Handling Transformations

For situations where the data source response isn't presented in a format that Power BI can consume directly, Power Query can be used to perform a series of transformations.

Static Transformations

In most cases, the data is presented in a consistent way by the data source: column names, data types, and hierarchical structure are consistent for a given endpoint. In this situation it's appropriate to always apply the same set of transformations to get the data in a format acceptable to Power BI.

An example of static transformation can be found in the TripPin Part 2 - Data Connector for a REST Service tutorial when the data source is treated as a standard REST service:

    Source = TripPin.Feed("https://services.odata.org/v4/TripPinService/Airlines"),
    value = Source[value],
    toTable = Table.FromList(value, Splitter.SplitByNothing(), null, null, ExtraValues.Error),
    expand = Table.ExpandRecordColumn(toTable, "Column1", {"AirlineCode", "Name"}, {"AirlineCode", "Name"})

The transformations in this example are:

  1. Source is a Record returned from a call to TripPin.Feed(...).
  2. You pull the value from one of Source's key-value pairs. The name of the key is value, and you store the result in a variable called value.
  3. value is a list, which you convert to a table. Each element in value becomes a row in the table, which you can call toTable.
  4. Each element in value is itself a Record. toTable has all of these in a single column: "Column1". This step pulls all data with key "AirlineCode" into a column called "AirlineCode" and all data with key "Name" into a column called "Name", for each row in toTable. "Column1" is replaced by these two new columns.

At the end of the day you're left with data in a simple tabular format that Power BI can consume and easily render:

Data in tabular form.

It's important to note that a sequence of static transformations of this specificity are only applicable to a single endpoint. In the example above, this sequence of transformations will only work if "AirlineCode" and "Name" exist in the REST endpoint response, since they are hard-coded into the M code. Thus, this sequence of transformations may not work if you try to hit the /Event endpoint.

This high level of specificity may be necessary for pushing data to a navigation table, but for more general data access functions it's recommended that you only perform transformations that are appropriate for all endpoints.


Be sure to test transformations under a variety of data circumstances. If the user doesn't have any data at the /airlines endpoint, do your transformations result in an empty table with the correct schema? Or is an error encountered during evaluation? See TripPin Part 7: Advanced Schema with M Types for a discussion on unit testing.

Dynamic Transformations

More complex logic is sometimes needed to convert API responses into stable and consistent forms appropriate for Power BI data models.

Inconsistent API Responses

Basic M control flow (if statements, HTTP status codes, try...catch blocks, and so on) are typically sufficient to handle situations where there are a handful of ways in which the API responds.

Determining Schema On-The-Fly

Some APIs are designed such that multiple pieces of information must be combined to get the correct tabular format. Consider Smartsheet's /sheets endpoint response, which contains an array of column names and an array of data rows. The Smartsheet Connector is able to parse this response in the following way:

raw = Web.Contents(...),
columns = raw[columns],
columnTitles = List.Transform(columns, each [title]),
columnTitlesWithRowNumber = List.InsertRange(columnTitles, 0, {"RowNumber"}),
RowAsList = (row) =>
        listOfCells = row[cells],
        cellValuesList = List.Transform(listOfCells, each if Record.HasFields(_, "value") then [value]
                else null),
        rowNumberFirst = List.InsertRange(cellValuesList, 0, {row[rowNumber]})

listOfRows = List.Transform(raw[rows], each RowAsList(_)),
result = Table.FromRows(listOfRows, columnTitlesWithRowNumber)
  1. First deal with column header information. You can pull the title record of each column into a List, prepending with a RowNumber column that you know will always be represented as this first column.
  2. Next you can define a function that allows you to parse a row into a List of cell values. You can again prepend rowNumber information.
  3. Apply your RowAsList() function to each of the rows returned in the API response.
  4. Convert the List to a table, specifying the column headers.