Module 7 · AI in the Real World — Use Cases & Value
Patterns by Function: Operations, Marketing, Support, Finance
70 min
Learning objectives
- Match common AI patterns (classification, prediction, generation, extraction) to business functions
- Cite concrete, realistic use cases in operations, marketing, customer support, and finance
- Explain which patterns favor automation versus augmentation in each function
A handful of patterns power most use cases
Across every function, the same few AI patterns reappear: classifying items into categories, predicting a number or likelihood, extracting structured information from messy text or documents, generating draft content, and recommending the next best option. Learning to recognize the pattern behind a request makes any new use case easier to scope.
Use case — A specific, bounded task where AI produces a defined outcome — narrow enough to measure and own.
Operations
Operations is rich in repetitive, data-heavy tasks. AI commonly forecasts demand, predicts equipment failure, optimizes routing and scheduling, and extracts data from invoices and shipping documents — freeing staff from manual data entry and reactive firefighting.
| Use case | Pattern | Typical mode |
|---|---|---|
| Demand forecasting for inventory | Prediction | Augmentation (planner decides) |
| Predictive maintenance on machines | Prediction | Augmentation (schedule a check) |
| Invoice / document data extraction | Extraction | Automation, review low-confidence cases |
| Delivery route optimization | Optimization | Automation (dispatcher can override) |
Marketing
Marketing uses AI to segment customers, predict who is likely to churn or convert, personalize recommendations, and generate first-draft copy and creative variations for human editors to refine.
| Use case | Pattern | Typical mode |
|---|---|---|
| Customer segmentation | Classification / clustering | Augmentation |
| Churn / propensity prediction | Prediction | Augmentation (targeting list) |
| Product recommendations | Recommendation | Automation |
| Draft ad / email copy variants | Generation | Augmentation (human edits & approves) |
Customer support
Support sees high volumes of repetitive questions, making it one of the most popular places to apply AI. Systems route tickets, suggest answers to agents, summarize long conversations, and answer routine questions directly — with escalation to humans for anything complex or sensitive.
| Use case | Pattern | Typical mode |
|---|---|---|
| Ticket routing / prioritization | Classification | Automation (low-confidence to a human) |
| Agent answer suggestions (copilot) | Generation / retrieval | Augmentation |
| Conversation & case summarization | Generation | Automation with light review |
| Self-service FAQ assistant | Generation / retrieval | Automation with human escalation |
Watch out
Customer-facing generative assistants can produce confident but wrong answers (hallucinations). Ground them in approved knowledge sources, constrain their scope, and provide an easy path to a human for anything they're unsure about or that carries legal/financial weight.
Finance
Finance combines large transaction volumes with high error costs, so it favors a mix of automation for detection and augmentation for decisions. AI flags anomalies and likely fraud, extracts data from financial documents, and drafts analyses — but humans approve consequential actions.
| Use case | Pattern | Typical mode |
|---|---|---|
| Fraud / anomaly detection | Classification | Augmentation (analyst reviews flags) |
| Invoice & receipt processing | Extraction | Automation with exception review |
| Cash-flow / revenue forecasting | Prediction | Augmentation |
| Drafting financial summaries | Generation | Augmentation (analyst verifies) |
Example — Same pattern, different function
Ticket routing in support and fraud flagging in finance are both classification problems — sort each item into a category. A team that has built one has a head start on the other; a key difference is error tolerance (other differences include how rare the positive cases are and how adversarial the domain is) — which is why routine support routing is often automated, with low-confidence cases reviewed, while fraud flags typically go to an analyst.
Map any request to its pattern (classify, predict, extract, generate, recommend) and its error tolerance. Those two facts largely determine feasibility and whether to automate or augment.
Knowledge check
Quick practice — not part of your exam score.
Routing an incoming support ticket to the correct team is an example of which AI pattern?
Why is fraud detection in finance typically run as augmentation (analyst reviews flags) rather than full automation?
A marketing team wants AI to write first-draft email copy that staff then edit. This is best described as:
Sign in to track your progress and mark lessons complete.
Sign in