Mapping Business Problems to AI Patterns

Overview
In the PMI Cognitive Project Management for AI (CPMAI) methodology, the first phase is Business Understanding. A core activity in this phase is mapping business problems to AI patterns. This means translating a real-world business need, such as reducing churn, detecting fraud, or automating document intake, into one or more of the Seven Patterns of AI. Once the pattern is known, the team can better judge data needs, feasibility, scope, success criteria, and risk.

Why It Is Important
Most AI projects that fail do so for business reasons rather than technical ones. Common causes are unclear problem definitions, misaligned expectations, the wrong technology, or scope that is too large. Mapping the problem to an AI pattern helps in several ways:
- It clarifies what is being built. Saying we want AI is vague. Saying we need a predictive analytics pattern to forecast which customers will churn in the next 90 days is actionable.
- It drives data requirements. Each pattern needs different data. Recognition needs labeled images, audio, or text. Predictive analytics needs historical outcome data. Anomaly detection needs a solid baseline of normal behavior.
- It supports the AI Go/No-Go decision. If the problem fits no AI pattern, or a simple rule-based or traditional automation solution would work, AI may not be justified.
- It enables reuse and benchmarking. Patterns let teams draw on known approaches, vendors, and lessons learned from similar projects.
- It helps with scoping and iteration. CPMAI favors small, iterative projects. Breaking a big vision into individual patterns allows a minimum viable, fast first iteration.
- It aligns stakeholders. Patterns give business and technical people a shared vocabulary.

What It Is: The Seven Patterns of AI
CPMAI uses seven patterns that describe what an AI system does, regardless of the algorithm underneath:

1. Recognition
The system identifies and understands unstructured data such as images, video, audio, handwriting, and text.
Examples: facial recognition, OCR of invoices, speech-to-text, identifying defects on a production line from camera images, classifying documents.
Keywords: identify, classify images, read, transcribe, detect objects, extract from documents.

2. Conversational and Human Interaction
The system interacts with people through natural language, in text or voice, in both directions.
Examples: chatbots, voice assistants, machine translation, content generation, summarization in a dialogue context.
Keywords: chat, talk, answer questions, understand intent, translate, natural language dialog.

3. Predictive Analytics and Decision Support
The system uses past and current data to predict future outcomes or help humans make better decisions.
Examples: demand forecasting, churn prediction, credit risk scoring, predictive maintenance, price optimization advice.
Keywords: forecast, predict, estimate, likelihood, score, recommend a decision to a human.

4. Patterns and Anomalies
The system finds patterns, relationships, or outliers in data, including items that deviate from normal behavior.
Examples: fraud detection, cybersecurity intrusion detection, finding unusual sensor readings, discovering customer segments, spotting data errors.
Keywords: unusual, outlier, suspicious, deviation, cluster, hidden relationship.

5. Hyperpersonalization
The system builds a profile of each individual and tailors content, offers, or experiences to them.
Examples: product recommendations, personalized learning paths, individualized medical treatment, tailored marketing.
Keywords: each customer, individual, tailored, recommend for this user, one-to-one.

6. Goal-Driven Systems
The system learns through trial and error, often with reinforcement learning, to find the best way to reach a goal.
Examples: game-playing AI, dynamic resource allocation, real-time bidding strategies, scenario simulation and optimization.
Keywords: optimize toward a goal, find the best strategy, learn by trial and error, maximize reward.

7. Autonomous Systems
The system acts on its own with little or no human involvement, both physically and virtually.
Examples: self-driving vehicles, autonomous drones, warehouse robots, self-managing software agents, autonomous document processing that acts with no human in the loop.
Keywords: without human intervention, self-driving, independent action, minimize human labor.

How It Works: The Mapping Process
Step 1: Define the business problem precisely. State the pain point, who it affects, and its cost. Example: claims processors spend 40 percent of their time manually reading scanned forms.
Step 2: Define the desired outcome and success metrics. Use business measures such as reduced processing time, increased accuracy, lower cost, or higher revenue, not just model accuracy.
Step 3: Ask whether AI is actually needed. If the logic can be fully written as fixed rules, traditional automation or programming may be cheaper and more reliable. AI fits best when the problem needs learning from data, handles variability, or involves unstructured inputs.
Step 4: Identify the pattern or patterns. Ask what the system must do: recognize, converse, predict, find anomalies, personalize, optimize toward a goal, or act autonomously. Look at the verbs in the problem statement.
Step 5: Decompose multi-pattern solutions. Many real solutions combine patterns. A customer service bot may use Conversational (dialog), Recognition (reading an uploaded receipt), and Hyperpersonalization (tailored answers). CPMAI advises tackling one pattern per iteration where possible, so you can deliver value early and lower risk.
Step 6: Assess feasibility for each pattern. Check data availability and quality, required accuracy, human-in-the-loop needs, ethical and regulatory constraints, and ROI.
Step 7: Make the AI Go/No-Go decision and scope the first iteration. Choose the smallest valuable piece to build first, then move into Data Understanding.

Worked Example
A bank says: we want AI to stop losses from suspicious card transactions and explain decisions to customers via chat.
- Flagging suspicious transactions maps to Patterns and Anomalies.
- Scoring the probability of fraud for an analyst maps to Predictive Analytics and Decision Support.
- Explaining decisions to customers in chat maps to Conversational.
A CPMAI-aligned approach would start with the anomaly detection piece, since it has the highest business value and clear data (transaction history). The chatbot would follow in a later iteration.

Common Pitfalls
- Starting with a technology, such as we want to use generative AI, instead of a business problem.
- Trying to deliver every pattern at once, which creates a boil-the-ocean project.
- Confusing Predictive Analytics with Hyperpersonalization. Prediction concerns an outcome. Hyperpersonalization concerns treating each individual uniquely.
- Confusing Autonomous with Goal-Driven. Autonomous removes humans from the action. Goal-driven learns an optimal strategy through trial and error.
- Ignoring whether data exists to support the chosen pattern.
- Using AI where a simple rule would work.

Exam Tips: Answering Questions on Mapping Business Problems to AI Patterns
1. Memorize the seven patterns and their signature verbs. Recognition means identify or understand unstructured data. Conversational means interact in natural language. Predictive means forecast or support decisions. Patterns and Anomalies means find outliers or relationships. Hyperpersonalization means tailor to an individual. Goal-Driven means find the best path to a goal through trial and error. Autonomous means act without humans.
2. Read for the primary objective. Scenarios often include distracting details. Ask what the core business outcome is. If a scenario mentions images but the goal is forecasting demand, the main pattern is Predictive Analytics.
3. Watch for key qualifiers. Words like each user or individual point to Hyperpersonalization. Unusual or suspicious point to Anomalies. No human involvement points to Autonomous. Reward or trial and error point to Goal-Driven.
4. Business first, technology second. In CPMAI, the correct answer nearly always begins by clarifying the business problem and success criteria before picking algorithms, tools, or vendors. Rule out answers that jump straight to model selection.
5. Consider whether AI is needed at all. If a scenario describes fixed, well-defined logic, the best answer may be that a non-AI solution is more appropriate. CPMAI values honest Go/No-Go decisions.
6. For multi-pattern scenarios, choose the iterative answer. When asked how to proceed, pick the option that breaks the solution into patterns and delivers the highest-value or most feasible pattern first.
7. Tie the pattern to data. Questions may ask what to check next. After identifying the pattern, the logical next step is to confirm that suitable data exists, which is the Data Understanding phase.
8. Distinguish similar patterns carefully. A recommendation engine for each shopper is Hyperpersonalization, even though it involves prediction. A chatbot that reads a photo uses both Conversational and Recognition. Choose the pattern that best matches the question's emphasis.
9. Eliminate extreme or premature answers. Options that promise full autonomy immediately, skip stakeholder alignment, or deploy before validating the business case are usually wrong.
10. Remember where this fits. Pattern mapping belongs to Phase I, Business Understanding. If a question asks when to identify AI patterns, the answer is early, before data preparation and modeling.

Sample Question
A retailer wants to reduce stockouts by knowing how many units of each product will sell next month at every store. Which AI pattern best applies?
A. Recognition
B. Predictive Analytics and Decision Support
C. Hyperpersonalization
D. Autonomous Systems
Answer: B. The goal is to forecast a future quantity. Nothing in the scenario involves unstructured data recognition, individual tailoring, or acting without humans.

Summary
Mapping business problems to AI patterns turns vague ambitions into well-defined, feasible, data-driven AI projects. Start with the business problem, decide whether AI is justified, match it to one or more of the seven patterns, break the work into iterations, and confirm data feasibility. On the exam, focus on the core objective, the keywords, and the CPMAI principle of business-first, iterative delivery.