Every business thinks it replies quickly, and most are measuring the wrong thing or nothing at all. The usual evidence is a feeling: we are pretty good at getting back to people. Then someone scrolls to a question under a post from eleven days ago, with three likes and no answer.
Response time is worth measuring because it is one of the few social media numbers that maps directly onto whether someone buys. It is also one of the easiest to measure dishonestly. This guide covers how to measure it so the number means something, what target to set, and the changes that actually make a team faster.
What counts as a response
This is where most measurement quietly goes wrong, and it goes wrong in the flattering direction.
A reply counts. Something the customer can read, that answers or acknowledges their message.
A label does not count. Filing a conversation under “pricing” is useful triage, and it is invisible to the person waiting. Any system that treats labelling as responding will report an excellent response time for a queue nobody has answered.
Assignment does not count. Giving a conversation an owner is a promise, not a reply.
Marking something done does not count, unless a reply went with it. Closing an unanswered question is the single worst thing that can happen in a queue, because it both fails the customer and hides the failure.
A like is a grey area. On a compliment, a like is a reasonable acknowledgement and arguably a complete response. On a question, it is not. If your reporting counts likes, keep them separate.
Insist on this distinction when you set up any tool. Picifer’s inbox is built around it: a rule can label, assign, set sentiment and mark an item done, and adding a label is deliberately not counted as a first response, precisely so that triage cannot make the numbers look better than the service.
The two numbers that run a queue
You need fewer numbers than you think. Two, in fact, and they do different jobs.
Median first response time tells you about a typical conversation. Use the median rather than the mean, because one message that waited a fortnight will drag an average into fiction while the median stays honest.
The age of the oldest unanswered conversation tells you whether anything is rotting. This is the uncomfortable one, and the one worth putting somewhere visible. If the oldest open item is five days old, no improvement in the median matters this week.
Together they catch both failure modes. A good median with an old backlog means you answer easy things and avoid hard ones. A bad median with no backlog means you are slow but thorough. The fixes are different.
| Measure | Why it matters | The trap |
|---|---|---|
| Median first response time | What a typical person waits | Using the average instead, which one bad case ruins |
| Age of the oldest open item | Whether anything has been abandoned | Closing things to clear it, which hides the problem |
| Open count by kind | Where the backlog actually is | Treating comments, reviews and questions as one pile |
| Percentage answered within target | Consistency, not speed | Setting a target so loose that everything passes |
| Reviews answered | Reviews are read for years | Counting replies to five-star reviews as the same work |
Working hours, not calendar hours
If you do not promise overnight cover, do not measure yourself against it. A message that arrives at 9pm on Friday and is answered at 9:10 on Monday is a ten-minute response in working hours and a sixty-hour response on a calendar. One of those numbers describes your service and the other describes your opening times.
Measure in working hours, and then do the one thing that makes the calendar number irrelevant: say when you reply. A line in your profile or pinned post, naming the hours someone can expect an answer, converts a slow weekend from a failure into an expectation that was met.
There is one exception worth knowing. On the networks that allow private messages, a business can usually only reply inside a window that starts from the customer’s last message, often 24 hours. A message that arrives on Friday evening may be unanswerable by Monday morning, whatever your working hours say. If private messages are part of your service, that window, not your opening times, sets the schedule.
Setting a target you will actually hit
Pick targets per kind of conversation, not one number for everything.
- Questions and comments: same working day.
- Complaints: within a few hours in working time, because other people are reading.
- Reviews: within one or two working days. Fast enough to show you noticed, slow enough to write it calmly. Our guide to responding to negative reviews covers the writing.
- Private messages: inside the network’s own reply window, which usually means the same day.
- Mentions that are not questions: when you get to them. Not everything needs a target.
Then set the targets at a level you can hit on a bad week, not a good one. A target you miss half the time is not a target, it is a source of guilt, and the team will stop looking at it.
The three changes that actually make a team faster
Almost every slow queue is slow for one of three reasons, and none of them is headcount.
One queue instead of five apps
Most delay is not reply time, it is discovery time. A question waited two days because nobody opened that app for two days. Gathering comments, mentions and reviews into one list converts a search problem into a work problem, and that single change usually produces the biggest improvement anyone measures.
It also makes the honest number visible for the first time. Plenty of businesses discover, on the first day of using one queue, that their real response time was several times what they believed.
A named owner for a named time
“Whoever sees it first” works until two people are busy. Replace it with a rota so specific it is boring: one person owns the morning check, one owns the late afternoon, and both know what to escalate and to whom. For a single-person business, the equivalent is a time of day, in the diary, treated like any other appointment.
Saved replies for the repeated questions
Look at a month of conversations and count the questions. In most businesses, a handful of questions make up a large share of the queue: opening hours, delivery, parking, whether you do a particular thing, how to book. Write a good answer to each once, keep them in one place, and add one specific sentence when you send them so they do not read as a template.
This is the only genuinely safe automation in the list. Everything that needs judgment should be written or at least read by a person before it is sent. One automated reply to a real complaint undoes months of good replies, which is why the rules in Picifer’s inbox can send a saved reply but can never hide or delete anything.
Out of hours, holidays and the quiet weeks
The queue does not know you are closed, and most bad response-time months are one closure nobody planned for.
Say when you reply. A line in each profile naming your reply hours converts a slow weekend from a failure into an expectation met. It costs one sentence and it is the single highest-value change on this list.
Decide what happens over a holiday. One person checking once a day for ten minutes is usually enough to stop anything rotting. Agreeing that in advance is the difference between a two-day backlog and a two-week one.
Pause the queue when something happens. If there is bad news locally or in your industry, scheduled posts keep going out regardless, and a cheerful promotion landing in the middle of it generates exactly the messages you least want to answer slowly. Name the person who may pause the schedule, and let them do it without asking.
Watch the quiet weeks. Counter-intuitively, response times often get worse when volume is low, because nobody opens the queue when they do not expect anything in it. A fixed time of day solves this and nothing else does.
What not to do about response time
Do not chase a published benchmark. Numbers about how fast customers expect a reply vary wildly by source, industry and how the question was asked. Your own trend over three months is worth more than any of them, because it is measured on your audience.
Do not optimize the median by ignoring the hard ones. The fastest way to improve a median is to answer every easy question and leave every difficult one. That is also the fastest way to lose a customer, which is why the age of the oldest open item sits next to it.
Do not count a vanity version of the metric. Response time measured only on messages you chose to answer is not response time. Our vanity metrics entry covers the general shape of this mistake.
Do not automate your way out of a staffing problem. If the queue is genuinely too big for the people on it, the honest options are fewer channels, clearer hours or more people. An auto-reply that promises a response nobody will send is worse than silence.
A four-week plan to improve it
Week one: measure honestly
Get every conversation into one queue and look at the real median and the real oldest item. Do not change anything yet. The first number is the only baseline you will ever get.
Week two: clear the backlog and name owners
Answer everything older than a few days, even the awkward ones, especially the awkward ones. Then assign who owns which part of the day, in writing.
Week three: write the saved replies
Count the repeated questions from the last month and write the top five answers. Put them where whoever is on duty can reach them in two seconds.
Week four: set the target and publish your hours
Pick a target per kind of conversation that you hit on a bad week. Put your reply hours in your profiles, so a slow weekend is an expectation rather than a failure.
After a month, compare the median and the oldest open item with week one. If both improved, keep going and change one thing at a time. If the median improved and the oldest item got worse, you have started avoiding the hard conversations, which is the one failure this measurement exists to catch.
If you want the rest of the routine, our guide to running a social media inbox covers the daily session this measurement sits inside, and comment moderation by network covers the few conversations that need an action rather than a reply.



