The number is about saturation, not statistics
Interviews are qualitative, so the goal is not a big enough sample to be significant. The goal is saturation: the point where new conversations stop surprising you. When you can finish the buyer's sentences and the same complaints, workarounds, and priorities keep coming up, you have your signal. For one buyer segment, that often lands somewhere between eight and fifteen conversations.
Quality beats quantity, by a lot
Five well-run interviews with the right people beat thirty bad ones with the wrong crowd. Bad interviews lead the witness, pitch the idea, and collect polite encouragement. Good ones ask about the buyer's real past behavior, not their hypothetical future enthusiasm. That is the core of the Mom Test: ask about what they already do and have already paid for, not whether they like your idea.
Talk to one segment at a time
Fifteen interviews spread across five different buyer types is really three interviews per segment, which tells you almost nothing. Pick one narrow ideal customer profile and saturate it before you widen. If you are validating two segments, you need roughly the full count for each, since their pains and language rarely match. Narrow first, then expand once one segment is clear.
When you need more (or can get away with fewer)
Go higher when the segment is broad, the answers conflict, or the purchase is high-stakes and complex, like enterprise buying. You can stop sooner when the pattern is screaming at you by interview six. The signal you want is not a magic count, it is consistency: the same problem, described the same way, by people who already spend time or money trying to solve it.