Design for AI: How to Spot AI Opportunities in Your Product (Framework + Prompts)
The 3-Step AI Opportunity Framework: find the right AI use cases in your product, map user needs to AI capabilities, and choose the right AI experience type.
TL;DR: Wondering where to add AI in your product? This guide shows designers, PMs, and founders how to find real AI opportunities: turn user needs into "How Might We" questions, map them to what AI does well (automate, generate, summarize, predict, personalize), then use the included prompt to test each idea for feasibility. Plus the four AI experience patterns (assistive, agentic, hybrid, ambient) so you choose the right one from the start.
Where Can We Add AI?
Everyone is being told to add AI. Leadership wants it, competitors are already doing it, and somewhere on your roadmap there's a vague line item that just says "AI."
But designing for AI isn’t just about adding a chatbot. It's a shift in how we think about control, interaction, and expectations. And the strategy around how do it well is less about the technology and more about the decisions around it.
If you’re building products powered by AI or just exploring, this guide gives you a 3-step framework and prompt to find real AI opportunities.
You’ll learn how to:
Spot the right use cases for AI by mapping user needs to AI capabilities
Understand what AI is actually good at
Choose the right type of AI experience: assistive, agentic, hybrid, or ambient
Turn user research into a grounded, buildable list of AI opportunities
You might like this one! A practical reference for designers, product managers, and founders: 100 Principles for Designing AI Products 💫
Start With the People, Not the Tech
I get this question all the time from fellow designers, mentees, clients, and product leaders: “How do I know where to add AI?”
Start with:
“What pains am I solving for my customers?”
“Where is my user stuck?”
”Can AI solve this in a better way?”
Don't think of AI as a feature, but as a new way to remove friction, add clarity, or unlock outcomes that used to be too expensive, slow, or complex.
Look for what people actually do, what frustrates them or slows them down.
What they wish was easier, faster, clearer, more delightful.
Problems and pain points
Repetitive or complicated flows
Tasks that are confusing, slow, or manual
Moments where people keep asking for help
Things users wish were easier, faster, or more personal
Understand AI Capabilities
AI tools are powerful, but they need the right data to work. At a high level, AI models learn from data like text, images, audio, video, logs.
Here are the core things AI can do well:
Automate: Handle repetitive tasks like tagging, sorting, formatting, organizing
Generate: Create content (text, code, images, audio, video)
Guide: Help users navigate complex flows or make better decisions
Summarize: Turn long content into short, useful insights
Understand: Interpret text, visuals, or voice in more human ways
Predict: Spot patterns, flag what matters, forecast outcomes
Personalize: Adjust content or responses to feel more relevant to each user
You don’t need to know every model. But knowing these basic capabilities will help you ask better questions and create intentional experiences.
The 3-Step AI Opportunity Framework
AI is powerful, but it’s not always the right answer. Before jumping in, ask yourself:
Is this a problem that actually needs AI?
Where does AI give people superpowers they didn’t have before?
Some tasks are better handled with simple logic or automation rules.
That’s what this framework helps you find. The framework maps user needs to AI capabilities, so you spot the moments where AI actually earns its place.
Let's get started!
Step 1: Extract User Needs
Start from research, interviews, support tickets, behavioral data, or direct pain points.
What’s frustrating, slow, or confusing for users?
Where do people need more guidance, speed, or support?
Write each need as a clear, short statement.
Use this prompt to turn user pains into clear need statements:
Here are raw notes from my user research (interviews, support tickets, feedback):
[paste your notes here]
Turn these into a short list of clear user needs. For each one:
- Write it as one simple sentence, from the user's point of view
- Focus on the real problem, not a solution
- Group similar pains into a single need
Important: Use only what's in my notes.
- Do not invent data, examples, or needs that aren't there
- Do not interpret meaning or assume anything I didn't say
- If something is unclear or missing, flag it instead of guessingImportant: Strong AI ideas start with real needs. The more your inputs come from actual research and data, the more useful and relevant your AI opportunities will be.
Step 2: Brainstorm “How Might We” Questions
For every user need, write a How Might We (HMW) question to spark creative solutions.
Examples:
User need: People don’t know which option to pick
HMW help them make faster, better decisions?User need: Users struggle to find key information
HMW surface what matters most at the right time?
Use this prompt to reframe those needs into How Might We questions:
Here is a list of user needs from my research:
[paste your user needs here]
For each need, write 2 to 3 "How Might We" questions. Each one should:
- Start with "How might we..."
- Stay broad enough to allow different solutions
- Focus on the user's goal, not a specific feature
Only work from the needs I listed. Don't add new needs, and don't assume a solution inside the question.Important: A good How Might We question reframes a need as a possibility. The broader you frame it, the more space you create for AI to add real value.
Step 3: Explore Relevant AI Capabilities
Explore what AI can do to solve these problems. Use this prompt with your AI tool of choice to identify realistic ways AI can help.
Provide the context! Paste in the user needs and HMW questions from Steps 1 and 2, and add context about what you're working with: your product, your users, and your current tech or constraints. The more real context you give it, the more grounded and buildable the suggestions come back.
Use this prompt to map your needs to realistic AI capabilities:
I’m working on identifying valuable and realistic AI opportunities for our product/solution.
Below is a list of real User Needs and corresponding How Might We (HMW) questions that we gathered through research.
For each pair, suggest a few ways AI could help, using practical, feasible capabilities like:
- Summarization
- Prediction
- Personalization
- Recommendation
- Classification
- Anomaly detection
- Natural language understanding/generation
- Automation of repetitive tasks
- Data insights or visualization
For every suggestion, include:
- AI Capability being used
- Brief description of how the AI would work and solve the problem
- A Feasibility & Complexity Score (Low, Medium, or High) based on current AI maturity and likely integration effort
Avoid vague or overly futuristic ideas, stay grounded in what’s actually buildable today.
Return your response in a clean, structured format.Choose the Right Type of AI Experience
Understanding what kind of AI you’re building helps shape the experience from the start. Is your AI meant to automate a task fully or augment the user’s abilities?
Here are 4 common patterns:
Assistive AI
Feels like a helpful assistant sitting beside you, ready to jump in with suggestions, shortcuts, or clarifications.
Feels like: a guide, helper, or co-pilot
Ideal for: support roles, writing help, suggestions, light automation
Examples: Grammarly, Notion AI, Google Docs Smart Compose
Agentic AI
Feels more like a junior employee than a co-pilot, you delegate a task, and it takes action on its own.
Feels like: a decision-maker or autonomous agent
Ideal for: complex execution, planning, task generation
Examples: GitHub Copilot, AutoGPT, AI task agents (e.g. Adept)
Hybrid AI
Feels like a tool that adapts between guiding and acting, sometimes it helps, sometimes it takes over.
Feels like: a shape-shifter, blending assistant and agent
Ideal for: workflows where AI supports and occasionally drives
Examples: Notion AI, Google Gemini, AI copilots in productivity tools
Ambient AI
Feels like the system just knows, subtle and often invisible, working in the background without needing prompts.
Feels like: an intelligent observer or invisible guide
Ideal for: personalized suggestions, prioritization, insights
Examples: Spotify recommendations, Gmail priority inbox, AI-powered dashboards
Loop In the Right People Early
You don’t have to figure this out on your own.
Collaborate early with:
Your developers
A data scientist or ML engineer
External AI consultants
Important guiding questions:
Have we considered the people, culture, and context of use?
What’s the real value for the user?
What type of human-machine relationship do we need to build?
How will we measure success, short and long term?
Are we building for trust (clarity, control, and confidence)?
What data does the AI produce, and how will people use it?
What happens if the AI fails or behaves unexpectedly?
Can users understand how the AI works?
How will the experience evolve as the AI improves?
Are we protecting privacy and reducing bias?
Can this be misused or cause harm?
How will we test and monitor the AI over time?
How does it impact user behavior?
Thanks for reading! 🫶
I hope this helped bring more clarity about how to uncover real AI opportunities.
Drop a comment or share your approach. I'd love to hear it. 💬
More UX + AI Reads 👇🏽
Get the full library of AI skills, prompts, and frameworks! 💎
For designers and product people who want to thrive in the AI era:
The AI UX Risk Skill | The Idea Studio Skill | 100 Principles for Designing AI Products | Control Evaluation Prompt | Framework for Discovering Problems
+more!
New resources keep coming 👌
I’m Ileana Marcut, founder of Creative Glue Lab, a design + development studio focused on digital products, AI-native tools, and systems. I write UX+ AI, where I share practical insights at the intersection of UX, AI, and product strategy.
I appreciate the actionable advice and simple prompts here. It genuinely helps me see how AI can solve real-world pain points without getting lost in technicality