Call center KPIs for messaging apps: what to measure when agents type instead of talk
Short answer
In messaging apps, classic call center KPIs change meaning: average handle time loses weight, while first response time, service level, missed conversations and resolution in a single conversation move to the front. New metrics appear that calls never had: concurrent conversations and promises kept. Measure them in working hours and together, never one at a time.
Silence in a chat doesn’t ring. That’s why nobody notices it.
On a call, a missed call is obvious right away: the phone rang and nobody picked up. In chat, a client writes, waits, and then simply leaves. No signal at all. If your call center moved to messaging apps but kept its call-based KPIs, the report will stay green while clients slip away.
Below: which metrics carry over from calls, which change meaning, and which appear for the first time. With formulas, but without anyone else’s benchmarks: you’ll have to find those in your own data.
What changes compared with calls
- An agent handles several conversations at once. Being busy no longer means one client.
- The client replies with pauses. A conversation can last an hour and consist of six messages.
- A missed conversation doesn’t ring. It just sits there until someone notices.
- Everything said is in writing. Quality can be judged from the text, not from a sample of recordings.
Metrics that stay at the top
First response time
How long the client waited for the agent’s first reply.
Formula: from the client’s message to the first reply, counting working hours only. Look at the median and the share of long waits, not the average: one morning message answered after lunch will skew the average without showing where the problem is.
Service level
A classic call center metric, and in chat it’s even clearer.
Formula: conversations where the first reply came within your target, divided by all conversations for the period. The target is set by the team lead: every agent knows it, and it’s the same for the whole shift.
Missed conversations
The equivalent of abandoned calls, only quieter.
Formula: conversations where the client wrote and got no reply by the end of the working day, divided by all incoming conversations. It’s worth counting separately the conversations that broke off on a client’s question: the agent replied once and disappeared.
Resolution in a single conversation
The equivalent of first call resolution.
Formula: requests closed without the client coming back on the same topic within a few days, divided by all closed requests. Pick one window for “a few days” and don’t change it.
Metrics that change meaning
Average handle time
On calls, it’s the main productivity metric. In chat it says almost nothing: the length is set by the client, who replies whenever they can. Keep it in the evaluation, and agents will start closing conversations before the issue is resolved.
Use it for one thing only: finding request types that take too much back-and-forth. That’s a reason to improve a template or a process, not to penalize an agent.
Occupancy
On a call, an agent is either talking or available. In chat, they can handle five conversations and keep up, or three and drown. Don’t look at time online. Look at the pair: the number of concurrent conversations and how fast the replies in them are.
Metrics calls never had
Concurrent conversations
How many conversations an agent handles in parallel. Find your own limit: the number at which reply time starts to grow and missed conversations appear. That’s the workload ceiling for your line.
Promises kept
The agent wrote, “I’ll check and get back to you within an hour.” Did they? In chat, promises are written down, so they can be checked.
Formula: promises kept on time, divided by all of the agent’s promises for the period.
Handoffs between shifts
How many conversations pass to the next shift without an explicit handoff. If the morning shift starts by reading someone else’s chats from scratch, the client waits twice.
Quality scores
A sample review of conversations against clear criteria: accuracy of the answer, tone, next step. A few reviewed conversations per agent per week beat hundreds read without feedback. Put conversations with rudeness, threats, complaints or refund demands at the top of the review. And agree that the agent brings such a conversation to a senior the same day, instead of waiting for it to surface in a complaint.
How not to ruin your metrics
- Count working hours only. Otherwise night-time messages will distort everything.
- Use metrics in pairs. Speed without single-conversation resolution means fast, empty replies. Resolution without speed means a queue.
- Don’t compare different lines. Sales, support and collections run at different rhythms.
- Don’t penalize a number without the conversation. Open the chat first, then draw conclusions.
- Find your baseline in your own data. Observe for the first month, then set the target.
A short checklist
- First response time is counted in working hours
- Every agent knows the service level target
- Missed conversations are dealt with the same day, not in the monthly report
- Single-conversation resolution uses one fixed window
- Handle time is not part of an agent’s evaluation
- You know the ceiling for concurrent conversations
- Agents’ promises get checked
- Shift handoffs are explicit
A call center in chat isn’t measured by how loud it is, but by how many clients didn’t leave in silence.
You can see who’s waiting. You can see who replied.
Questions and answers
Which call center KPIs work for messaging apps?
First response time, service level (the share of conversations answered within your target), missed conversations, resolution in a single conversation, promises kept and quality scores from a sample.
Why doesn't average handle time work in chat?
An agent handles several conversations at once, and the client replies with pauses. A long conversation can be perfectly normal, and a short one can end with nothing. Handle time shows the client's pace, not the agent's work.
How do you calculate service level in chat?
The share of conversations where the first reply came within the set target, out of all conversations for the period. Time is counted in working hours only.
How many conversations can an agent handle at once?
It depends on how complex the requests are. Find your own limit: when more concurrent conversations start slowing replies and increasing missed ones, you've passed it.