Topic 3: How to work effectively with LLMs
How to work effectively with LLMs
Large Language Models can be incredibly useful for product and business analysis work, but the quality of their output depends heavily on how you work with them.
An LLM works best when you treat it less like a search engine that gives you an instant answer and more like a working partner that you direct, question and refine. Give it a clear task, enough context to understand the situation, and boundaries around what you need.
Start with clarity and context
A vague request usually produces a broad or generic response. Before asking an LLM to help, make three things clear:
- Task: What exactly do you want it to do?
- Context: What does it need to know about the product, users, problem or business situation?
- Boundaries: What should it consider, avoid or not assume?
For example, instead of asking “Create some user stories”, explain the feature, target user, business objective, known requirements and any constraints. This gives the LLM something meaningful to work with.
Tell the LLM what a useful answer looks like
Don’t leave the output format entirely to the model. Ask for the information in a form that helps you do your job.
Depending on the task, you might ask for:
- a summary or structured table
- user stories and acceptance criteria
- assumptions and unanswered questions
- options with advantages and disadvantages
- risks or dependencies
- recommendations with reasoning
This is especially useful when you want to take the output into a workshop, backlog, requirements document or stakeholder conversation.
Break complex work into smaller steps
One giant prompt is rarely the best way to tackle a complicated problem.
For example, when analysing customer feedback, you could first ask the LLM to identify recurring themes, then group the problems, then identify possible opportunities, and finally help prioritise them.
Working step-by-step gives you opportunities to inspect the thinking, add information and change direction before moving further.
Review and challenge the response
An LLM can produce an answer that sounds confident and convincing even when information is missing, or an assumption is incorrect.
Don’t just ask, “Is this correct?” Challenge the output:
- What assumptions have you made?
- What information might be missing?
- Which parts are uncertain?
- What alternative interpretation should we consider?
- What questions should I ask before making this decision?
This turns the LLM from simply a content generator into a useful tool for exploring and challenging your thinking.
Keep refining through conversation
Your first prompt doesn’t need to produce the final answer.
Add missing context. Correct assumptions. Ask follow-up questions. Request another option. Challenge weak reasoning. Ask the model to improve a particular section.
Think of the interaction as an iterative conversation, where each exchange improves the quality and usefulness of the result.
DSM Tip
Treat the LLM as a working partner, not an answer machine.
Give it direction, work with its response, and apply your own professional judgement before using the output in your product or project.
