AI Products: What Users Actually Want
Built AI features for 6 different products. Most failed. Here’s what I learned.
What We Thought Users Wanted
- Chat interfaces for everything
- AI-generated content
- Automated decision-making
- Conversational UI
What Users Actually Wanted
- Faster ways to do existing tasks
- Confidence in AI suggestions
- Easy ways to override AI
- Clear value proposition
The Pattern That Works
AI as assistant, not replacement.
Don’t: “AI writes your code” Do: “AI suggests completions, you decide”
Don’t: “AI generates your report” Do: “AI drafts report, you edit and approve”
Users want augmentation, not automation.
The Features That Flopped
Fully automated workflows. Users didn’t trust them. Needed human checkpoints.
Chat for everything. Most tasks are faster with traditional UI. Chat adds friction.
AI-generated content without editing. Users wanted control over final output.
The Features That Worked
Smart suggestions. AI proposes, user accepts/rejects. Fast and clear.
Context-aware assistance. AI understands what you’re doing and offers relevant help.
Automated busy work. AI handles repetitive tasks users hate. Clear time savings.
Key Lessons
Start with pain points. Don’t add AI because it’s cool. Solve actual problems.
Make AI optional. Some users won’t trust it. That’s okay. Give them traditional options.
Show confidence scores. When AI is uncertain, say so. Users appreciate honesty.
Allow easy overrides. AI will be wrong sometimes. Make it trivial to correct.
Measure time saved. That’s the metric users care about. Not accuracy or sophistication.
The Test
Before building an AI feature, ask:
Does this save time? Is the value obvious? Can users trust it? Is there an escape hatch?
If any answer is “no,” rethink the feature.
The Bottom Line
Good AI products make users more effective. They don’t try to replace users.
Build augmentation, not automation. Users will thank you.