Microsoft Power Automate Interview Questions & Answers

Flows are the individual automation files or processes. Subflows organise smaller tasks or problem statements within a main flow, making the automation easier to develop, debug, and maintain.

Subflows divide a flow into small chunks that are easier to develop, debug, and maintain. As a best practice, the main flow holds high-level steps and calls the subflows, which are defined on separate pages.

Run Subflow invokes a subflow that exists inside the current flow, called wherever the process logic requires. Run Desktop Flow invokes a separate desktop flow that you created or that was shared with you.

The Run Desktop Flow action, which lets one desktop flow call and execute another.

A Region groups multiple actions or steps together in a logical manner without affecting the flow's execution. For example, if a process has multiple stages, you can group the steps for each stage into a separate region.

A Label marks a point in the flow that a Go to action can jump to, letting you alter the normal top-to-bottom execution based on process logic.

Use a Label to mark the target step and a Go to action to jump back to that label, redirecting execution as the logic requires.

Modules and actions can be classified as Business, Non-business, or Blocked, which is used by data loss prevention policies.

The Set variable action stores and manipulates data, for example adding two numbers together and storing the result for later use in the flow.

A String stores text and is written in quotes, while an Integer stores whole numbers used for calculations. Mixing them without conversion causes errors, for example concatenation instead of addition.

Use the Set variable action with an expression, for example %Var1 + Var2%, wrapping the arithmetic in percentage signs so PAD evaluates it as a calculation.

Use the Convert text to number action so arithmetic can be performed, because PAD treats text input as strings by default.

Insert the Convert text to number action before the calculation, so the text values are treated as numbers and added rather than joined.

Use the Set variable action and enter %True% or %False% depending on the value you want to store.

Sensitive variables hide a value so it is not visible during debugging and is not logged when desktop flows are triggered from the cloud portal. They protect data like passwords and keys.

Use the notation %VariableName.length% to get the number of characters or items.

Open the action and turn off the produced-variable option for any output you do not need, keeping the variables pane clean.

The Generate random number action, where you set the minimum and maximum bounds of the range.

Use array notation with the Set variable action: multiple single-dimension arrays separated by commas and enclosed in curly brackets, which builds the datatable without the Create datatable action.

Use the notation %VariableName[RowNumber][ColumnNumber]% to address a specific cell by its row and column index.

Three: Simple loops (Loop) that iterate a set number of times, Loop condition that iterates while a condition holds, and For each loops that iterate through a list or datatable.

A For each loop, which iterates over every item or row in the collection.

Use a Loop for a fixed number of iterations, a Loop condition to repeat while a condition is true, or a For each to process every item in a list, choosing the type that matches the data and stopping logic.

Initialise a counter variable before the loop, use a Loop condition that checks the counter value, and increment the counter at the end of each iteration.

Set a counter to 1, use a Loop condition that runs while the counter is less than or equal to 4, take a screenshot inside, and increment the counter each iteration.

An If action evaluates a condition and runs the enclosed actions only when it is true, letting the flow make decisions such as branching on a value or status.

Nested conditions are If actions placed inside other If actions, letting you check multiple criteria in sequence, for example checking a status only after confirming a record exists.

Use a While or Loop condition when you repeat based on a condition being met (for example, until a value is found), and a For Each when you iterate over a known collection of items.

The loop may never start, run infinitely, or skip iterations, because the condition check depends on the counter. Always initialise the counter before the loop and increment it consistently inside.

Use the Get first free row action or the row count of the read data to set the loop's end value, so it processes every populated row and stops at the last one.

Start the loop index at 2 (skipping the header row) and iterate to the last populated row, addressing each row by its index.

Use the Display message action, drag it into the workspace, and enter the text you want shown in the message box.

Message box title, the message to display, the message box icon (Information, Warning, Error, Question), the buttons shown, the default button, and whether the box stays on top.

Use the Display input dialogue action with a title, prompt message, and input type, which pauses the flow and stores what the user enters in a variable.

It forces the message box to stay in front of all other windows, ensuring the user sees and responds to it instead of it being hidden behind the automated application.

Use %[Var1,Var2,Var3]% or %['Col1','Col2','Col3']% array syntax in a single Write to Excel worksheet action to populate multiple adjacent columns at once.

Use a single Write to Excel worksheet action with array syntax %[Val1,Val2,Val3]% to write all values in one operation, rather than separate actions per column, which prevents code bloat.

Enable the First line contains column names option in the Read from Excel worksheet action so PAD maps the column headers correctly.

Use the Resize columns/rows action in the Excel category to autofit column widths.

Use the Attach to running Excel action to connect to the already-open workbook.

It returns the first empty row or column, so the bot knows where to append new data without overwriting existing records.

Read, write, delete, execute macros, and apply Excel formatting, among other operations available in the Excel action group.

Treating Excel as a database lets you run SQL queries against it for faster filtering, joining, and bulk operations on large datasets, which is more efficient than row-by-row Excel actions.

The worksheet name, referenced with square brackets and a dollar sign (for example [Sheet1$]), acts as the table name in the SQL query against the Excel file.

Split Text divides a string by a delimiter and returns a list of items, whereas Crop Text extracts a single portion of text between two specified flags.

Use Split Text to break a delimited string, such as comma-separated values, into individual list items for further processing.

The Crop Text action, which extracts the portion of text between two markers.

PAD uses zero-based indexing: the first item is at index 0, the second at index 1, and so on.

Address items by their zero-based index, for example %ListVariable[0]% to retrieve the first item from the resulting list.

Use the expression %List.Count%, where List is the name of your list variable.

Use the Set variable action with array syntax, for example %['FirstItem','SecondItem','ThirdItem']%, to define the list directly.

Use the Recognize entities in text action, which detects and extracts many such entity types from a block of text.

Terminate Process forcibly ends a running application process and works even if the browser is not open, while Close Browser only works on an already-open browser instance and errors if none exists. Use Terminate Process in the Init state to safely reset.

No. Use Terminate Process in the Init state instead, because the browser may not be open at that point and Close Browser would throw an error.

Instead of an exact match, use Contains or Starts with operators in the selector configuration so it stays valid even when attribute values change dynamically.

Edit the selector and replace the dynamic part with a Contains or Starts with operator so it matches the element regardless of changing values.

Remove the hard-coded dynamic attribute (such as the exact number or a changing ID) and replace it with a Contains or Starts with rule, so the selector matches by a stable part of the element.

The button may not exist on later iterations, or the selector matched only the first element. Add an If element exists check before the click and use a stable selector to handle each iteration.

Add an If UI element exists conditional check before the click action, so the bot clicks only when the element is present, preventing runtime errors.

Passing parameters directly in the URL skips multiple UI navigation steps, making the automation faster and more reliable.

Use the Go to webpage action with the Navigate to URL option.

Not always. You may need to configure the extraction to include the header row or map columns manually, depending on the page structure.

Configure a Pager in the Extract data action so PAD follows the next-page link and collects data across all pages into one dataset.

A Pager is the next-page navigation element on a website. You select it during data extraction so PAD automatically moves through paginated results and gathers all records.

Pagers are web elements that let you navigate between multiple pages of results, used by the data-extraction action to collect data across pages.

PAD supports Internet Explorer, Firefox, Chrome, and Edge.

Three ways: by title, by URL, and by using the foreground window.

Buttons, radio buttons, checkboxes, dropdown lists, and tables or grids, among other interface controls.

Use the Send physical click option in the click action, which sends a hardware-level click when the emulated click is not acknowledged by the webpage.

You can extract page-level properties such as the title, URL, and other document attributes exposed by the Get details of webpage action.

The Recorder cannot capture logic such as loops or conditions, and it is not suitable for Excel automation where dynamic data handling is required.

No. The Recorder only captures UI interactions and cannot record conditional branches, loops, or any decision-making logic, which must be added manually.

Hold Ctrl and left-click on the element to select it.

The icon at the start of the element's name indicates its type, differentiating web elements from desktop application elements.

Three ways: by Window UI element, by Window instance or handle, and by title or class.

Wait, Wait for process, Wait for file, Wait for window content, Wait for image, Wait for window, Wait for mouse, Wait for shortcut key, and Wait for CMD session.

Actions such as Print document, Get default printer, Set default printer, Show desktop, Lock workstation, and Log off user.

Use the Set clipboard text action to place text on the clipboard, then paste it into the target application.

Use the Set key state action to turn on the Caps Lock key, then provide the input so it is entered in uppercase.

Add to datetime, Subtract dates, and Get current date and time, among other date and time operations.

Use the Convert datetime to text action with a custom format string, or Convert text to datetime to parse a string into a date, specifying the exact format pattern.

Uppercase MM represents the month while lowercase mm represents minutes. Using the wrong case produces an incorrect value, for example minutes where you expected the month.

Specify the exact format string dd/MM/yyyy in the date conversion action, using uppercase MM for month, so the output matches the UK convention regardless of system locale.

PAD provides native CMD session actions (Open, Read from, and Write to CMD session) that run commands without opening a visible command prompt, which is more reliable than sending keystrokes to a terminal window.

Use the Read from CMD session and Write to CMD session actions to perform read and write operations without opening a visible command prompt window.

Native CMD actions run commands directly and capture output reliably, avoiding the timing and focus problems that come with sending simulated keystrokes to a terminal.

Set the file filter (for example *.pdf) in the Get files in folder action so it returns only matching file types.

Get the files in the folder, loop through each, compare each file's creation or modified date against the current date minus N days, and delete the file when it is older than the threshold.

When enabled, each Write text to file action adds the content on a new line rather than appending to the existing line, keeping records separated by line breaks.

Include a counter variable in the filename, such as Screenshot_%Counter%.png, and increment the counter each iteration so every file has a unique name.

Extract text from PDF, Extract tables from PDF, Extract images from PDF, Extract PDF pages to a new PDF, and Merge PDF files.

Document automation is the process of extracting data from documents such as invoices, purchase orders, bank statements, contracts, or agreements, and using it in a flow.

A Native PDF contains selectable text that can be read directly, while a Scanned Image PDF is a picture of text that requires OCR to be read.

The Extract text from PDF action reads text content from the document.

Extract the full text, then use text actions such as Crop Text between known flags or a regular expression to isolate the specific field value.

It tells the action to preserve the layout and structure (such as tables and columns) so extracted data keeps its positional meaning instead of becoming a flat text block.

Extracted values frequently carry leading or trailing spaces and line breaks. Trim removes that whitespace so the value is clean for comparison or storage.

Use Get files in folder filtered to PDFs, then loop through each file, extracting and processing its content one by one.

OCR (Optical Character Recognition) converts images of typed, handwritten, or printed text into machine text. PAD includes the Windows and Tesseract engines for free, and can integrate paid engines like Abbyy or OmniPage.

The archive file path and the destination folder where the extracted contents should be placed.

Use UI automation or window actions to handle the native Save As dialog, since it is an operating-system window rather than a web element.

The browser must be configured to ask where to save each file (disabling automatic downloads to the default folder), so PAD can control the Save As path.

Exception handling is the practice of anticipating and managing errors so a flow can recover or fail gracefully instead of crashing, which is essential for reliable automation.

Unattended bots run without a person watching, so an unhandled error stops the entire process silently. Proper exception handling lets the bot log the issue, skip bad records, or exit cleanly.

On block error wraps a group of actions so one error handler covers the whole block, while individual on-error handling is set per action. Block-level handling is efficient for a sequence, action-level for pinpoint control.

The On block error action, which groups multiple steps together and applies error handling to them.

Two ways: at the block level, covering many actions at once, and at the individual action level.

Use the Get last error action to retrieve details of the most recent error for logging or decision-making.

Use the Get last error action, which returns the details of the most recent error.

Two: design-time errors, tied to action configuration and caught during development (such as an empty mandatory field), and run-time errors or exceptions that occur during execution (such as an invalid file path).

Type (error or warning), Description of the issue, the Subflow containing the action, and the Line number of the action that caused it.

Go to Advanced in the error handling settings, select the exception type, click New rule, and specify the subflow to run for that error.

Wrap the record-processing actions in a block with On block error set to continue flow run, so a bad record is logged or skipped and the loop moves to the next item.

Use error handling in the Init block to detect the failure, log it, close any open resources, and terminate the flow cleanly before the main process begins.

Retrieve email messages from Outlook, Send email messages through Outlook, Process email messages in Outlook, and Save Outlook email messages.

Use the Retrieve or Process email action, specify the mailbox folder, and set filters such as sender, subject, or unread status to return only matching messages.

Read the email body as text, then use text actions such as Crop Text between known markers or a regular expression to isolate the code.

Use the Send email (SMTP) action configured with Gmail's SMTP server, an app password, and the appropriate port to send log messages.

Use smtp.gmail.com as the server with port 587 and TLS (or 465 with SSL), authenticating with the Gmail address and an app password rather than the normal account password.

Use the Open SQL connection action and provide a connection string, or select the database provider and enter the server name, username, password, and database name.

The database provider, server name, database name, and authentication details (username and password, or integrated security).

The provider, server or data source, initial catalog (database name), and credentials, assembled into the connection string the Open SQL connection action uses.

Closing the connection releases the database resource and avoids leaving open connections that can exhaust the connection pool or lock resources.

SELECT reads and returns rows matching a condition without changing data, while UPDATE modifies existing rows. In automation, SELECT fetches records to process and UPDATE writes results back.

Concatenate the variable into the SQL text using percentage-sign notation, for example including %myVariable% in the query string so its value is inserted before execution.

Open a SQL connection, read the Excel data into a datatable, loop through the rows, and run an INSERT statement for each row (or a bulk insert), then close the connection. Column mapping must match the table schema.

Run an ALTER TABLE statement through the Execute SQL statement action, specifying the new column name and data type.

The action returns a SQL connection variable that holds the open connection, which you pass to subsequent Execute SQL statement and Close SQL connection actions.

A config file stores settings such as file paths, URLs, credentials references, and parameters outside the flow, so values can be changed without editing and republishing the automation.

Use input and output arguments (variables): the parent flow passes values in when calling the child flow, and the child returns values through its output arguments.

When one desktop flow runs another, input and output variables exchange values between the parent and child flows.

Flow A passes values into Flow B through Flow B's input arguments when it calls it. Flow B processes them and returns results through its output arguments, which Flow A reads after the call completes.

Use a loop with a short wait, reading the value each iteration, and store or compare it. A Loop condition with a timed delay lets you sample the changing value at a controlled interval.

It creates and switches focus to the new worksheet, so subsequent Excel write actions target the newly added sheet unless you explicitly re-select another.

Cloud flows handle scheduling, process mining, connector integrations, and AI capabilities like AI Builder. Desktop flows handle UI automation, API automation, and database automation on the local machine.

The AI Hub is a centralized cloud location where users create, manage, and use various AI models, such as document processing and object detection.

Pre-built models are ready-to-use for common scenarios (like invoices or ID cards) with no training required, while custom models are trained on your own sample documents for specific formats.

Yes. The cloud offers several pre-built models, including ID card reader, entity extraction, classification, language detection, and finance or receipt models.

Define the fields to extract, upload and tag sample documents, train the model on those samples, test its accuracy, and publish it for use in flows.

A custom document processing model typically needs at least five sample documents per layout, with more improving accuracy.

Publishing makes a trained model available for use in flows. Until a model is published, it exists only in training and cannot be called by an automation.

It uses OCR to convert scanned images and non-selectable PDF text into machine-readable text, either through desktop OCR engines or AI Builder's cloud document models.

Define the table field in the model, tag the rows and columns in sample documents so the model learns the structure, train and publish, then read the returned table rows in the flow.

Advanced tagging lets you mark table boundaries, merge or split columns, and label multi-row or variable-length tables so the model correctly captures structured tabular data.

Leave that field untagged for that sample. The model learns the field is optional and will return an empty value when it is absent, rather than failing.

Yes, through OCR and AI Builder models, though accuracy on handwriting is lower than on printed text and depends on legibility.

Create separate collections (layouts) within the same custom model and tag samples for each, so the model recognises multiple formats and applies the right extraction per document.

The invoice processing model reads standard invoice fields out of the box, but its limitation is that it expects common invoice layouts and may miss non-standard or heavily customised formats, which need a custom model.

The model returns the object name and the object count, showing how many instances of each object were found in the image.

Document models need about five images, while Object Detection needs at least 15 images per object, with 50 or more recommended for better accuracy.

No. Object Detection is available only as a custom model; there is no pre-built version.

AI Builder object detection supports common objects, logos, and custom objects that you train on your own images.

Category classification trains on labelled text records, and Dataverse provides the structured table that stores the text and its category labels for the model to learn from.

A mapping error occurs when the source columns (for example from Excel) do not correctly match the target Dataverse table columns, so the import cannot place the data. Fixing it means aligning column names and data types.

Create the entity in the model, then tag examples of it in your sample text so the model learns to recognise that entity type.

Highlight the full multi-word phrase as a single tag during the tagging step, so the model treats the whole span as one entity rather than separate words.

An API (Application Programming Interface) lets two systems communicate directly, so an automation can send and receive data from another application without using its user interface.

The endpoint URL is the specific web address the API request is sent to, identifying the resource or operation the automation wants to access.

API responses arrive as JSON text. Converting JSON to a custom object turns it into a structured value whose fields you can address directly, instead of parsing raw text.

The Parse JSON action in the cloud, which takes JSON and a schema and outputs addressable fields.

Provide the required credentials in the HTTP request, such as an API key, bearer token, or basic authentication header, matching what the API expects.

Power Automate Cloud connects to many services, including SharePoint, OneDrive, Google Drive, Trello, Box, and social platforms, through its connectors.

A trigger (the event that starts the flow), actions (the steps that run after the trigger), and conditions (the logic that decides which actions run).

A work queue is a managed list of transaction items to process, maintained in the Power Automate cloud portal, that bots pull from to distribute and track work.

In the Power Automate cloud portal, go to Work queues and open the queue by name to see all items with their status, priority, and data.

The main work queue actions are Process work queue items, Update work queue items, Add work queue items, and Requeue item with delay.

Use the Add work queue item action, defining the work queue, priority, name, input data, expiry, and processing notes.

The work queue item, its new status, and the processing result.

It re-adds a work queue item to its original queue after a delay, so a failed or deferred item is retried later.

A Loader bot gathers data and adds items to the work queue, while a Performer bot picks items from the queue and processes each transaction. Separating them enables parallel, scalable processing.

Multibot architecture splits work between a Loader bot that populates a shared work queue and one or more Performer bots that process items concurrently, allowing the workload to scale across machines.

Separating them lets loading and processing run independently and in parallel, makes each flow simpler to maintain, and allows multiple performers to work the same queue without conflicting with loading.

Its status changes to in-progress (locked) so no other bot processes the same item, then moves to completed or failed based on the outcome.

It updates the queue item's status to completed (with a successful processing result) using the Update work queue item action.

The Process work queue items action loops until the queue has no remaining pending items, so the performer stops automatically when the queue is empty.

A scheduled flow runs a process automatically at a set time or interval, like an alarm, for example every day at 6:00 AM or every hour. Use it for recurring, time-based automation.

A flow name, a start date, a start time, and the frequency defining how often it repeats.

Configure the recurrence to repeat on specific days of the week (for example, selecting only weekdays or only Saturday and Sunday) in the schedule settings.

Manually (instant), on a schedule (recurrence), or automatically from an event such as a new email or a new file.

An event-based trigger starts a flow when something happens in OneDrive, such as a file being created or modified in a watched folder.

Use an email-based trigger (for example, When a new email arrives) that starts the flow whenever a matching message is received.

The trigger exposes the attachments as a collection, which you loop through to save or process each attachment individually.

Machine Runtime connects a local machine to the Power Automate cloud so cloud flows can trigger and run desktop flows on that machine, including unattended.

It runs on supported Windows versions (Windows 10 and 11, and Windows Server editions) with the runtime installed and the machine registered to the environment.

Get the computer name from Windows system settings or by running the hostname command in Command Prompt, then use it to register the machine.

In the cloud action that runs the desktop flow, open Settings via the ellipsis and increase the timeout duration so the flow has enough time to complete.

No. Power Automate Cloud does not currently offer built-in version control for flows.

Copilot can suggest actions, explain steps, generate expressions, and help build parts of a flow from a natural-language description, speeding up development.

By describing the goal in plain language, the developer can have Copilot generate the logic, expressions, or scripts, lowering the coding barrier.

Use scripting when a task cannot be done with built-in actions, or when a script is far more efficient, such as complex string manipulation or calling functionality PAD does not expose directly.

VBScript, JavaScript, PowerShell, and Python.

A digital agent is an AI-powered assistant built in Copilot Studio that answers questions and performs tasks using instructions, knowledge sources, and connected actions.

Sources such as uploaded documents, websites, SharePoint sites, or Dataverse tables, which the agent draws on to answer questions.

Instructions define the agent's behaviour, tone, scope, and rules, guiding how it responds and keeping it focused on its intended purpose.

Restrict it to specified knowledge sources and disable general web search, so it answers only from the approved content.

Topics are defined conversation paths that give short, specific answers for particular high-priority queries, instead of relying on general knowledge-source responses.

Topics let a developer script precise, crisp answers for specific or urgent queries (like an emergency), overriding the agent's general behaviour for those cases.

Power Fx expressions add logic to topics, for example calculating values, formatting variables, or making conditional decisions within the conversation flow.

Publishing makes the agent live on its channels (website, Teams, email, and others) so end users can interact with it.

Configure an email-based trigger so an incoming message starts the agent's flow, which then reads, processes, and responds to the inquiry.

Basic Authentication, using the ServiceNow instance URL, a username (often admin), and a generated password.

The ServiceNow instance URL, valid credentials, and the connector or custom action configured with authentication so the agent can create and read tickets.

Add an Ask a question node between the existing fields and the confirmation step, create a variable to store the response, and insert that variable into the final ticket creation and confirmation.

Read the ticket's priority, then use conditional logic (If or Switch) to assign a resolution time per priority level, and return that value to the user.

Columns such as a unique ID, the requester or contact, the request or summary text, category, status, and a timestamp, matching the data the agent captures.

Generate the summary, then use the Add or Update row action to write each summary field into the matching Dataverse column.

Fall back to keyboard shortcuts (such as Ctrl+C), Send physical click, image-based actions, or OCR to read the text when the application does not expose selectable UI elements.

A solution is a container that packages flows, agents, tables, and connections so they can be moved together between environments, such as from development to production.

An unmanaged solution is editable and used in development, while a managed solution is locked and read-only, used in test and production to keep the deployed version consistent.

In the source environment, add the agent to a solution, export the solution (choosing managed or unmanaged), which produces a zip file you then import into the target environment.

In the target environment, use Import solution, upload the exported zip file, map any required connections, and complete the wizard to deploy its contents.

Create a Data Loss Prevention (DLP) policy that classifies or blocks specific connectors and actions, preventing their use across the environment.

No. The suspended desktop flow will not be available to select in the cloud flow's dropdown until the policy issue is resolved.

In Power Automate, go to Data and then Connections to view and manage the connections your flows use.

Give clear names to variables and UI elements, divide the flow into logical subflows, focus the required window when working across multiple windows, and focus a text field before using Send keys.

Automatic scheduled triggers, triggering a flow via a desktop shortcut or URL, access to premium and custom connectors, and process mining.

Yes. The same desktop flow can be triggered attended (with a user present) or unattended (on a registered machine without supervision), depending on how it is run.

In the Power Automate cloud portal under the desktop flow's run history and the machine or machine group's monitoring view, which show run status and results.

A PDD outlines the business process to be automated and is usually prepared by the end user or business analyst. An SDD describes how the automated solution is built and is prepared by the developer to explain the to-be process.

Present the automation projects and hands-on work you have genuinely done, framed honestly against your total experience, rather than overstating years, and emphasise the tools and processes you have actually built.

Understanding of RPA tools and feasibility, process assessment for automation suitability, writing PDDs, familiarity with exception handling and work queues, and the ability to bridge business needs with automation design.

Prepare hands-on examples from real projects, know core concepts (variables, loops, selectors, exception handling, work queues), be ready to explain design decisions, and practise articulating how you solved specific automation challenges.