The controller against the algorithm
A controller gets through about fifty calls a shift. The system goes through all of them against one checklist.
A 2% sample does not show what happens on the line
Controllers listen to random calls, agents know that almost nothing is checked, and the manager sees scores that do not match the customers’ complaints.
Randomness
The sample fills up with calm calls while the conflict goes unheard.
Subjectivity
Two controllers give the same call different scores. The argument takes longer than the check.
Lateness
A mistake in the script is spotted a week later, once it has been repeated hundreds of times.
One checklist for people and for the system
The criteria are the ones your controllers already use. The system applies them the same way to every call and shows exactly where in the conversation a criterion was not met.
- Greeted the caller and introduced themselves
- Established what the enquiry was about
- Verified the customer’s identity
- Did not interrupt the customer
- Offered a solution and stated the timescale
- Said goodbye and asked whether anything was unclear
Every call scored by the next morning
The manager opens not a sample but the picture of the day: scores by agent, by criterion and by topic. Controllers listen to the contentious calls deliberately rather than at random.
A weak criterion turns into a practice run
The analytics sees that an agent regularly interrupts customers — and assigns them a scenario in the AI Trainer for exactly that. A week later the score on that criterion is compared with the old one.
1
The analytics finds the failed criterion
2
The trainer provides a scenario for that skill
3
The next calls show whether it helped
Still have questions?
What happens to the controllers?
They stop listening to random calls and take on the contentious ones: agents’ appeals, conflict conversations, calibrating the criteria. The system handles the volume, people handle the judgement.
How does the system understand “did not interrupt”?
From overlapping speech: the moments when the agent starts talking before the customer has finished. The threshold is configurable and every such moment can be listened to.
Can a score be challenged?
Yes. Agents see their own scores and can appeal against a specific criterion; the controller listens to the fragment and either confirms or changes the score.
We will check your calls against your own checklist
Send us the checklist and a day of recordings. We will return a score for every call and show where they differ from your controllers’ scores.
