Key points
- Vendors in this market put manual QA coverage at roughly 2 to 5% of calls.
- The strongest setups are hybrid: AI scores every call and managers review the ones it flags.
- For regulated operators, where the recordings are processed matters as much as how well they are scored.
Contact centre QA has worked the same way for decades. A manager plays a recording, fills in a scorecard and moves on. Suppliers of QA software commonly put the share of calls reviewed this way at around 2 to 5%, and one estimates that an analyst can get through eight to ten calls a day.
What a QA team hears, before and after
Manual sampling: about 3 calls in every 100
AI scoring: every call
AI changes the arithmetic. Speech-to-text turns each recording into a transcript, and a language model checks that transcript against every line of the scorecard. Coverage goes from a sample to everything.
What changes for managers
The job moves from finding calls to acting on them. Level AI, one of the suppliers in this market, describes the strongest setups as hybrid: the AI scores every call and passes the ones that most need attention to a person. Managers stop spending their day listening for problems and spend it coaching.
It also changes how fair the process feels. When every call is scored against the same criteria, an agent is no longer judged on the three calls that happened to be picked that month.
Four questions to ask before you buy
- Can it use our scorecard? A generic rubric produces scores your team will not recognise. The system should mark calls the way your QA managers already do.
- Can a manager see why a score was given? Each mark should point to the part of the conversation it was based on.
- Where are the recordings processed? Most products need calls uploaded to the supplier's cloud. For telecoms operators, banks and public bodies, that is often not permitted.
- Who has the final say? AI should do the first pass. A person should own the result.
How we approach it at NexAura
We built UMAR around those four questions. It scores calls against the client's own scorecard, shows the evidence for each mark and leaves QA managers in charge of the outcome. It can also run entirely inside the client's own network, including networks with no internet connection, so recordings never leave the building.
For Special Communications Organization, that meant roughly ten times as many calls reviewed for a similar operating cost. The case study sets out how it was done.