You generally want your peak detector to be faster than your sample rate from the ADC. (note: please write ADC [Analog to Digital Controller Converter], not ATD -- use the well-established acronyms). When you want to sample the peak within your sampling interval, then you know it's going to be ready and holding a peak for you. Your peak detector can be self-resetting (via a resistor across the cap, i.e. using the RC time constant) or it can be reset via a transistor across the capacitor, using an output signal from the micro after you've read the peak. If you use a self-resetting peak-detector then you need the ADC to sample it fast enough to get the reading before it decays. If you use a reset transistor, then your ADC can be slower than your peak detector and you can reset it after each reading.
To design the peak detector you need to know your input signal, and design it so that it is fast enough to track and hold the peaks that you want (i.e. choose the proper op-amp and cap)
Peaks are always going to be within some time period that you establish. You can wait 1 us or 1 second for a peak to arrive. It's completely up to you and usually depends on what you are sampling. You just need to make the peak detector hold it that long, and you sample it at the required rate.
You can't just decide that I'll wait for the peak to arrive and trigger the ADC when it does, because how long should you wait? You can't wait forever, so you have to wait some time period. This is going to be your ADC sample rate.
Yes, you can get fancy and do some tricky stuff with analog switches, holding caps and comparators to hold the previous peak value value and the current value and trigger the ADC using the comparator output when the current value is greater than the previous value. But you generally don't need that added complication, if your peak detector is fast enough you can just sample the peak value periodically.