This post is a part of the AB-620: Designing and Building Integrated AI Agent Solutions in Copilot Studio Exam Prep Hub.
This topic falls under these sections:
Plan and configure agent solutions (30–35%)
   --> Create and monitor agent flows in Copilot Studio
      --> Add input and output parameters

Note that there are 10 practice questions (with answers) at the end of each section to help you solidify your knowledge of the material. Also, there are 4 practice tests with 30 questions each available from the hub's main page below the exam topics section.

Introduction

One of the key capabilities of Microsoft Copilot Studio is enabling agents to exchange information with users, workflows, Power Automate flows, connectors, REST APIs, child agents, and enterprise systems. This exchange of information is made possible through input and output parameters.

Input parameters allow an agent to receive the information necessary to perform a task, while output parameters allow the agent to return the results of that task for use in subsequent conversation steps or workflows.

Well-designed parameters make agents more reusable, modular, reliable, and easier to integrate with enterprise systems.

For the AB-620 certification exam, you should understand how input and output parameters work, how they are used within Copilot Studio, Power Automate, connectors, and REST APIs, and the best practices for designing parameter-driven agent flows.

What Are Parameters?

A parameter is a named piece of information that is passed into or returned from an action, flow, API, or agent.

Parameters allow different components to communicate without hardcoding values.

Examples include:

• Customer ID
• Order number
• Product name
• Employee ID
• Email address
• Approval decision
• Invoice amount
• Support ticket number

Parameters make workflows flexible and reusable.

Input Parameters

An input parameter is information supplied to an action before it executes.

The action requires this information to complete its work.

Examples:

• Customer ID
• Product SKU
• Reservation date
• Ticket number
• Email address
• Employee ID

Without required input parameters, many actions cannot execute successfully.

Output Parameters

An output parameter is information returned after an action completes.

Examples:

• Customer name
• Ticket status
• Confirmation number
• Order total
• Error message
• Success flag
• Appointment time
• Generated document URL

Output parameters allow the conversation or workflow to continue using the returned information.

How Parameters Flow Through an Agent

A typical sequence is:

1. User submits a request.
2. Agent gathers required inputs.
3. Inputs are passed to an action.
4. Action executes.
5. Output parameters are returned.
6. Agent uses outputs in later conversation steps.

Example:

User:
“Check the status of order 54321.”

Input parameter:

• Order Number = 54321

Connector executes lookup.

Output parameters:

• Status = Shipped
• Delivery Date = Friday
• Tracking Number = 998877

The agent presents the results naturally to the user.

Sources of Input Parameters

Input values can come from multiple sources.

User Input

Collected directly during conversation.

Example:

“What is your employee ID?”

Conversation Variables

Previously collected values can be reused.

Example:

Customer ID gathered earlier in the conversation.

Entity Recognition

The AI extracts values automatically.

Example:

“Schedule a meeting tomorrow at 3 PM.”

Parameters extracted:

• Date
• Time

System Values

Generated automatically.

Examples:

• Current date
• Current time
• User identity
• Environment information

Previous Actions

Outputs from one action often become inputs for another.

Example:

Action 1 returns:

• Customer ID

Action 2 uses:

• Customer ID

This chaining enables sophisticated workflows.

Types of Parameters

Common parameter types include:

Text (String)

Examples:

• Customer name
• Email address
• Product name

Number

Examples:

• Quantity
• Invoice total
• Age

Boolean

Examples:

• Approved
• Active
• Completed

Possible values:

• True
• False

Date and Time

Examples:

• Appointment date
• Purchase date
• Deadline

Currency

Examples:

• Order amount
• Invoice balance
• Refund amount

Arrays (Lists)

Examples:

• Product list
• Employee list
• Search results

Objects

Complex data structures containing multiple fields.

Example:

Customer

• Customer ID
• Name
• Email
• Address
• Status

Objects simplify passing related information together.

Required vs Optional Parameters

Required Parameters

Must be supplied before execution.

Example:

Customer ID

Without it, the lookup cannot proceed.

Optional Parameters

Improve flexibility but are not mandatory.

Example:

Preferred language

The workflow can continue even if omitted.

Parameter Validation

Input values should always be validated.

Validation checks may include:

• Required fields
• Correct format
• Numeric ranges
• Date validity
• Allowed values
• Maximum length
• Minimum length

Proper validation reduces execution errors.

Example Validation

User enters:

Email = john@email

Validation detects an invalid email format.

The agent asks:

“That doesn’t appear to be a valid email address. Could you enter it again?”

This improves user experience.

Default Values

Optional parameters may have default values.

Example:

Language = English

If the user provides no language preference, the workflow uses the default.

Using Parameters in Power Automate

Power Automate flows commonly receive input parameters.

Examples:

Inputs:

• Employee ID
• Purchase Amount
• Department

Flow performs approval.

Outputs:

• Approved
• Approver Name
• Approval Date

The outputs are returned to Copilot Studio.

Using Parameters with Connectors

Connectors require parameters for operations.

Example:

SQL query

Inputs:

• Customer ID

Outputs:

• Customer record

Example:

Outlook connector

Inputs:

• Recipient
• Subject
• Body

Outputs:

• Success
• Message ID

REST API Parameters

REST APIs use parameters in several ways.

Path Parameters

Example:

GET /customers/{CustomerID}

CustomerID is a path parameter.

Query Parameters

Example:

GET /orders?status=Open

Status is a query parameter.

Request Body

POST operations often send parameters in JSON.

Example:

{
 "Name":"John",
 "Department":"Finance"
}

Using Output Parameters

Output values may be used to:

• Display results
• Make decisions
• Trigger another action
• Update variables
• Call another connector
• Generate Adaptive Cards
• Start approvals

Outputs often drive the next stage of a workflow.

Chaining Parameters

Complex workflows often chain outputs into subsequent inputs.

Example:

Action 1

Input:

• Customer Email

Output:

• Customer ID

↓

Action 2

Input:

• Customer ID

Output:

• Orders

↓

Action 3

Input:

• Order Number

Output:

• Shipment Status

This creates intelligent multi-step automation.

Error Handling

Missing or invalid parameters should be handled gracefully.

Examples:

Missing required value

Agent:

“I’ll need your employee number before I can continue.”

Invalid date

Agent:

“Please enter a valid future date.”

Good error handling improves usability.

Security Considerations

Sensitive parameters require additional protection.

Examples:

• Passwords
• Tokens
• Credit card numbers
• Personal information
• Medical records

Best practices include:

• Encrypt sensitive data.
• Limit parameter visibility.
• Apply least privilege.
• Avoid logging confidential values.
• Validate all external input.
• Follow organizational DLP policies.

Parameter Naming Best Practices

Good parameter names should be:

• Descriptive
• Consistent
• Simple
• Easy to understand

Good examples:

• CustomerID
• EmployeeEmail
• PurchaseAmount
• OrderNumber

Poor examples:

• Value1
• Temp
• InputA
• DataX

Designing Reusable Parameters

Reusable components should:

• Use standardized parameter names.
• Minimize required inputs.
• Return consistent outputs.
• Follow organizational naming conventions.
• Document expected values.
• Validate all inputs.

Reusable parameters simplify enterprise integration.

Common Mistakes

Avoid:

• Poor parameter names
• Missing validation
• Hardcoded values
• Returning excessive data
• Inconsistent naming
• Missing required inputs
• Ignoring null values
• Exposing sensitive information

Best Practices

• Clearly define all inputs and outputs.
• Validate every input.
• Use descriptive names.
• Prefer reusable parameter structures.
• Return only necessary information.
• Handle missing values gracefully.
• Protect confidential parameters.
• Document expected formats.
• Use consistent naming across flows.
• Test all parameter scenarios before deployment.

Exam Tips

For the AB-620 exam, remember these key points:

• Input parameters provide information needed to execute actions.
• Output parameters return results for later use in conversations and workflows.
• Parameters can originate from users, variables, previous actions, entities, or system values.
• Connectors, Power Automate flows, child agents, and REST APIs all use parameters extensively.
• Required parameters must be supplied before execution, while optional parameters improve flexibility.
• Validate inputs to reduce execution errors and improve user experience.
• Outputs often become inputs for later actions in multi-step workflows.
• Sensitive parameters should be protected through appropriate security controls.
• Use descriptive naming conventions to improve maintainability.
• Well-designed parameters make agents modular, reusable, and easier to integrate.

Practice Exam Questions

Question 1

A Copilot Studio action retrieves customer information from a CRM system. Which value is most likely to be an input parameter?

A. Customer ID

B. Customer Name

C. Account Balance

D. Support Ticket Status

Correct Answer: A

Explanation: The Customer ID is supplied to the action so it can locate the correct customer record. The other values are typically returned as outputs.

Question 2

What is the primary purpose of an output parameter?

A. To authenticate a connector

B. To return information after an action completes

C. To publish an agent

D. To configure security roles

Correct Answer: B

Explanation: Output parameters return the results of an action, allowing the agent or workflow to use that information in subsequent steps.

Question 3

Which source can automatically provide input parameters through natural language understanding?

A. Connection references

B. Environment variables

C. Entity recognition

D. Audit logs

Correct Answer: C

Explanation: Entity recognition extracts structured values such as dates, times, locations, and names directly from user conversations.

Question 4

Why should required input parameters be validated before executing an action?

A. To improve dashboard reporting

B. To increase API quotas

C. To reduce licensing costs

D. To prevent execution errors caused by missing or invalid data

Correct Answer: D

Explanation: Validation ensures that required information is complete and properly formatted before an action is executed.

Question 5

A Power Automate flow returns an approval result to Copilot Studio. What type of parameter is the approval decision?

A. Input parameter

B. Environment variable

C. Output parameter

D. Authentication token

Correct Answer: C

Explanation: The approval decision is produced by the flow and returned to Copilot Studio, making it an output parameter.

Question 6

Which parameter type would be most appropriate for representing whether a purchase request has been approved?

A. Boolean

B. Currency

C. Date

D. Array

Correct Answer: A

Explanation: Approval status is naturally represented as a Boolean value (True or False).

Question 7

A workflow retrieves a Customer ID from one action and uses it in the next action to retrieve recent orders. What design pattern is being used?

A. Environment isolation

B. Parameter chaining

C. Adaptive Card rendering

D. Authentication delegation

Correct Answer: B

Explanation: Parameter chaining uses the output of one action as the input to another, enabling multi-step business processes.

Question 8

Which naming convention is considered the best practice for parameters?

A. Value1

B. TempData

C. CustomerID

D. InputA

Correct Answer: C

Explanation: Parameter names should be descriptive, meaningful, and consistent to improve readability and maintainability.

Question 9

Which type of REST API parameter is typically embedded directly within the URL path?

A. Request body parameter

B. Path parameter

C. Output parameter

D. Header parameter

Correct Answer: B

Explanation: Path parameters are incorporated directly into the endpoint URL, such as /customers/{CustomerID}.

Question 10

Why should organizations avoid logging sensitive input parameters such as passwords or personal information?

A. It improves conversation speed.

B. It simplifies API development.

C. It reduces connector licensing costs.

D. It helps protect confidential information and supports security and compliance requirements.

Correct Answer: D

Explanation: Sensitive parameters should be protected from unnecessary exposure to reduce security risks and comply with privacy regulations and organizational governance policies.

Go to the AB-620 Exam Prep Hub main page

This post is a part of the AB-620: Designing and Building Integrated AI Agent Solutions in Copilot Studio Exam Prep Hub.
This topic falls under these sections:
Plan and configure agent solutions (30–35%)
   --> Plan an agent solution
      --> Design agents for internal or external audiences

Note that there are 10 practice questions (with answers) at the end of each section to help you solidify your knowledge of the material. Also, there are 4 practice tests with 30 questions each available from the hub's main page below the exam topics section.

Introduction

One of the first architectural decisions when designing an AI agent in Microsoft Copilot Studio is determining who the intended users are. The audience influences nearly every aspect of the solution, including authentication, security, data access, integrations, conversation design, deployment channels, governance, and compliance.

In general, AI agents are designed for one of two broad audiences:

• Internal audiences – employees, contractors, and trusted business partners within an organization.
• External audiences – customers, vendors, suppliers, citizens, or members of the public.

Although the same technologies can often be used for both, their design requirements differ significantly. An internal HR assistant, for example, has access to confidential employee information and typically uses Microsoft Entra ID authentication, whereas a customer support chatbot on a public website may serve anonymous users and require high scalability.

For the AB-620 exam, you should understand how audience selection affects architecture, security, deployment, integrations, Responsible AI considerations, and user experience.

Why Audience Planning Is Important

Identifying the intended audience early in the planning process helps determine:

• Authentication requirements
• Authorization model
• Data sources
• Enterprise integrations
• Conversation style
• Deployment channels
• Security controls
• Compliance requirements
• Scalability needs
• Monitoring strategy

Design decisions that work well for an internal employee assistant may be inappropriate for a customer-facing chatbot.

Internal Audiences

Internal audiences consist of people who belong to or work closely with the organization.

Examples include:

• Employees
• Contractors
• Consultants
• Internal support staff
• Executives
• Managers
• Business partners (when granted organizational access)

Internal users generally operate within the organization’s identity and security infrastructure.

Common Internal Agent Scenarios

Examples include:

• IT help desk
• HR self-service
• Employee onboarding
• Benefits assistance
• Payroll inquiries
• Leave management
• Knowledge management
• Project support
• Finance assistance
• Internal policy search

These agents typically improve employee productivity and reduce repetitive administrative work.

Characteristics of Internal Agents

Internal agents commonly include:

• Microsoft Entra ID authentication
• Single Sign-On (SSO)
• Role-Based Access Control (RBAC)
• Access to confidential business data
• Enterprise system integrations
• Department-specific permissions
• Microsoft Teams deployment
• Microsoft 365 Copilot integration
• Detailed audit logging

Internal agents often operate within trusted enterprise environments.

External Audiences

External audiences include individuals outside the organization.

Examples include:

• Customers
• Prospective customers
• Vendors
• Suppliers
• Patients
• Students
• Citizens
• Website visitors
• Members of the public

External agents are often designed to improve customer experience and reduce support costs.

Common External Agent Scenarios

Examples include:

• Customer support
• Product information
• Order tracking
• Appointment scheduling
• FAQ assistants
• Banking support
• Insurance claims assistance
• Travel booking assistance
• Healthcare appointment requests
• Retail shopping assistance

These agents are commonly deployed through public-facing channels.

Characteristics of External Agents

External agents often require:

• Public website deployment
• Mobile application integration
• Customer authentication (when needed)
• High availability
• High scalability
• Simplified conversations
• Strong security protections
• Protection against malicious users
• Rate limiting
• Monitoring for abuse

Unlike internal agents, external agents must often support thousands or millions of users.

Comparing Internal and External Agents

Authentication Considerations

Authentication requirements differ significantly depending on the audience.

Internal Authentication

Most internal agents use:

• Microsoft Entra ID
• Single Sign-On (SSO)
• Multi-Factor Authentication (MFA)
• Conditional Access

This allows employees to securely access business resources using their corporate identities.

External Authentication

External agents may support:

• Anonymous users
• Customer accounts
• OAuth providers
• Social identity providers
• Customer Identity and Access Management (CIAM) solutions

The authentication model depends on the business scenario and the sensitivity of the information being accessed.

Authorization Considerations

After users authenticate, authorization determines what they can access.

Examples include:

Internal HR agent:

• View employee benefits
• Submit leave requests
• Access payroll information (based on role)

External customer agent:

• View personal orders
• Check warranty status
• Update personal profile
• Access support tickets

Authorization should always follow the principle of least privilege.

Conversation Design

Different audiences require different conversation styles.

Internal Conversations

Internal users generally:

• Understand company terminology
• Know business processes
• Expect technical accuracy
• Need detailed responses
• Frequently request operational assistance

Conversations can safely include organizational terminology and internal references.

External Conversations

External conversations should be:

• Friendly
• Clear
• Concise
• Easy to understand
• Free of internal jargon
• Guided with simple instructions

Customer-facing agents should avoid exposing internal business processes or confidential information.

Enterprise Integrations

Internal agents commonly integrate with:

• Microsoft Dataverse
• Microsoft Dynamics 365
• Microsoft SharePoint
• Microsoft Teams
• Microsoft Graph
• Human Resources systems
• ERP systems
• IT Service Management (ITSM) platforms

External agents typically integrate with:

• CRM systems
• Customer support platforms
• E-commerce systems
• Order management systems
• Payment services
• Public APIs

The audience influences which systems the agent should access.

Knowledge Sources

Knowledge sources should also be selected based on the intended audience.

Internal knowledge sources include:

• Employee handbook
• HR policies
• Technical documentation
• Internal procedures
• Project documentation
• Corporate SharePoint sites

External knowledge sources include:

• Product documentation
• Public FAQs
• User manuals
• Marketing materials
• Public knowledge bases
• Customer support articles

Never expose confidential internal knowledge through an external agent.

Security Considerations

Internal Agents

Security planning includes:

• Microsoft Entra ID
• RBAC
• Conditional Access
• Audit logging
• Data Loss Prevention (DLP)
• Least privilege

External Agents

Additional protections often include:

• Input validation
• Rate limiting
• CAPTCHA (where appropriate)
• Web Application Firewalls (WAF)
• Abuse detection
• Prompt injection protection
• Secure customer authentication

External agents face a broader range of security threats due to public accessibility.

Responsible AI Considerations

Audience affects Responsible AI planning.

Internal agents should:

• Protect confidential business information
• Respect employee privacy
• Limit access to sensitive records
• Support human escalation for sensitive requests

External agents should:

• Clearly disclose AI-generated responses
• Protect customer privacy
• Avoid collecting unnecessary personal information
• Escalate complex or sensitive issues to human representatives
• Provide transparent limitations

Responsible AI principles apply equally to both audiences but are implemented differently based on context.

Scalability Considerations

Internal agents generally support a relatively predictable number of users.

External agents may experience:

• Seasonal demand spikes
• Marketing-driven traffic
• Large numbers of concurrent users
• Global usage

Planning for external audiences often requires greater emphasis on performance, scalability, and resilience.

Governance Considerations

Governance differs based on the intended audience.

Internal governance focuses on:

• Identity management
• Department ownership
• Internal compliance
• Data classification
• Enterprise integrations

External governance emphasizes:

• Customer privacy
• Regulatory compliance
• Public content review
• Brand consistency
• Service availability
• Incident response

Monitoring

Both internal and external agents should be monitored.

Internal monitoring may focus on:

• Employee adoption
• Conversation success rates
• Workflow completion
• Productivity improvements

External monitoring often includes:

• Customer satisfaction
• Abandonment rates
• Escalation frequency
• Response quality
• Performance
• Security events
• Abuse detection

Common Design Mistakes

Avoid these common mistakes:

• Using internal terminology in customer-facing agents.
• Exposing confidential enterprise knowledge through external agents.
• Applying weak authentication to internal agents.
• Overcomplicating conversations for external users.
• Ignoring scalability for public deployments.
• Granting excessive permissions.
• Using the same security model for both audiences without evaluating risk.
• Failing to monitor production usage.

Best Practices

When designing agents for internal or external audiences:

• Clearly identify the intended audience before designing the solution.
• Select authentication methods appropriate for the audience.
• Apply Role-Based Access Control (RBAC) and least-privilege access.
• Choose deployment channels that align with user workflows.
• Use appropriate enterprise knowledge sources.
• Design conversations that match user expertise.
• Protect confidential organizational data.
• Plan for expected scale and availability.
• Implement Responsible AI practices.
• Continuously monitor and improve the solution after deployment.

Exam Tips

For the AB-620 exam, remember the following:

• Audience selection influences nearly every architectural decision.
• Internal agents typically use Microsoft Entra ID, Microsoft Teams, and enterprise integrations.
• External agents commonly use websites, mobile apps, and customer-facing systems.
• Internal agents access confidential organizational data, while external agents should expose only approved public or customer-specific information.
• Apply the principle of least privilege to both internal and external solutions.
• Use conversation styles appropriate to the audience.
• External agents generally require greater scalability and stronger protections against public threats.
• Responsible AI principles apply to all agents regardless of audience.
• Monitoring and governance remain important throughout the solution lifecycle.

Practice Exam Questions

Question 1

A company wants to build an AI agent that allows employees to check payroll information and submit vacation requests through Microsoft Teams. Which audience is the agent primarily designed for?

A. External customers

B. Business partners without company accounts

C. Internal employees

D. Public website visitors

Correct Answer: C

Explanation: Payroll and vacation information are internal business functions intended for employees. Microsoft Teams and Microsoft Entra ID are common platforms for internal employee-facing agents.

Question 2

Which deployment channel is generally the best choice for an AI agent designed to answer questions from visitors browsing a company’s public website?

A. Microsoft Teams

B. Public website

C. Employee portal

D. Microsoft 365 Copilot

Correct Answer: B

Explanation: Public websites are the most common deployment channel for customer-facing AI agents that provide product information, FAQs, and support.

Question 3

Which authentication approach is most appropriate for an internal HR assistant used only by employees?

A. Anonymous access

B. Shared administrator account

C. Microsoft Entra ID with Single Sign-On

D. Public API key authentication

Correct Answer: C

Explanation: Internal enterprise agents typically use Microsoft Entra ID with Single Sign-On to provide secure, seamless authentication for employees.

Question 4

A customer-facing AI agent should avoid exposing which type of information?

A. Public product documentation

B. Marketing content

C. Public FAQs

D. Internal HR policies and confidential business documents

Correct Answer: D

Explanation: External agents should only expose approved public or customer-specific information. Confidential internal documents should never be available to public users.

Question 5

Why do external AI agents typically require greater scalability than internal agents?

A. They always use Microsoft Teams.

B. They generally serve larger and less predictable user populations.

C. They never require authentication.

D. They cannot access enterprise systems.

Correct Answer: B

Explanation: External agents often serve thousands or millions of customers and may experience significant traffic spikes, requiring scalable architectures.

Question 6

Which conversation characteristic is generally most appropriate for an external customer-facing AI agent?

A. Use extensive internal acronyms and technical terminology.

B. Assume users understand company-specific business processes.

C. Provide concise, user-friendly responses using plain language.

D. Require detailed technical knowledge before beginning the conversation.

Correct Answer: C

Explanation: External users benefit from clear, concise, and jargon-free conversations that are easy to understand.

Question 7

An organization is designing an AI agent that retrieves employee handbook information from SharePoint. Which knowledge source is most appropriate?

A. Public product catalog

B. Internal SharePoint knowledge repository

C. External social media posts

D. Public marketing website

Correct Answer: B

Explanation: Employee handbooks are internal documents and should be stored in secured enterprise repositories such as SharePoint with appropriate access controls.

Question 8

Which security practice should be applied to both internal and external AI agents?

A. Allow unrestricted access to all enterprise systems.

B. Use anonymous authentication for all users.

C. Eliminate audit logging to improve performance.

D. Apply the principle of least privilege when granting permissions.

Correct Answer: D

Explanation: Least privilege is a fundamental security principle that limits permissions to only those necessary, reducing security risks for both internal and external solutions.

Question 9

What is one key governance difference between internal and external AI agents?

A. Only external agents require monitoring.

B. Internal agents do not require compliance planning.

C. External agents place greater emphasis on customer privacy, public content review, and brand consistency.

D. Internal agents never integrate with enterprise systems.

Correct Answer: C

Explanation: While governance is important for all AI agents, external solutions must also address customer privacy, public communications, and brand reputation in addition to security and compliance.

Question 10

An architect is determining authentication methods, deployment channels, enterprise integrations, and conversation style before building a Copilot Studio agent. What is the most important planning activity that should be completed first?

A. Select the programming language for connectors.

B. Create Adaptive Cards for every conversation.

C. Design the monitoring dashboard.

D. Identify whether the agent is intended for an internal or external audience.

Correct Answer: D

Explanation: Identifying the target audience is a foundational planning activity because it influences authentication, security, deployment, knowledge sources, integrations, governance, and the overall user experience.

Go to the AB-620 Exam Prep Hub main page