In one minute
Separate discovery and problem definition from solution exploration and delivery.
Diverge to gather perspectives, converge to define a defensible challenge; diverge again to develop options, then converge through testing. Each transition is a decision gate, not a promise that work moves only forward.
The problem it addresses
A team commits to a solution before it has understood whose problem it is solving.
A clinic wants an AI booking bot after complaints about long waits. Interviews reveal that unclear referral rules, not booking speed, drive most repeat calls.
When to use it
- Use Double Diamond when a team commits to a solution before it has understood whose problem it is solving.
- The team can examine: People affected, field observations, a decision sponsor and a testable outcome.
- A relevant situation is: A clinic wants an AI booking bot after complaints about long waits. Interviews reveal that unclear referral rules, not booking speed, drive most repeat calls.
When not to use it
Do not use it to replace evidence, accountable judgement or affected people's participation.
Treating the four phases as a waterfall or using a workshop vote as user evidence.
Inputs required
People affected, field observations, a decision sponsor and a testable outcome.
Step-by-step process
1. Bound the decision and name its owner.
Write the scope, decision owner and people who can veto or revise the result before starting the workshop.
2. Collect the necessary evidence and affected perspectives.
Collect and date the necessary evidence: People affected, field observations, a decision sponsor and a testable outcome.
3. Discover competing accounts and unmet needs.
Interview the people who make, receive and support the service; keep contradictory accounts and observation dates separate.
4. Define one challenge and list excluded interpretations.
Write the problem from the affected person's perspective and record why alternative explanations were set aside.
5. Develop several materially different responses.
Include a process, communication and technology option so the favoured product idea is not the only candidate.
6. Test small, including accessibility and unintended effects.
Use a small real-world trial with an agreed success signal, an accessibility check and a route to stop harm.
7. Deliver, measure and reopen the frame when evidence changes.
Record what was learned at the delivery gate and return to discovery if the original problem statement no longer holds.
8. Record the output, next test and review trigger.
Store the artifact with its evidence links, owner, bounded next test and date or trigger for review.
AI automation lens
AI may sort authorised notes and expose missing evidence. It must not invent observations, silently decide for affected people or present a generated map as validated.
Visual model
Text alternative: Discover competing accounts and unmet needs; Define one challenge and list excluded interpretations; Develop several materially different responses; Test small, including accessibility and unintended effects; Deliver, measure and reopen the frame when evidence changes.
Interactive example
Scenario: A clinic wants an AI booking bot after complaints about long waits. Interviews reveal that unclear referral rules, not booking speed, drive most repeat calls.
Worked answer: Reframe the challenge around referral clarity; test a clearer message and a human escalation route before automating booking.
Facilitation notes
Record disagreements before synthesising; ask whose evidence is missing; state who may revise the result.
Expected output
- An inspectable decision artifact, its assumptions and evidence, a responsible owner, a bounded next test and a review date.
- Deliver, measure and reopen the frame when evidence changes.
- A named owner, test and review date.
Common mistakes
Treating the four phases as a waterfall or using a workshop vote as user evidence.
Quality checklist
- The decision and system boundary are explicit.
- Affected people and alternative accounts are included.
- Each important claim has a source or is marked as an assumption.
- An owner, safeguard and disconfirming signal are named.
Template
Knowledge check
Question: What mistake would undermine this method in the scenario?
Answer: Treating the four phases as a waterfall or using a workshop vote as user evidence.
Related tools
References
- Double Diamond — method source (opens in a new tab). Accessed 2026-09-28.
- Independent practice reference (opens in a new tab). Accessed 2026-09-28.
Method profile
The method structures a discussion and a test; it does not establish causal proof or guarantee success.