Evaluation · Contact centre
Call quality evaluation for contact centres
Quality teams usually listen to one percent of calls and draw conclusions about all of them. Zero-Shot builds systems that transcribe conversations and score them against your quality form: coverage extends to every call, and each score points back to the sentence that produced it.
What we build
- 01Transcription with speaker separation, tuned for your call conditions
- 02Criterion-level scoring against your existing quality form
- 03A reason for every score, linked to the sentence in the transcript
- 04Mandatory-phrase checks for statements required by regulation
- 05Agent and team trend reporting, with agreement against human reviewers
How it runs
01
Encode the quality form
Your form is taken criterion by criterion. Unmeasurable criteria like 'was polite' are either defined properly or removed.
02
Calibrate against reviewers
The same calls are scored by both the system and a human reviewer. Low agreement means the criterion definition is the problem, not the model.
03
Widen gradually
Criteria with high agreement move to full coverage first; the rest stay under human review.
04
Connect to training
Criteria that score low become training topics. Measurement that is not connected to training only produces reports.
What we measure
- Score agreement between system and human reviewer, per criterion
- Share of calls reviewed — moving from sample to coverage
- Miss rate on mandatory phrases
- Word error rate of transcription under your real line conditions
- Movement in low-scoring criteria after training
When this is the wrong fit
The system is not built to make employment decisions on its own; it shows a score with its reasoning and a human decides. If the quality form is not written down, or reviewers disagree substantially with each other, that work comes first.
Common questions
Is speech recognition good enough for our calls?
It performs well on stable lines at normal noise levels; heavy accents, poor lines and overlapping speech raise the error rate. That is why we measure word error rate on your own recordings and report it, instead of quoting a general figure.
Won't agents see this as surveillance?
They will, and it needs to be discussed openly. Transparency helps: every score and the sentence behind it are shown to the agent too, with a route to challenge it. Hidden scoring systems damage trust and stop working over time.
Do recordings leave our infrastructure?
They do not have to. Transcription and scoring can run on your own servers, in which case audio never leaves the network. For cloud deployments, where each piece of data goes is written down.
Other services
What we've written on this
- Where to start with enterprise AIHow to choose a first AI project: which tasks qualify, in what order to proceed, and what to expect in the first eighty days.
- Writing prompts for productionWhy a prompt that works in a demo breaks in production. Treating prompts like code: structure, versioning, boundaries and measuring every change.
- Why AI makes things up, and how to reduce itWhy language models state false things confidently. The real cause of hallucination, which countermeasures work, and which do not.
Last updated: 2026-09-05
Let's talk about this one
Tell us what you want to build and within two days we'll tell you who takes it and how we'd measure it.