User Personas and Use Cases for AI Solutions

Overview
In the PMI Cognitive Project Management for AI (PMI-CPMAI) framework, the first phase is Business Understanding, which belongs to the Identify Business Needs and Solutions domain. Within it, user personas and use cases are the main tools for turning a vague wish such as "we want AI" into a defined problem. That problem must be measurable, centred on real users, and actually solvable with AI. This guide explains why these tools matter, what they are, how they work in an AI project, and how to answer exam questions about them.

1. Why User Personas and Use Cases Are Important
Research often cited in the CPMAI body of knowledge says that a large share of AI projects fail. The usual causes are not bad algorithms but a poor understanding of the business problem, the users, and the data. Personas and use cases address these risks directly.

Benefits
• They keep the solution focused on people, not technology. AI built for no specific user is "a solution looking for a problem." Personas make sure someone real will use the system and benefit from it.
• They define scope and success criteria. A well-written use case states the inputs, outputs, triggers, actors and expected outcomes. This makes it possible to set KPIs and acceptance criteria and to calculate ROI.
• They show whether AI is needed at all. Use cases show whether a task needs pattern recognition, prediction, personalization or other cognitive abilities. If not, rules-based automation or a process change may be enough.
• They drive data requirements. Knowing who uses the output, and when, tells the team what data is needed, at what frequency, and with what quality. This prepares Phase II, Data Understanding.
• They support trustworthy AI. Personas expose concerns such as explainability needs, bias risks, privacy expectations and the level of human oversight each user group needs.
• They improve adoption and change management. Users who see their pain points addressed are more likely to trust and use the system.
• They support go/no-go decisions. Use cases can be ranked by value, feasibility and risk. This helps the team start with small, achievable iterations, in line with CPMAI's iterative approach of "think big, start small, iterate often."

2. What They Are

User Persona
A user persona is a semi-fictional, research-based profile representing a group of people who will interact with, be affected by, or benefit from the AI solution. A typical AI persona includes:
• Name, role and job title (for example, "Priya, Claims Adjuster")
• Goals and motivations
• Pain points and frustrations with the current process
• Tasks and workflows, plus the decisions the persona makes
• Technical skill and comfort with AI
• Trust and explainability needs (does this person need to know why the AI made a recommendation?)
• Environment and constraints, such as device, time pressure, regulations and accessibility
• Success measures from the persona's point of view

Kinds of persona
• Primary personas are the direct users of the AI output.
• Secondary personas are indirect users or approvers, such as managers who read reports.
• Affected parties are people impacted by AI decisions who do not use the system. Examples are loan applicants scored by a model, or patients triaged by an algorithm. These people are especially important for ethics and bias analysis.
• Operational personas include data stewards, MLOps engineers and model monitors who maintain the system.

Use Case
A use case describes a specific scenario in which an actor (often a persona) interacts with a system to achieve a goal. In AI projects it is tied to a cognitive need. A typical AI use case includes:
• Use case name and ID
• Actor or persona
• Business problem or objective
• Trigger or precondition
• Main flow and alternate flows
• AI pattern applied (see below)
• Inputs (data) and outputs (prediction, classification, recommendation, generated content)
• Human-in-the-loop points
• Success metrics, both business KPIs and model metrics
• Constraints and risks, including ethical, regulatory and data-availability risks

The Seven Patterns of AI
CPMAI often links use cases to these seven patterns:
1. Hyper-personalization: treating each individual as unique, for example tailored recommendations.
2. Recognition: identifying objects, speech, images or text in unstructured data.
3. Conversation and human interaction: chatbots, voice assistants, generative interaction.
4. Predictive analytics and decision support: forecasting and guiding decisions.
5. Patterns and anomalies: fraud detection and outlier discovery.
6. Goal-driven systems: reinforcement learning and optimization, such as finding the best route or strategy.
7. Autonomous systems: acting with minimal human involvement.
Mapping each use case to one or more patterns clarifies the type of AI, its complexity, and its data needs.

3. How It Works in a CPMAI Project

Step 1: Identify the business problem. Start with the business need, not the technology. Ask what pain or opportunity exists and what it costs.

Step 2: Identify stakeholders and users. Use interviews, workshops, surveys, observation and journey mapping. Include end users, decision makers, affected parties and operational staff.

Step 3: Build personas. Combine research findings into a small number of representative personas. Validate them with real users so they rest on evidence rather than assumptions.

Step 4: Map current-state journeys. Record how each persona does the task today. Highlight bottlenecks, repetitive judgment tasks, delays and errors, since these are where cognitive technology could help.

Step 5: Write candidate use cases. For each pain point, describe how an AI capability would help the persona. Link each use case to an AI pattern.

Step 6: Check AI suitability. Ask whether the problem really needs AI. Signs of a good AI fit:
• The task needs learning from data.
• Rules are too complex or change too often to code by hand.
• Large volumes of data exist.
• Some error is tolerable.
If a simple rule or a process fix would work, AI may be the wrong answer.

Step 7: Assess feasibility and value. For each use case, consider:
• Business value and ROI
• Data availability and quality
• Technical feasibility
• Organizational readiness
• Ethical and regulatory risk
• Time to value

Step 8: Prioritize and scope. Choose a small, high-value, feasible use case for the first iteration. Avoid "boiling the ocean." Define the minimum viable scope and success criteria.

Step 9: Define success metrics. Link business KPIs (cost reduction, cycle time, customer satisfaction) to model metrics (accuracy, precision, recall). Also set the acceptable performance threshold for each persona.

Step 10: Feed later phases. Personas and use cases shape:
• Data requirements (Phase II)
• Data preparation (Phase III)
• Model selection (Phase IV)
• Evaluation criteria (Phase V)
• Operationalization and monitoring (Phase VI)
Revisit them at each iteration.

Worked Example
Business need: An insurer has slow claims processing.
Primary persona: Priya, a claims adjuster with 8 years of experience. She handles 40 claims a day and spends most of that time reviewing photos and documents. She distrusts "black box" tools and needs reasons she can use to justify her decisions.
Affected party: The policyholder, who wants quick, fair decisions.
Use case: "Automated damage assessment." A policyholder uploads photos. A recognition model estimates damage severity and flags possible fraud (the patterns-and-anomalies pattern). The system gives Priya a recommendation with a confidence score and highlighted image regions. Priya approves or overrides it, keeping a human in the loop.
Metrics:
• Business: 30% reduction in cycle time.
• Model: at least 90% agreement with expert adjusters.
• Fairness: no disparity across regions.
Data needs: historical claim photos labeled with final outcomes.

4. Common Pitfalls
• Starting with a technology ("Let's use a large language model") instead of a user need.
• Generic personas based on assumptions rather than research.
• Ignoring affected parties who never use the system.
• Use cases with no measurable success criteria.
• Choosing use cases for which no suitable data exists.
• Scoping too broadly at first instead of iterating.
• Forgetting the trust, transparency and explainability needs of each persona.
• Treating personas as static documents instead of revisiting them each iteration.

5. Exam Tips: Answering Questions on User Personas and Use Cases for AI Solutions

Tip 1: Business need first, technology last. If an answer option jumps straight to selecting an algorithm, buying a platform or building a model before the problem and users are understood, it is almost always wrong. The CPMAI order is business understanding, then data understanding, then preparation, then modeling.

Tip 2: Look for the option that involves real users. Answers involving interviews, workshops, observation or validating personas with actual stakeholders usually beat answers based on assumptions or on the sponsor's opinion alone.

Tip 3: Ask "Is AI the right fit?" Some scenarios describe problems a simple rules engine or process change could solve. The best answer may be to question the need for AI or to recommend a non-AI solution. CPMAI values honest go/no-go decisions.

Tip 4: Prefer small, iterative scope. When asked which use case to pursue first, choose the one with clear value, available data, manageable risk and a quick win. Avoid the most ambitious, enterprise-wide option.

Tip 5: Connect use cases to measurable outcomes. Good answers define success in business terms (KPIs, ROI) and link them to model performance thresholds. Be wary of options that define success only as "high accuracy."

Tip 6: Remember data feasibility. A brilliant use case without data is not feasible. If a question mentions missing, poor-quality or inaccessible data, the right answer often involves reassessing or re-prioritizing the use case.

Tip 7: Include affected parties and ethics. Watch for scenarios where AI decisions affect people who are not users, such as applicants, patients or citizens. Correct answers consider bias, fairness, transparency, privacy and human oversight for these groups.

Tip 8: Recognize the seven AI patterns. Expect questions such as "A retailer wants to tailor offers to each shopper. Which pattern applies?" The answer is hyper-personalization. Learn short examples of each pattern.

Tip 9: Explainability depends on the persona. A data scientist may accept a complex model. A regulator, loan officer or doctor may need explanations. If a question emphasizes user trust or regulation, choose answers that prioritize interpretability and a human in the loop.

Tip 10: Use the PM's perspective. You are the project leader. Choose actions such as facilitating, clarifying, documenting, validating and aligning stakeholders. Avoid options where the PM makes technical decisions alone or skips stakeholder alignment.

Tip 11: Watch for scope creep and vague requirements. When stakeholders request "AI for everything," the best response is to break the request into specific, prioritized use cases tied to personas.

Tip 12: Iteration means revisiting. If new insights appear in later phases (for example, data reveals a different user behaviour), the correct move is often to return to Business Understanding and refine the personas and use cases. Pushing ahead is usually wrong.

Tip 13: Eliminate extreme or absolute answers. Options with "always," "never," "fully automate immediately" or "skip stakeholder review" are usually distractors.

Sample Question 1
Question: A hospital executive asks the PM to "implement AI to improve patient care." What should the PM do FIRST?
A) Select a deep learning vendor
B) Gather historical patient data
C) Identify user personas and specific use cases with clinicians and patients
D) Build a prototype chatbot
Answer: C. Business understanding and user-centered problem definition come first.

Sample Question 2
Question: Three use cases have been identified. Use case X has high value but no available labeled data. Use case Y has moderate value, good data and low risk. Use case Z has high value but major regulatory concerns. Which should be pursued in the first iteration?
Answer: Y. It is feasible and low risk, and it delivers a quick win consistent with iterative AI delivery.

6. Quick Memory Aid: "PUMP-D"
• Personas: who uses or is affected by the AI?
• Use cases: what scenario and goal?
• Metrics: how will success be measured?
• Pattern: which of the seven AI patterns applies?
• Data: is the needed data available and suitable?

Summary
User personas tell you who the AI serves and what they need, including trust and explainability. Use cases tell you what the AI must do, in what scenario, with what data, and how success is measured. Together they form the basis of CPMAI's Business Understanding phase. On the exam, favour answers that are user-centered, business-first, data-aware, ethically conscious, measurable and iterative.