Quality & conversation intelligence
Score every conversation, then review the ones that matter.
Calls, e-mails, chats and AI-agent conversations go through the same scorecards. Each criterion cites the turns that justify it, and reviewers start with the conversations that need a person.
Talk balance
Score
8.4 / 10Scorecard
- MetIdentity checked · turns 3–4
- MetCard fees explained · turn 9
- MissedNext steps confirmed
Products
- Debit card
- Card fees
Sentiment
NeutralHow it works
From recording to review
The same pipeline for every channel.
Collect
Upload recordings, e-mails or chat exports, send them through the API, or import AI-agent conversations automatically.
Transcribe
Calls are separated by speaker and transcribed in Derja, French or English.
Protect
Card numbers, IBANs, ID numbers, phones and e-mails are masked before storage and analysis.
Score
Each criterion is checked with the turns that prove it. A failed critical item fails the interaction.
Review
Low scores, low confidence, escalation risk and unresolved cases go to review, plus a random share.
Built in
What comes with it.
One QMS for every channel
Recorded calls, e-mail threads, chats (web, WhatsApp, Messenger, SMS) and AI-agent conversations in one place.
Scorecards with evidence
Scorecards per channel with weights and critical items. Each criterion shows the turns that prove it.
SLA and response metrics
For e-mail and chat: first response, response times, longest wait and handle time against your targets.
Outcome, sentiment and risk
Outcome, sentiment, customer effort, first-contact resolution and escalation risk for each interaction.
Natural-language search
Ask “customers who asked about a late tow truck”. Matches are verified and shown with a quote.
Saved searches and alerts
Re-run a search hourly, daily or weekly and see new matches in the notification bell and by e-mail.
Products & services catalog
Each call is tagged with the products discussed and how clearly they were explained.
Ingestion API and Derja transcription
Send interactions from other systems through the API. Recordings are transcribed by speaker in Derja, French or English.
Example scenario
Finding a product that is poorly explained
Illustrative example, not a customer case. A quality manager in a bank wants to know how card fees are explained on the phone.
- 01Upload last week’s recorded calls
- 02Calls are tagged with the products discussed
- 03Search “calls where the card fees were not explained clearly”
- 04Save the search with a daily alert
- 05Open the matching calls and plan coaching
Control
Scores you can check
The score is computed from your scorecard weights. The AI brings evidence and a confidence for each criterion, not an unexplained grade.
- Evidence turns and confidence for every criterion
- Personal data masked before analysis
- Scorecards per channel, with critical items
- Human review workflow
- Data scoped to your workspace
Quality questions
Quality management and conversation intelligence, answered.
What is a quality management system for customer service?
It reviews customer interactions against a scorecard to see what goes well and what needs coaching. Liberrex applies scorecards to calls, e-mails, chats and AI-agent conversations, with evidence for each criterion and a review workflow.
Which interactions can Liberrex analyse?
Recorded calls (WAV, MP3, M4A, OGG, FLAC and more), e-mail threads (.eml, .mbox), WhatsApp exports, chat transcripts in CSV or JSON, interactions sent through the ingestion API, and the conversations of your Liberrex AI agents.
Does transcription work for Tunisian Arabic calls?
Yes. Calls are separated by speaker and transcribed with Liberrex Tunisian Arabic (Derja) speech recognition, as well as French and English. Search turns a question into Derja, French and English phrasings, so a French question can find Derja calls.
How is a quality score calculated?
The AI checks each scorecard criterion and cites the turns that support it, with a confidence. The score is then computed in code from your weights, and a failed critical criterion fails the interaction.
Can I search conversations in plain language?
Yes. Ask a question such as “customers who asked about a late tow truck”. The search ranks the matching passages, verifies each candidate and shows a quote. Saved searches can alert you hourly, daily or weekly.
How is personal data handled in quality analysis?
Card numbers, IBANs, national ID numbers, phone numbers and e-mail addresses are masked before transcripts are stored and analysed, and data stays scoped to your workspace.