Skip to main content
Next Step PM
Back to the library

Design · Add-on

Design for responsible AI and human oversight

Users need to know when AI is involved, how far to trust it and how to correct it. Design for mistakes, fairness and human oversight from the start.

About 2-5 daysFast-movingLast reviewed 5 November 2026

Who this applies to

Skills it builds
Design and user experienceTechnical and AI fluency
PM types
AI
Product types
Data or AI product
Starting points
All
Ways of working
All
Experience
All
Size of organisation
All

Steps, by experience

Beginner (under 1 year)

  1. Tell users clearly when they are dealing with AI.
  2. Show how confident the AI is, or why it gave an answer, where that helps.
  3. Let users correct, override or report wrong answers.
  4. Decide where a human must check before anything important happens.
  5. Test with a range of users to spot unfair results.
  6. Plan what the product does when the AI fails or is unavailable.

Intermediate (1 to 4 years)

  1. Design transparency, control and feedback into the experience.
  2. Define human review points by risk.
  3. Test for fairness and failure.

Experienced (5+ years)

  1. Run a structured risk assessment using a recognised framework.
  2. Set red lines for uses you will not support.
  3. Build governance with legal, security and ethics colleagues.

Who can do this

Owner: Product manager

Can be delegated to

  • UX Designer: Design transparency, correction and fallback
  • Policy / Legal / Compliance: Assess legal and ethical risks
  • Data Scientist / ML Engineer: Test for bias and failure modes

Reviewed by: Policy / Legal / Compliance, Data Scientist / ML Engineer

Tool examples

Figma
Design error, low-confidence and fallback states, not just the happy path.
Confluence
Keep an AI risk register linked to design decisions.

Definition of done

  • AI use is clear to users
  • Users can correct or override
  • Human review points defined
  • Fairness and failure tested

Common pitfalls

  • Hiding that AI is involved
  • Designing only for correct answers
  • No human check on high-impact decisions

If you lead the team

What good looks like: Users know when AI is involved, can correct it, and people stay in control of important decisions.

Ask your PM

  • How do users know AI is involved, and how do they correct it?
  • Who checks for unfair or harmful outcomes?

Warning signs

  • AI makes important decisions with no human check.
  • No one owns fairness or harm.

Do these first

Sources