Feedback & Validation
When Asha first sketched Pantry Pal, eight friends looked at the mock-up and every one said "I'd totally use that." Then she asked how often they opened the recipe apps already on their phones. Almost never, it turned out. Kind words are not evidence.
The rule
Trust what people do more than what they say they will do.
Feedback is what testers tell you and what you see them do. Validation is evidence that the problem is real and your answer helps: someone changes what they do, or gives up something — time, effort, data, money — to use it.
Ways to collect behaviour instead of compliments:
- Ask about the past, not the future. "When did you last throw food away? What did you eat that night?" beats "Would you use an app that…?"
- Set the pass mark before the test. "At least 3 of 6 testers cook one suggested dinner this week." Decide it afterwards, and almost any result can be called a success.
- Separate the idea from the AI's quality. If people like the concept but dislike the ideas it produced, you have a prompt or model problem, not a product problem — and the reverse is also possible.
- Look for patterns. One tester's wish is a note. The same stall from three testers is a finding.
Then act on it and test again. Validation is a loop, not a verdict.
The shortcut
Pass mark → watch → count → decide. Write the pass mark, watch real use, count behaviour against it, then choose one of three: carry on, change something and re-test, or go back to Define.
Pass mark
Asha writes it down first: "At least 3 of 6 testers cook one suggested dinner within a week."
Why new people? The first group now knows what she hopes to hear, and their answers would lean that way.
Drills
Answer fast — trust the rule.
Which question gives the most reliable evidence?
Why write the pass mark before the test?
Testers like the Pantry Pal idea but say the suggested dinners are boring. What have you learned?
One tester out of six asks for a meal-planning calendar. What should Asha do?
Remember
- Trust what people do more than what they say they will do.
- Pass mark → watch → count → decide, with the pass mark written first.
- Trap: counting compliments as validation, or mistaking weak AI output for a weak idea.
