Calm teams do not collect evidence to feel informed. They collect it to change the next decision.
Concept of the week · Let Evidence Decide
A test is only useful when it changes what the team does next.
Research becomes calm when evidence has somewhere to go: a product decision, a deferred risk, or a deliberate no. Under pressure, loose findings create verification burden: people reread notes, defend memories, and argue from preference when the team should be reducing uncertainty.
So we trace every meaningful observation to the next decision it changes.
In a formative usability test for a patient monitor, a nurse reaches the confirmation screen for an alarm profile, pauses, and asks whether the setting will apply to the current patient or the whole bed. Today, that might be captured as “some users were unsure at confirmation.” It is accurate, but too soft; it does not tell design, product, engineering, or risk what needs to change. Tomorrow, the team writes it differently: observation — state boundary unclear; workload — nurse must verify scope before acting; decision — confirmation must name patient, bed, profile, and persistence; next check — repeat the task without clarification. The finding no longer floats as research output. It becomes a design, labeling, and test decision.
In an information technology [IT] admin console, the same failure occurs when an operator hesitates to save a global configuration, feature flag, or access rule. Calm teams do not argue whether hesitation “matters”; they ask which decision it changes and how they will know the change worked.
One thing to remember: Every unowned finding turns into extra triage, implementation ambiguity, regression cases, and later support work.
How teams ship this: Add a required decision status to every usability finding — change now, monitor, reject with reason, or defer with owner and review date — and block handoff until one is selected.
Asset & resource
Finding-to-Decision Card
Use one card per meaningful observation:
Observed behavior — What did the person actually do, say, miss, repeat, or hesitate over?
Task context — Where in the workflow did it happen?
Workload created — What extra checking, recall, rework, or uncertainty did it create?
Decision changed — What design, labeling, risk, training, or test decision does this affect?
Follow-up proof — What would show that the change worked?
Continuous Discovery Habits by Teresa Torres. Read it for the discipline of connecting customer input to choices a product trio can own, especially through Opportunity Solution Trees.
Light wisdom & reflection
“Research is formalized curiosity. It is poking and prying with a purpose.”
Where is your team still collecting evidence without naming the decision it should inform?
Mindful practice
Outside work, this often shows up in small decisions that linger because we keep gathering opinions instead of choosing a next step.
What have I already observed enough to act on?
What decision am I avoiding by calling it “more information”?
Next time you ask for one more opinion, notice whether you need evidence — or permission.
Bonus chapter: AI can prepare the evidence trail, not own the decision trail
Artificial intelligence can speed up usability research, but it can also make weak synthesis appear more confident than it is. A transcript summary is not yet a finding. A cluster of repeated phrases is not yet a decision. A generated recommendation is not yet a safe product change.
Used effectively, AI can reduce documentation overhead associated with the Finding-to-Decision Card. It can draft observation summaries, pull supporting quotes, compare sessions, find repeated hesitation points, and suggest possible workload categories such as extra steps, hidden checks, state loss, or verification burden. That is useful work, especially in formative studies where teams need to move from one prototype round to the next.
But in healthcare and regulated systems, AI should not own the final interpretation. The team still needs to check the source recording, confirm the use context, assess risk, and decide whether the evidence changes labeling, interaction design, training, risk control, or verification. AI can prepare the evidence trail; humans must own the decision trail.
A calm AI workflow would be simple: summarize the session, extract candidate observations, map each one to a possible decision, then require human review before anything enters the design backlog. The goal is not more automated insight. The goal is less unowned evidence.
Best,
Andreas Walden
Share this with someone designing for clarity.