Why delay doesn't only cost you responses
After a few days people don't remember details, they remember an overall impression. So even those who do answer write something vaguer and more extreme: the peak survives — the thing that annoyed or delighted them — and everything in between disappears, which is the actionable part.
It's why late feedback always looks polarised: it isn't that people split, it's that after a delay only those with a peak to report bother answering.
The right moment for the three typical cases
After support: as soon as the conversation closes, within the hour. Timing matters so much here that one question sent immediately beats a full questionnaire sent tomorrow.
After a purchase: depends what you sold. If the experience *is* the purchase, immediately. If the value shows up in use, wait long enough for it to have been used — a week for an object, a month for a recurring service.
After a course or event: at the end, while people are still in the room or still have the recording open. The next day you collect half the answers, and they're from the ones who were angry or delighted.
Don't ask the same person twice
Every feedback request spends a little of the recipient's goodwill. Asking a second time soon after doesn't double the data: it halves the response to the next one, and the one after that never arrives.
If you need to measure over time, rotate recipients instead of pressing the same ones: a hundred different people each month are worth more than ten people asked ten times.
The one thing that matters as much as timing
Saying what happens next. Feedback asked for and then visibly ignored is worse than feedback never requested: next time that person won't answer, and they'd be right. One line on the thank-you screen is enough — «we read every one, and the next round changes accordingly» — as long as it's true.
Ask the same day for experiences and after first use for products — and don't ask the same person twice.