There are fewer moments that need sound
If a button changes colour and position the moment it is pressed, the user already knows what happened. Repeating that change with a sound turns feedback into duplication. Actions in which hand and eye move together—scrolling, moving a cursor, switching tabs—usually work better in silence.
A sound can do more when it confirms that a send has finished, a slow task has completed, or a state has changed while the screen was out of view. Its value is highest not at the action itself but where certainty about the result is needed.
Timing comes before timbre
The same short tone feels like a control response when it arrives instantly at a press, and like a completion signal when it arrives after processing. Too early, and it promises an outcome that has not happened; too late, and the user has already looked elsewhere. Product sound is less about finding a beautiful note than deciding at which end of a state that note belongs.
Duration follows the same rule. A brief response that follows the finger and a tail that confirms completion do not need to last the same length. Whether the user should act next or wait should change the beginning and end of the sound.
One product, several signals
A product contains clicks, successes, warnings, errors, and background updates. Give them all the same bright sound and their meanings disappear. Give every state a wholly different effect and the product becomes a room of notifications competing with one another.
Separate the signals a person must notice from the ones the screen already makes clear. Then distinguish the weight of success, caution, and error among the remaining signals. Only after that order is in place can an individual sound support the system instead of breaking it.