The average hides the problem
Nine leads answered in two minutes and one answered eighteen hours later average out to under two hours — a number that sounds fine and describes nothing that happened. The nine were never at risk. The tenth was the one you lost.
Track the median instead: half your leads were answered faster, half slower. It moves only when your actual behaviour changes, and it cannot be rescued by a handful of fast replies.
Measure from submission, not from when you saw it
The clock starts when the customer hits send, not when you opened the app. This feels unfair and is exactly the point: the customer's experience of your speed does not care when the notification surfaced.
If your marketplace shows you a lead timestamp, use it. The gap between that timestamp and your first message is the whole measurement.
Split the number by time of day
Almost every home-services business has two very different response times wearing one number: a good one during working hours, and a terrible one outside them. Averaging them together produces a figure that describes neither and hides the half you can actually fix.