:

Analytics and anomalies

Select the angry Pac-Man symbol, then expand Find anomalies at the bottom of the list of properties.


Choose to add anomaly markers:

Adding anomalies

Click on the Add link to show any anomalies.


The lower the sensitivity, the less "anomalies" you'll detect:

High sensitivity Low sensitivity
The default 70% sensitivity Applying 40% sensitivity

Formatting anomalies

You can format your anomalies (and the area showing where normally data would be expected to occur) in a fairly obvious way.  For example, these settings:

Anomaly formatting settings

Some fairly tasteless formatting!


Would give this line chart:

Formatted line chart

The green expected area was a particularly bad idea.

Explaining anomalies

If you want to see more details about a particular anomaly, select it:

Selecting an anomaly

Here I've gone for the biggest anomaly - the figure for 15th April 2020, which is over twice what you would expect the highest value to be given the other data.

This displays an Anomalies pane;

Anomalies pane

An explanation of what's happening, followed by a list of possible explanations.

I'd love at this point to show you a list of the explanations, but instead this is what I see:

Error doing analysis

The message I get at this point.

Instead, I'll switch to Microsoft's explanation of how to work with anomalies at this point, to show the sort of thing you should get:

Explanations list

You can click on each explanation to see a chart explaining it in more detail.


Adding explanatory fields

Explanations consider all possible fields, but you might want to limit the scope of your investigation (for our example, you might theorise that the anomaly will be explicable by reference to the town and/or product data):

Explain by fields

Here I've dragged the TownName and ProductName fields onto the Explain by card, to try to explain my anomalies in those terms only.


All in all, an impressive feature!  You can see a summary of the algorithm used here, but it uses Fourier transforms, so you're going to need at least a degree in maths to understand it (I don't!).

This blog has 0 threads Add post