How we use AI
Version of 22 October 2026
Mira scores uses AI to help contact-center quality teams review customer conversations. This page explains exactly what the AI does, what it does not do, and where people stay in control.
1. What the AI does
The AI works on conversations. On the Smart AI and Elite AI plans it does three things: — It transcribes recorded customer-service conversations, with speaker labels. — It drafts answers to the customer's own quality scorecard from the transcript, quoting the passage behind each answer and flagging as unverified any question the transcript does not settle. — It produces supporting notes for the evaluator: a summary, sentiment, compliance flags, coaching suggestions, and answers to classification questions the customer defines, such as contact reason or outcome. Call metrics such as talk ratio, hold time and dead air are measured by signal processing on our own servers, not by an AI model.
2. What the AI does not do
— It does not submit evaluations. Everything it produces is a draft until an evaluator submits it, and there is no setting that turns drafts into final evaluations automatically. — It does not make or recommend employment decisions. Hiring, pay, promotion, discipline and dismissal are outside the product. — It does not identify people by their voice or their face. The product has no biometric identification. — It does not learn from your data. Conversations are never used to train any model.
3. A person makes every evaluation
An evaluation exists only once a named evaluator at the customer has reviewed it and submitted it under their own name. The evaluator can accept, correct or overrule any answer, and the original AI answer, the human answer, who changed it and when are all recorded. Customers can run blind calibration sessions in which their evaluators and the AI score the same conversations independently. That shows, question by question, where the AI disagrees with their people, so they can adjust their guidance or stop relying on the AI for that question.
4. The people being evaluated
Agents can see their own evaluations, including the transcript evidence behind each answer and the context the analysis read. They can dispute any evaluation through a recorded workflow that a person resolves.
5. How customers may use the results
Our Terms of Service prohibit customers from basing decisions that affect an employee — pay, performance management, discipline or dismissal — on AI output that a person has not reviewed. Where local law regulates automated decision-making about workers, customers remain responsible for complying with it, and the product is built so that human review is the default rather than an option.
6. AI is optional
AI is not required to use Mira scores. The Essential plan is a complete manual quality program with no AI at all; AI features are included only in the Smart AI and Elite AI plans. See Plans.
7. Data and model training
Language processing runs on Google Gemini Enterprise, under enterprise terms that exclude customer data from model training. Conversations are not used to train Google's models, ours or anyone else's. Card numbers and other structured identifiers are masked before a transcript is stored or analysed. More in Security & privacy and our Privacy Policy.
8. Questions
For questions about how the AI works, or for our AI documentation as part of a procurement or vendor review, write to contact@mirascores.com.
Questions about any of this?