Tutorial 2 of 5

Stakeholders, Workflow, and Responsibility Design

COMP 327 Capstone Project — Assignment 6: Discipline-Specific Responsible AI Integration Strategy

Suggested completion time: 120–150 minutes

Required prior work: Completed Tutorial 1 (project brief); Unit 20 materials

Capstone period: Weeks 12–13 (Middle Week 12)

Connects to: Tutorial 3 (framework selection), Tutorial 4 (governance)

1. Purpose of This Tutorial

This tutorial helps you map the proposed AI-supported workflow and allocate responsibility for your Capstone Project. Building on the scope and purpose you defined in Tutorial 1, you will identify who is affected, how the workflow will operate, and who is accountable for each step.

A responsible AI integration strategy is not just about technology—it is about people, processes, and authority. This tutorial ensures that you design a workflow that is not only effective but also reviewable, contestable, and accountable. You will produce a stakeholder map, workflow diagrams, a responsibility matrix, and an escalation/fallback pathway.

📌 Why this matters

A well-designed workflow with clear responsibility is the core of a responsible AI integration. Without it, your strategy is merely a collection of ideas. This tutorial transforms your proposal into a practical, governable design that can be implemented (or simulated) with confidence.

2. Connection to the Capstone Project

The COMP 327 Capstone Project requires you to design a Discipline-Specific Responsible AI Integration Strategy. Your strategy must demonstrate that you have considered not only the technical aspects but also the human, organisational, and governance dimensions.

This tutorial directly addresses several required components of your final report:

Stakeholders and Affected Non-Users (Section 6 of the recommended structure)
Proposed AI Role and Human-Only Activities (Section 8)
Current and Proposed Workflow (Section 9)
Information, Sources, Permissions, and Authority (Section 10)

The outputs from this tutorial will feed directly into those sections and will also inform your framework selection (Tutorial 3) and governance design (Tutorial 4).

🔗 Links to other tutorials

Tutorial 1 provided the scope and purpose; you will now map stakeholders and workflows for that defined problem.
Tutorial 3 will use your workflow and responsibility design to select and apply relevant frameworks (especially PARTNER).
Tutorial 4 will transform your responsibility gaps into governance controls and safeguards.

3. Learning Outcomes

By completing this tutorial, you will be able to:

Identify and categorise all relevant stakeholders, including intended users, practitioners, decision-makers, affected non-users, beneficiaries, risk bearers, information subjects, and communities.
Map the current workflow (without AI) and the proposed workflow (with AI) for your bounded problem.
Define a bounded AI role and clearly distinguish AI-assisted activities from human-only activities.
Allocate authority for review, decision, action, correction, appeal, incident response, and stop authority using a PARTNER responsibility matrix.
Design an escalation and fallback pathway to handle workflow failures and maintain human accountability.
Recognise stop conditions related to inadequate responsibility or authority and take corrective action.

4. Key Concepts

The following concepts are essential for designing responsible workflows and allocating authority. You will apply them directly to your capstone.

4.1 Stakeholder Mapping

A stakeholder is any person, group, or organisation that has an interest in or is affected by the AI integration. Stakeholder mapping is the process of identifying and categorising these parties to understand their needs, roles, and influence.

Key categories include:

Intended users – those who will directly interact with the AI system.
Practitioners and decision-makers – professionals who use the system's outputs to make decisions.
Affected non-users – people who do not use the system but are impacted by its outcomes.
Beneficiaries – those who stand to gain from the integration.
Risk bearers – those who may suffer harm or adverse consequences.
Information subjects – individuals whose data is processed.
Communities – groups that require meaningful participation or consultation.

4.2 Workflow Mapping

A workflow is a sequence of tasks, decisions, and handoffs that achieve a specific outcome. You will map the current-state workflow (without AI) and the proposed-state workflow (with AI) to visualise how the integration changes the process.

Workflow mapping helps identify:

Where AI is introduced and what it contributes.
Where human judgment remains essential.
Potential failure points and bottlenecks.
Opportunities for oversight and review.

4.3 PARTNER Responsibility Matrix

PARTNER is a framework for allocating responsibility in human-AI workflows. The exact expansion is provided in the course materials; you should refer to that for the full definition. In essence, PARTNER helps you assign who is accountable for each step, what authority they have, and how they review, correct, or escalate issues.

You will use PARTNER to create a responsibility matrix that clarifies:

Who reviews AI outputs.
Who decides and acts.
Who has correction and appeal authority.
Who has incident response and stop authority.

💡 Framework reminder

The PARTNER framework is one of several you can select. It is particularly relevant when human-AI allocation and authority are central to your project. You will formalise your framework selection in Tutorial 3.

4.4 Human-Only Activities

Human-only activities are tasks that should never be delegated to AI, regardless of capability. These typically include:

Final consequential decisions (e.g., hiring, firing, diagnosis, sentencing).
Activities requiring professional judgment or ethical discretion.
Tasks that involve direct human interaction with sensitive populations.
Activities that preserve human agency, dignity, or relationships.
Review and approval of AI-generated outputs before they affect people.

4.5 Escalation and Fallback

Escalation is the process of moving a decision or issue to a higher authority when it cannot be resolved at the current level. Fallback is the alternative pathway when the primary workflow fails (e.g., AI system unavailable, errors detected, or human reviewer unavailable). A responsible design includes clear escalation and fallback procedures.

Stakeholder Mapping – identify all affected parties and their roles

Workflow Mapping – visualise current and proposed processes

PARTNER – framework for responsibility allocation

Human-Only – activities never delegated to AI

Escalation/Fallback – pathways when things go wrong

Authority – review, decision, action, correction, stop

5. Retrieval Practice

Before you begin the guided activities, recall what you already know about stakeholders, workflows, and responsibility.

1. Give an example of an "affected non-user" in an AI-supported workflow.

Think about someone who doesn't use the system but is impacted by its output.

2. What is the difference between a "reviewer" and a "decision-maker" in a workflow?

Consider the level of authority: review suggests checking, decision means final say.

3. Why is it important to identify human-only activities?

Think about accountability, ethics, and preserving human agency.

4. What is a "stop authority" and why is it necessary?

Who can halt the process if something goes wrong? This is critical for safety and accountability.

5. List two reasons a workflow might require a fallback pathway.

Consider system failure, error detection, or human unavailability.

6. Worked Synthetic Example

This example continues from the Education stream project introduced in Tutorial 1. The student has defined a bounded problem: an AI-supported system that generates structured, accessible text summaries of lecture recordings and slide content for students with visual impairments, with human review by accessibility specialists and instructors.

6.1 Stakeholder Map

Figure 1: Stakeholder Map (text description)

Intended Users: Students with visual impairments (who receive summaries).
Practitioners: Accessibility specialists (who review and format summaries); Instructors (who approve and distribute).
Decision-Makers: Accessibility office director (approves overall process).
Affected Non-Users: Other students (may perceive unequal treatment); University administration (budget and policy).
Beneficiaries: Students with visual impairments (improved access); University (compliance with accessibility legislation).
Risk Bearers: Students if summaries contain errors; Instructors if they rely on inaccurate summaries.
Information Subjects: Students whose recorded voices are processed (lecture content).
Communities: Disability student groups; Accessibility advisory committee.

6.2 Current-State Workflow (No AI)

Figure 2: Current Workflow (text description)

Instructor delivers lecture (no recording).
Student with visual impairment requests materials from accessibility office.
Accessibility specialist manually transcribes or converts slides (if available) — can take 7–10 business days.
Materials delivered to student.
Instructor not directly involved in accessibility process.

6.3 Proposed-State Workflow (With AI)

Figure 3: Proposed Workflow (text description)

Instructor delivers lecture (recorded with consent).
AI system processes recording and slides to generate structured text summary (draft).
Accessibility specialist reviews draft: corrects errors, adds formatting, ensures accessibility.
Instructor reviews final summary: verifies accuracy and approves for distribution.
Accessibility office distributes to student(s).
Student provides feedback (optional) for improvement.
If AI fails or errors exceed threshold, fallback: manual transcription (slower, but reliable).

6.4 AI-Role Statement and Human-Only Activities

AI Role: Generate a first draft of structured text summary from lecture recording and slides. Provide timestamps and key concepts. AI does not make final decisions about accuracy or suitability.

Human-Only Activities:

Final review and correction of AI-generated summary.
Approval and distribution of final materials.
Handling of student complaints or requests for alternative formats.
Making decisions about which lectures are prioritised.
Instructor judgment about content accuracy and appropriateness.

6.5 PARTNER Responsibility Matrix (Excerpt)

Table 1: Excerpt from PARTNER matrix for Education example
Activity	Review Authority	Decision Authority	Action Authority	Correction / Appeal	Stop Authority
AI draft generation	System (automated checks); Accessibility specialist (review)	Accessibility specialist (proceed to instructor)	AI system (generation)	Accessibility specialist (can correct)	Accessibility specialist (can stop and fallback)
Final approval	Instructor	Instructor	Accessibility office (distribution)	Student (appeal to instructor)	Instructor (can stop distribution)
Incident response (e.g., error in summary)	Accessibility specialist	Accessibility office director	Accessibility specialist (correct, communicate)	Student (report issue)	Director (halt use if necessary)

6.6 Escalation and Fallback Pathway

Escalation: If accessibility specialist cannot resolve an error, escalate to director. If instructor disputes summary, escalate to department head.
Fallback: If AI system unavailable (e.g., technical failure), revert to manual transcription process (existing workflow) until system restored.
Stop: If errors exceed acceptable threshold, director can suspend AI-generated summaries and require manual process.

✅ Example complete

The student has clearly mapped stakeholders, workflows, responsibility, and fallback pathways, providing a solid foundation for the rest of the capstone.

7. Guided Activity

Using your project brief from Tutorial 1, work through the following steps to map your stakeholders, workflows, and responsibility. Record your answers in a working document.

7.1 Identify Your Stakeholders

List all stakeholders for your proposed AI integration. Use the categories below:

Intended users: Who will directly use the AI system?
Practitioners and decision-makers: Who will use the system's outputs to make decisions or take actions?
Affected non-users: Who is impacted but does not use the system?
Beneficiaries: Who stands to gain?
Risk bearers: Who may be harmed?
Information subjects: Whose data is processed?
Communities or groups: Who requires meaningful participation?

✏️ Your stakeholder list:

Write your stakeholders here.

7.2 Map Current and Proposed Workflows

Describe the current workflow (without AI) and the proposed workflow (with AI) as a sequence of steps. Identify where AI is introduced and what it does, and where humans intervene.

✏️ Current workflow:

Describe the current process.

✏️ Proposed workflow:

Describe the proposed process with AI.

7.3 Define the Bounded AI Role

Write a concise statement that defines exactly what the AI will do and, just as importantly, what it will not do. Be specific about the boundaries.

✏️ AI-role statement:

Write your AI-role statement here.

7.4 Identify Human-Only Activities

List all activities that must remain human-only in your proposed workflow. Explain why each activity cannot be delegated to AI.

✏️ Human-only activities:

List and justify human-only activities.

7.5 Distinguish Levels of Authority

For each major step in your workflow, identify who has authority for:

Review – checking the output
Decision – making a binding choice
Action – executing the decision
Correction / Appeal – fixing errors or challenging outcomes
Stop – halting the process

✏️ Authority mapping:

Describe who holds each type of authority for each step.

7.6 Design Escalation and Fallback Pathways

Identify what happens when the workflow fails (e.g., AI errors, human unavailable, system down). Who is escalated to? What is the fallback process?

✏️ Escalation and fallback:

Describe your escalation and fallback plans.

7.7 Identify Accessible Human Alternatives

For any AI-driven service, there must be an accessible human alternative for those who cannot or choose not to use the AI. Describe what that alternative is and how it is provided.

✏️ Accessible human alternative:

Describe the human alternative.

7.8 Check Stop Conditions

Review your work against the stop conditions for Tutorial 2:

Have you identified all affected people? If not, revise.
Is there a clear accountable owner for each step? If not, revise.
Does your proposed AI role include any consequential decisions (e.g., hiring, diagnosis)? If so, revise to keep those decisions human-only.
Is human review meaningful (not a rubber stamp)? Ensure reviewers have authority and capacity.
Is there a qualified reviewer available? If not, revise or simulate.
Do affected people have a way to correct errors or appeal decisions? If not, add a mechanism.
Is there an accessible human alternative? If not, add one.
Is there a clear stop authority? If not, assign one.

⚠️ If any stop condition is triggered, revise your design before proceeding.

8. Student Production Activity

Now you will produce the complete set of deliverables for this tutorial. These products will become key sections of your final capstone report.

📄 Required Student Products

Stakeholder map – a structured list or diagram of all stakeholders, categorised by role.
Affected-non-user analysis – a brief narrative describing how non-users are impacted and what safeguards are in place.
Current-state workflow – a step-by-step description or diagram of the process without AI.
Proposed-state workflow – a step-by-step description or diagram of the process with AI, clearly showing where AI intervenes.
AI-role statement – a precise statement of what the AI does and what it does not do.
Human-only activity list – a list of tasks that remain exclusively human, with justifications.
PARTNER responsibility matrix – a table mapping each workflow activity to review, decision, action, correction, and stop authorities.
Escalation and fallback pathway – a description of how failures are handled and what alternative processes exist.
Text alternative for every workflow diagram – a written description of any visual diagram you create, ensuring accessibility.

Format: You may use a word processor, spreadsheet (for the matrix), or diagramming tool. Ensure all diagrams have text alternatives. Keep all products together in your evidence-retention folder.

✅ Check stop conditions again

Before finalising, re-run the stop condition checklist. If any condition is met, revise your design. A responsible design is one that can withstand scrutiny.

9. Evidence-Retention Box

Add the following items to your evidence-retention folder. You will need them for the integrated analysis in Tutorial 3 and the governance design in Tutorial 4.

Your stakeholder map
Your affected-non-user analysis
Current and proposed workflow descriptions (and diagrams, if any)
AI-role statement
Human-only activity list
PARTNER responsibility matrix
Escalation and fallback pathway
Text alternatives for any visual diagrams
Any feedback or notes from discussions

10. Common Problems and Corrective Actions

Table 2: Common responsibility design problems and corrections
Problem	Description	Corrective Action
Stakeholders incomplete	Only users and decision-makers are identified; non-users and risk bearers are overlooked.	Use the full stakeholder categories listed in the guided activity. Ask: "Who else is affected, even indirectly?"
AI role too broad	AI is assigned tasks that should be human-only (e.g., making final decisions).	Review the human-only list and move any consequential decisions or judgments to human authority.
No clear accountability	The responsibility matrix does not assign a specific person or role to each step.	Be explicit: name the role (e.g., "Accessibility specialist") for each authority type.
Review is nominal	Reviewer lacks time, authority, or expertise to meaningfully check AI outputs.	Ensure reviewer has adequate resources and authority to reject or correct outputs. Document the review process.
No accessible alternative	Users who cannot use AI have no other option.	Design a non-AI pathway that is equivalent in outcome (even if slower).
No stop authority	No one can halt the workflow if problems arise.	Assign a specific role with explicit stop authority and a trigger condition.

11. Responsible-Use Boundaries

The following boundaries apply to all workflow and responsibility design activities.

⛔ Do Not

Do not design a workflow that delegates consequential decisions (e.g., hiring, firing, diagnosis, sentencing, credit approval) to AI without human final authority.
Do not assume that AI can replace human judgment in areas requiring professional ethics, empathy, or discretion.
Do not design a workflow that lacks a human review step before any output reaches a person.
Do not rely on AI to handle sensitive personal information without explicit permission and security controls.

✅ Do

Design workflows that keep humans in the loop for all substantive decisions.
Ensure that every AI output is reviewable by a qualified person.
Provide clear escalation paths for when things go wrong.
Document all responsibilities in a transparent, accessible manner.

12. Three Equivalent Participation Pathways

You may complete this tutorial using any of the three pathways below.

Option A: Approved AI Tools

Use AI tools for bounded tasks such as:

Generating a first draft of a stakeholder list based on your problem description.
Suggesting potential workflow steps or failure points.
Drafting a responsibility matrix template.
Proposing escalation scenarios.

You must verify and adapt all AI-generated suggestions to your specific context. You remain accountable for the final design.

Option B: Supplied Capstone Records

Use supplied materials such as:

Sample stakeholder maps for your stream.
Generic workflow templates.
Example responsibility matrices.
Common escalation pathways.

Adapt these materials to your specific problem, ensuring they reflect your bounded scope and purpose.

Option C: Non-AI Project Development

Complete all activities using conventional methods:

Brainstorming with pen and paper.
Creating diagrams with drawing tools.
Using course readings to inform stakeholder identification.
Developing responsibility matrices manually.

All pathways assess the same learning outcomes. Choose the one that works best for you.

13. Accessibility Guidance

Ensure your workflow and responsibility designs are accessible to all audiences.

Provide text alternatives for all diagrams (flowcharts, maps).
Use clear, plain language in your responsibility descriptions.
Ensure your responsibility matrix has proper headers and is readable by screen readers.
If you create a diagram, also provide a numbered step list as an alternative.
Consider the needs of stakeholders with disabilities in your design — they should be able to participate and contest decisions.

♿ Tip

The text alternative for a workflow diagram should describe each step, the sequence, and the handoffs between human and AI. This is also a useful way to check completeness.

14. Self-Check Questions

1. Which of the following is not a stakeholder category you should consider in your map?

A) Intended users
B) Affected non-users
C) Competitors of the organisation
D) Risk bearers

Feedback: C is correct. Competitors are generally not stakeholders unless they are directly affected by the AI integration (which is unlikely for a bounded capstone). The other categories are essential.

2. What is the primary purpose of a PARTNER responsibility matrix?

A) To list all the AI tools you will use
B) To clarify who holds what authority at each workflow step
C) To calculate the cost of the AI system
D) To describe the technical architecture

Feedback: B is correct. PARTNER is about allocation of authority and responsibility. It is not a technical or financial document.

3. Which of the following is a valid human-only activity in a healthcare AI integration?

A) AI generates a preliminary diagnosis; a doctor reviews and makes final diagnosis
B) AI automatically prescribes medication
C) AI determines patient eligibility for a clinical trial
D) AI writes the patient's discharge summary without human review

Feedback: A is correct. The doctor's final diagnosis and prescription are human-only. Options B, C, and D delegate consequential decisions to AI without human final authority, which is inappropriate.

4. What is a "stop authority"?

A) The person who decides to end the project
B) The person who can halt the workflow if a critical issue arises
C) The person who stops the AI system from learning
D) The person who deletes all data

Feedback: B is correct. Stop authority is about pausing or halting the process to prevent harm. It is a key safeguard in responsible AI design.

5. Which of the following is a stop condition for Tutorial 2?

A) The workflow has more than 10 steps
B) There is no clear accountable owner for each step
C) The workflow uses both AI and non-AI steps
D) The workflow has a fallback pathway

Feedback: B is correct. Lack of clear accountability is a stop condition. The number of steps, mixed approaches, and fallback pathways are not stop conditions.

6. Why is it important to identify affected non-users?

A) To ensure they are included in the user training
B) To understand the broader impact of the AI integration and mitigate negative spillover effects
C) To increase the number of stakeholders for the map
D) To comply with data privacy laws

Feedback: B is correct. Non-users may be indirectly impacted, and their interests must be considered to ensure fairness and responsibility.

7. What is the role of an "escalation pathway" in a workflow?

A) To speed up the workflow by skipping steps
B) To provide a route for issues to be reviewed by higher authority when resolution is not possible at the current level
C) To replace the need for a fallback pathway
D) To enable the AI system to make decisions autonomously

Feedback: B is correct. Escalation moves issues upward when they cannot be resolved at the current level. It does not replace fallback; both are needed.

8. Which of the following would not be considered a human-only activity?

A) Final approval of a financial transaction
B) Reviewing and correcting an AI-generated summary
C) Performing a complex calculation using AI assistance
D) Making a hiring decision

Feedback: C is correct. Performing a calculation with AI assistance is not inherently human-only; the critical part is the decision or action based on that calculation. The human-only part is the final judgment and accountability.

15. Completion Checklist

I have reviewed the purpose and connection to the Capstone Project.
I have completed the retrieval practice questions.
I have studied the worked synthetic example and understand the stakeholder mapping, workflow design, and responsibility matrix.
I have completed the guided activity: identified stakeholders, mapped workflows, defined AI role, listed human-only activities, mapped authority, designed escalation/fallback, and checked stop conditions.
I have produced all required student products:
- Stakeholder map
- Affected-non-user analysis
- Current-state workflow
- Proposed-state workflow
- AI-role statement
- Human-only activity list
- PARTNER responsibility matrix
- Escalation and fallback pathway
- Text alternatives for all diagrams
I have saved all evidence in my evidence-retention box.
I have reviewed the common problems and corrected any issues.
I have confirmed that no stop conditions apply to my design.
I have reviewed the responsible-use boundaries and ensured compliance.
I have chosen a participation pathway and completed the tutorial accordingly.
I have ensured my products are accessible (text alternatives, clear structure).
I have completed the self-check questions and reviewed the feedback.

✅ Tutorial 2 Complete

You are now ready for Tutorial 3: Framework Selection and Integrated Analysis. Your workflow and responsibility design will inform your framework choices.

16. Section Summaries

Purpose

Map stakeholders, workflows, and responsibility to transform your proposal into a practical, accountable design.

Connection

This tutorial directly feeds into sections of your final report, especially stakeholder analysis, workflow, and responsibility.

Learning Outcomes

You can now identify stakeholders, map workflows, define AI role, allocate authority using PARTNER, and design escalation/fallback pathways.

Key Concepts

Stakeholder mapping, workflow mapping, PARTNER, human-only activities, escalation/fallback.

Retrieval Practice

Recalled key concepts to prepare for the activities.

Worked Example

Showed a complete stakeholder map, workflow diagrams, PARTNER matrix, and escalation/fallback for an Education stream project.

Guided Activity

Step-by-step prompts to build your own stakeholder map, workflows, and responsibility design.

Student Production

Produced nine concrete deliverables that will become part of your final capstone.

Evidence-Retention

Saved all products for later use.

Common Problems

Addressed issues like incomplete stakeholders, vague AI role, lack of accountability.

Responsible-Use Boundaries

Reinforced that consequential decisions must remain human, and AI outputs must be reviewable.

Participation Pathways

Offered three equivalent ways to complete the tutorial.

Accessibility

Guided you to ensure your designs are inclusive and provide text alternatives.

Self-Check

Tested understanding of stakeholder categories, authority, stop conditions, and human-only activities.

Completion Checklist

Confirmed all required work is done and no stop conditions are triggered.

📌 Looking ahead

In Tutorial 3, you will select and apply relevant frameworks to your problem, using the workflow and responsibility design as a key input.