The most expensive sentence in product design is "we'll test it after launch". A flaw found in a prototype costs an afternoon; the same flaw in production costs a sprint, a hotfix and a slice of user trust. Early testing is the cheapest insurance a project can buy.
Five users beat fifty opinions
You don't need statistical significance to find big problems — five representative users surface the vast majority of critical issues. Recruit for behaviour (shops online weekly) rather than demographics, and run sessions remotely to keep costs near zero.
Give tasks, not tours
"Find a blue sweater under $50 and check out" reveals everything; "what do you think of this page?" reveals nothing. Stay silent while they work, and note where they hesitate — hesitation is the bug report.
"Test the riskiest assumption first — if checkout is the bet, test checkout before the homepage."
Fix, then re-test the fix
A test without a fix loop is theatre. Rank findings by impact, redesign the top three within days, and run a second round on the same tasks. Two small rounds routinely outperform one big study — and keep momentum with stakeholders who watch clips, not reports.