In-Scope and Out-of-Scope AI Functionality

Introduction
In the PMI Certified Professional in Managing AI (PMI-CPMAI) framework, defining what an AI solution will and will not do is one of the most important early activities in the Identify Business Needs and Solutions phase. This phase corresponds to Phase I (Business Understanding) of the CPMAI methodology. In-scope and out-of-scope AI functionality is the explicit, documented boundary of an AI project. It states which business problems, cognitive capabilities, data, users and decisions the AI system will address, and which it deliberately excludes.

Why It Is Important
AI projects fail at a high rate. Common reasons include unclear objectives, scope creep, unrealistic expectations and a mismatch between what stakeholders imagine AI can do and what the data can support. Clear scope boundaries address these risks in several ways.

1. They prevent scope creep. AI generates excitement. Stakeholders often keep adding features, such as "Can it also predict X?" or "Can it chat with customers too?" Each addition may need new data, new models and new validation. Without firm boundaries, timelines and budgets grow out of control.

2. They manage expectations. Business leaders may believe AI is magic. Stating what is out of scope, such as fully autonomous decisions, real-time processing or certain languages, sets realistic expectations and protects trust.

3. They support the iterative, small-win approach. CPMAI favors starting small, delivering value quickly and iterating. A tight scope lets the team deliver a minimum viable AI solution, measure it against success criteria and then expand in later iterations.

4. They align data needs with the solution. Scope drives data requirements. If a capability is in scope, the team must confirm that suitable data exists. If the data does not exist, the capability may need to move out of scope. This link carries directly into Phase II (Data Understanding).

5. They address risk, ethics and compliance. Some functions carry regulatory, ethical or reputational risk. Examples include automated credit denial, medical diagnosis without human review and biometric identification. Explicitly excluding them, or keeping a human in the loop, reduces exposure and supports trustworthy AI.

6. They support go/no-go decisions. A clear scope lets the team judge AI feasibility and ROI. It also confirms that AI is the right approach at all, rather than conventional automation or rules-based logic.

What It Is
Scope definition for AI covers more than a traditional feature list. It typically includes the following dimensions.

- Business problem and objective: the specific problem the AI solves, for example "reduce invoice processing time by 40%."
- AI pattern(s): CPMAI uses seven patterns of AI. These are Recognition, Conversational/Human Interaction, Predictive Analytics and Decision Support, Patterns and Anomalies, Hyperpersonalization, Autonomous Systems, and Goal-Driven Systems. The scope states which pattern or patterns are included. Mixing too many patterns in one iteration is a warning sign.
- Functional capabilities: what the system will do, such as classifying documents, extracting fields or flagging anomalies.
- Level of autonomy: whether the AI recommends (decision support), acts with human approval (human-in-the-loop) or acts autonomously.
- Users and stakeholders: who uses the output and who is affected.
- Data boundaries: which data sources, time ranges, geographies, languages and formats are included.
- Operational boundaries: batch or real-time processing, volume, integration points and channels.
- Performance expectations: acceptable accuracy, precision and recall, and latency. AI is probabilistic, so 100% accuracy is never in scope.
- Explicit exclusions: features, use cases, populations or decisions intentionally left out, often with a note that they may be considered in future iterations.

How It Works: The Process

Step 1: Start with the business need. Identify the real pain point and the measurable business outcome. Ask: what problem are we solving, and how will we know we succeeded? Scope must trace back to business value.

Step 2: Determine whether AI is needed. CPMAI stresses that not every problem needs AI. If deterministic rules, traditional software or process change solve the problem, AI may be out of scope entirely. AI fits best where there is variability, pattern recognition or prediction from data.

Step 3: Identify the AI pattern(s). Map the need to one or more of the seven patterns. This clarifies the type of capability and the data and expertise required.

Step 4: Define the minimum viable scope. Choose the smallest valuable slice that can be delivered quickly. For example, start with invoice classification for one supplier region before expanding globally. CPMAI advises thinking big, starting small and iterating often.

Step 5: Check data availability and quality. For each candidate capability, confirm that enough relevant, quality, accessible and labeled data exists. Capabilities without adequate data are deferred or excluded.

Step 6: Assess risks, ethics and regulations. Identify functions that raise bias, privacy, safety, transparency or compliance concerns. Decide whether to exclude them, constrain them or add human oversight.

Step 7: Document in-scope and out-of-scope items explicitly. Record them in the project charter, AI solution definition or scope statement. Out-of-scope items should be written down, not merely implied.

Step 8: Get stakeholder agreement. Business owners, data teams, IT, legal and compliance, and end users should review and approve the boundaries.

Step 9: Manage changes through iteration. New requests go through change control or a backlog for future iterations, rather than being added mid-iteration. CPMAI is iterative, so out-of-scope items can return to scope in later cycles after re-evaluation, sometimes even after returning to Phase I.

Example
A bank wants AI to help with loan processing.

In scope (Iteration 1):
- Extracting data from submitted loan documents (Recognition pattern).
- Predicting default risk scores as decision support for human underwriters.
- Personal loans in one country, English-language documents.

Out of scope:
- Automated approval or denial without human review (regulatory and ethical risk).
- Mortgage and business loans (different data and rules; future iteration).
- A customer-facing chatbot (a different AI pattern; separate project).
- Real-time scoring at the point of application (infrastructure not ready).

Common Pitfalls
- Trying to boil the ocean with an overly broad first iteration.
- Including capabilities with no supporting data.
- Leaving out-of-scope items unstated, so stakeholders assume they are included.
- Promising full automation where human oversight is required.
- Defining scope by technology ("use deep learning") rather than by business outcome.
- Ignoring operationalization: where the model runs, how it is monitored and who acts on its output.

Exam Tips: Answering Questions on In-Scope and Out-of-Scope AI Functionality

1. Favor the smaller, focused scope. When an answer offers a narrow, achievable first iteration that delivers measurable value, it is usually correct. Answers that try to deliver everything at once are usually wrong.

2. Tie scope to the business need. The best answer traces functionality back to a defined business problem and success criteria. Distractors often focus on technology or exciting features.

3. Check data feasibility. If a scenario says data is unavailable, poor quality or unlabeled for a capability, the correct response is usually to exclude or defer that capability. Another valid response is to address the data gap first. Do not proceed as if the data exists.

4. Handle scope creep through process. If a stakeholder requests a new AI feature mid-project, the best answer usually involves one of these:
- assessing the request against the business objective;
- adding it to the backlog for a future iteration;
- following change control.
Immediately adding the feature is rarely correct. Flatly refusing without evaluation is also rarely correct.

5. Recognize when AI is not the answer. If the problem can be solved with simple rules or conventional automation, the correct answer may be to put AI out of scope.

6. Watch the level of autonomy. In high-risk domains such as healthcare, finance, hiring and legal, prefer answers that keep a human in the loop or limit the AI to decision support. Answers that make fully autonomous decisions in scope are usually wrong.

7. Prefer explicit documentation. Answers that explicitly document out-of-scope items and gain stakeholder sign-off are stronger than answers that leave boundaries implied.

8. Limit AI patterns per iteration. If a scenario combines several AI patterns in one initial project, the better choice is usually to focus on one pattern first.

9. Remember iteration. Out of scope does not mean never. CPMAI expects later iterations to revisit deferred functions. Choose answers that keep deferred items for future cycles.

10. Treat probabilistic outputs realistically. Any requirement for 100% accuracy or zero errors signals unrealistic scope. Prefer answers that define acceptable performance thresholds.

11. Know the phase. Scope definition belongs to Phase I (Business Understanding). If a question asks what to do before collecting data or building models, defining the business problem and scope is often the answer.

12. Use keywords to eliminate distractors.
- Correct answers tend to use words like iterative, minimum viable, measurable, stakeholder alignment, feasible, data availability and human oversight.
- Wrong answers tend to use words like all features, fully automate everything, skip, assume data exists or add immediately.

Summary
Defining in-scope and out-of-scope AI functionality turns a vague AI ambition into a focused, feasible and valuable project. It connects business needs to AI patterns, data realities, risk tolerance and iterative delivery. On the exam, choose answers that are business-driven, data-feasible, risk-aware, explicitly documented, stakeholder-approved and deliberately small to start, with expansion through later iterations.