Response time analysis

Response time analysis is the practice of measuring how long it takes to respond to incoming requests, then examining the full distribution of those times rather than a single average. A request that arrives at 09:00 and gets a first reply at 09:30 has a response time of 30 minutes. Across hundreds of requests you can compute an average, but the average alone hides the cases that hurt most.

The analysis part matters because response times are rarely evenly spread. A team can show a comfortable 20-minute average while one in twenty requests waits over two hours. Looking at percentiles, such as the median and the 90th and 95th, tells you what the typical customer experiences and what the unlucky tail experiences. That tail is usually where complaints, escalations and churn come from.

Start with the response time of each request, which is the response timestamp minus the request timestamp. The average is the sum of those individual times divided by the number of requests, but the analysis only becomes useful once you also compute the median and the upper percentiles. The median is the middle value when all response times are sorted, and the 95th percentile is the value below which 95 percent of responses fall.

Work an example. Ten requests have response times in minutes of 5, 6, 7, 8, 9, 10, 12, 15, 40 and 120. The average is about 23 minutes, dragged up by the two slow cases. The median is 9.5 minutes, which is what most requests actually felt like. The 95th percentile sits around 120 minutes. Three numbers tell a far truer story than the average on its own.

1. 1

   Capture both timestamps

   Record when each request arrived and when the first meaningful response was sent.

2. 2

   Compute per-request response time

   Subtract the request time from the response time for every request in the period.

3. 3

   Calculate the average

   Sum all response times and divide by the number of requests for the headline figure.

4. 4

   Add median and percentiles

   Sort the times and read off the median, 90th and 95th percentiles to see the distribution.

Knowing the 95th percentile crept up tells you there is a problem but not where it sits. A metric tree decomposes response time into the drivers that move it, so a slow tail traces back to a staffing gap, a routing delay or a spike in volume rather than a generic service issue. Response time is shaped by how fast requests arrive, how much capacity is on hand to absorb them, and how cleanly each request reaches the right person.

The tree below splits response time into incoming demand, available capacity and routing. Reading it, a slow tail at 17:00 might be a volume peak meeting a thin late shift, while slow mornings might be misrouted tickets sitting in the wrong queue. Each branch points to a different owner and a different lever.

Expectations vary sharply by channel. Live chat and phone are judged in seconds and minutes, email in hours, and social channels somewhere between. Set targets per channel and track the median and 95th percentile against them, because a single blended target will always flatter your fastest channel and punish your slowest. The ranges below are common first-response expectations rather than guarantees.

Improving response time is mostly about matching capacity to demand and clearing the friction that lets requests sit. Focus first on the tail, because shaving the slowest cases improves customer experience far more than nudging an already-fast median.

1. 1

   Reporting only the average

   The average masks a slow tail, so a healthy headline can sit on top of a poor experience for many.

2. 2

   One target across all channels

   A blended target flatters chat and unfairly penalises email, hiding real problems in both.

3. 3

   Measuring an auto-reply as a response

   Counting an automatic acknowledgement as the response makes times look fast while customers still wait for help.

4. 4

   Ignoring business hours

   Including overnight gaps in response time distorts the figure unless you measure within agreed service hours.