Chapter 1
Start with the Business Problem and Use Case
Learning Objective
Learn how to begin a GenAI project by translating a business problem into a safe, measurable AI use case.
What it means
A successful GenAI project does not start with a model. It starts with a business problem. Teams often fail because they begin by asking which LLM to use instead of asking what decision, document, task, or workflow needs improvement. The first step is to define the problem, users, input data, expected output, risk level, and success metrics.
Why it matters
GenAI can summarize, extract, classify, reason, and generate content, but not every problem needs GenAI. A simple rules engine may be better for deterministic logic. GenAI becomes valuable when the input is unstructured, ambiguous, or language-heavy — such as clinical notes, emails, call transcripts, policy documents, and case summaries.
Healthcare Example
A healthcare operations team receives clinical documents and needs to identify missing information before a case can move forward. A GenAI solution can summarize the document, extract diagnosis/procedure details, and identify missing sections. However, a human or validation layer should review high-risk decisions.
Architecture Flow
Common Mistakes
- Starting with a model before defining the use case.
- Trying to automate high-risk decisions without human review.
- Using GenAI for simple deterministic rules.
- Not defining success metrics before building the prototype.
Interview Q&A
Q: How do you start a GenAI project?
A: I start with the business problem, data sources, user journey, expected output, risk level, and measurable success criteria. Then I decide whether GenAI, rules, RAG, or a hybrid approach is appropriate.
Q: When should you not use GenAI?
A: I avoid GenAI when the task is deterministic, when exact calculations are required, or when a simple rules engine can solve the problem more safely and cheaply.
Architect Takeaway
A strong architect does not force GenAI into every workflow. The right approach is to choose GenAI only where language understanding, summarization, extraction, or reasoning adds measurable value.