Last updated: September 12, 2026
What call center coaching software actually does
Call center coaching software turns a scored call into a record you can point at later. It holds the clip, the criterion that failed, the change you agreed with the agent, and whether the next score on that criterion moved. Most of the value shows up when the agent can hear the scored moment and argue with it.
QA programs mostly die in the gap between the score and the conversation. A number lands in somebody's inbox on Tuesday, they read 82 and a criterion name they half remember, and nothing about Wednesday's calls is different. Run that for a quarter and you have built a reporting habit with a morale cost attached.
Software carries the smaller half of the fix. It finds the call, marks the line, keeps the record, and tells you when the follow up is due. Nobody has built the part where a person defensive about a call they barely remember actually hears it, so that half stays yours.
Picking the call worth coaching
The instinct is to grab the worst score of the week. A call that fell apart in six places is the least useful teaching material you have, because the agent already knows it went badly and there is no single thing to fix.
A critical fail is the exception and gets handled the day it surfaces, since a missed dispatch on an account that defines that call as an emergency cannot wait until Thursday.
Otherwise the call you want is the one that scored just under standard where the agent thinks it went fine. That distance between their read and the rubric is the actual coaching. Worth qualifying that, though, because it only works when the scores are trustworthy. If two of your supervisors would grade that same call four points apart, fix the variance before you book the session.
Finding those in a month of scored audio is a filtering job, which saved searches by criterion and by agent handle.
One session, start to finish
Six steps, and the order does most of the work.
- Pull one call and one criterion beforehand, and read the transcript around the flagged line so you are not discovering the call in front of the agent.
- Ask the agent how the call went before you say anything about the score.
- Play the moment, the clip itself rather than your summary of it.
- Name the criterion and read what it requires, in the words on the scorecard.
- Agree one change, specific enough that you could score it on the next call.
- Set the follow up date and say which call you will look at against it.
Prep is the part software genuinely shortens. A transcript with the scored line marked, and the audio sitting next to it, turns a fifteen minute scrub into about a minute, which decides whether a supervisor runs six sessions this week or runs two and reschedules the rest.
Let the agent score the call first
Leading with the number puts the agent in defense before they have heard anything. Ask them to score that one criterion themselves, out loud, before you give yours.
Scoring themselves lower than you did means the coaching is mostly finished. Matching your number lets you skip straight to the change. The useful case is an agent who scores the call higher, because you are about to find out whether they misread it or the criterion says something other than what you meant.
It takes longer the first two times. After that, agents show up having already listened to the call, which is the whole behavior you were trying to buy.
One agreed change per session
A session that names six problems produces no change at all. The agent leaves with a general sense of doing badly and nothing they could follow on the next call.
Keep it to one criterion with one agreed action and a date. An agent with four weak criteria gets four sessions across a month, starting with whichever one costs the account the most. Empathy criteria are where many sessions land and the hardest to run without drifting into a conversation about personality, so empathy phrases worth coaching keep it on observable behavior.
Skill gap or process gap
Before you coach anything, find out whether the agent is the variable.
Sort the criterion across the whole team for a month rather than reading one agent's sheet. Say you run 12 agents and the criterion is the callback number read back digit by digit. Ten agents pass it 94% of the time and two pass it 40%, so you have a skill gap and two sessions to run. If all 12 sit near 51%, coaching was never the answer.
A process gap treated as a skill gap is the most expensive mistake in a QA program. You spend the hours, the agents feel managed for something outside their control, and the number holds steady because the account script asks for the readback after the close instead of before it. Usual culprits are that script, the on-call schedule the agent works from, the field order on their screen, and the wording of the criterion. Fix one and every call after it improves.
When the rubric is the problem
Agents dispute scores. A jump in disputes right after a new scorecard goes live means the scorecard is being read, and the quiet programs are the ones worth worrying about.
In the session, open the call at the timestamp behind the disputed criterion and read the words that triggered it. Check the confidence value on that line if the score came from the machine.
Then decide which was wrong, the call or the criterion. A rewritten criterion fixes every call scored after it instead of one agent's Tuesday, so that outcome is the one to want. The rewrite happens on the scorecard the criterion came from, and everyone scoring against it needs to know it changed.
There is a limit here that gets glossed over a lot. Aircall and StackAI both land on automated scoring being reliable on rule based criteria and softer once a criterion asks for judgment, which matches what we see. A machine score on a tone criterion is a first pass, so check it yourself before carrying it into a room with an agent.
Two supervisors disagreeing about the same call is a different problem with its own procedure, and calibration when reviewers disagree covers the session format and how to track the variance. A criterion that keeps producing disputes belongs in the next calibration hour.
Documentation that survives a performance dispute
Six months later, in an HR conversation or a client escalation, "we coached her about this repeatedly" is worth nothing on its own. A dated record pointing at a specific call and a specific agreed action is what holds up.
Seven fields do it, and the CSV next to this article carries the same columns.
| date | agent | call_reference | criterion | agreed_action | follow_up_date | next_score |
|---|---|---|---|---|---|---|
| 2026-09-08 | Agent A | 09-05 14:22 inbound, 3:41 | Callback number read back digit by digit | Read the number back before confirming the message | 2026-09-22 | pending |
| 2026-09-08 | Agent B | 09-04 09:10 inbound, 1:58 | Account triage question asked before taking a message | Ask the triage question on the HVAC account before dispatch | 2026-09-22 | pending |
Call reference is the field people skip and the one carrying the weight. A record naming a criterion and a date is an assertion, while a record that opens the call is evidence the agent can listen to as well.
Measure the follow up score on that same criterion rather than the agent's total. SQM Group puts the industry average QA score at 85% with 90 to 99% treated as good and only 5% of agents ever reaching 100, which is a decent argument against coaching anyone toward a number. Nothing credible exists on how much coaching moves a score, and we are not going to invent it. The share of scored calls that produced a documented conversation is a metric in its own right, and it sits with coaching completion and follow up scores.
The cadence a team can actually hold
A weekly one on one with every agent sounds like the right answer and survives contact with about a dozen people.
Run the arithmetic. Twenty minutes of prep to pull the call, read around the flagged line, and decide what you are asking for, then roughly 25 minutes in the room. That is 45 minutes a session, so six agents costs about four and a half hours a week, which a supervisor with a floor to run can hold.
Thirty agents is 22 and a half hours. Over half the week, before the escalations, the schedule, and the client call. That cadence does not degrade gracefully at this size. It stops happening in week three, and the agents notice that it stopped.
So size it to the team you have. SQM Group reports that around 60% of call centers evaluate five or more calls per agent per month, and one real conversation about one of those five beats five scores nobody discusses.
What survives at 30 agents is a monthly session with everyone on the calendar, plus a weekly slot only for agents carrying an open agreed action or a critical fail from the last seven days. Drop the weekly slot for agents already at standard on every criterion first. Prep depth goes second, since ten minutes on one clip beats a 45 minute session that gets cancelled.
The only line in that arithmetic a tool can shrink is the prep, which is what playback beside the transcript is for.
Give up the follow up date last. A program that stops checking whether the change happened is back to emailing people numbers, which is where this started.
If you want the clip and the criterion in one place rather than spread across a spreadsheet and an inbox, a call review scores a month of your own recordings against your rubric and hands your supervisors the scored lines worth coaching, at a fixed pilot price.
Frequently asked questions
How do you coach a call center agent after a bad score?
Pull one call and one criterion, then ask the agent how the call went before you say the score. Play the moment, read what the criterion requires, and agree on one change you could score on their next call. Set the follow up date before they leave the room.
What should a coaching session cover?
One criterion, the clip it came from, an agreed action, and a date when you will score another call against that criterion. Sessions covering four criteria produce no change, because the agent leaves without an instruction they could follow.
How often should call center coaching happen?
With six agents, a weekly session at roughly 45 minutes each runs about four and a half hours and holds up. At 30 agents it costs over half a supervisor's week and collapses inside a month, so move everyone to monthly and keep the weekly slot for open actions and recent critical fails.
What do you do when an agent disagrees with a QA score?
Open the call at the timestamp behind that criterion, read the words that triggered the score, and check the confidence value if it was scored automatically. Then decide whether the call was wrong or the criterion was, and rewrite the criterion when it is at fault.
Does coaching software improve QA scores?
Any figure we published for that would be one we could not stand behind, so there is none here. Measure the follow up score on the same criterion for the same agent, plus the share of scored calls that produced a documented conversation.
Sources
- SQM Group: 85% industry average QA score, 90 to 99% treated as good, only 5% of agents reaching 100. Third party, unverified.
- SQM Group: around 60% of call centers evaluate five or more calls per agent per month. Third party, unverified.
- Aircall and StackAI on rule based scoring holding up better than tone scoring. Third party, unverified.
- Trinsic feature sheet and call notes, 2026-08-27, 2026-09-03, and 2026-09-10. Verified internally.
- Session length, prep time, and cadence arithmetic. Editorial judgment, worked as a hypothetical.