AI quality assurance for every client program
BPO quality teams usually review a small manual sample of each program and report on what they happened to hear. Merivex evaluates conversations against your own weighted scorecard, cites the transcript turn behind every score, and shows where quality is slipping by team, so a program review rests on evidence rather than a sample.
The sampling problem
A sample cannot defend a program review.
Manual sampling
With Merivex
Manual sampling
A reviewer hears a few conversations per agent each month.
With Merivex
Evaluation runs on the conversations you connect, up to your plan's monthly allowance: thousands of conversations rather than a handful.
Manual sampling
A score is one reviewer's judgment, hard to challenge after the fact.
With Merivex
Every criterion score links to the transcript turn it came from, so a disputed score is checked in seconds.
Manual sampling
Recurring problems surface when the client complains.
With Merivex
A behaviour becomes a pattern only when several evaluated conversations support it, each linked to its evidence.
Manual sampling
Coaching follows the loudest issue of the week.
With Merivex
Coaching plans are derived from those patterns and point to the conversations behind each step.
Built for outsourced operations
What a BPO quality lead gets
Your scorecard, weighted
Define the criteria and weights you are measured on. The rubric is versioned, so older scores keep the rubric they were made under.
Evidence you can put in front of a client
Each finding carries a rationale and a citation to the transcript turn, so a program review is a walk through evidence, not a debate about impressions.
Humans stay in charge
Reviewers accept, adjust or reject any AI evaluation. The decision, the reviewer and the time are recorded, and an adjusted score replaces the AI one.
Calibration before trust
Import your reviewers' scores and measure how closely Merivex agrees, overall and per criterion, before you rely on it.
Team and agent views
Quality, compliance and sentiment trends by team and by agent, across shifts and sites.
Read-only connection
Connect the database your contact center platform writes to (PostgreSQL, MySQL, Microsoft SQL Server, MongoDB, SQLite or Supabase) or import CSV, JSON or XLSX. Connectors only ever read.
Getting started
From an export to your first evaluations
01
Connect or import
Point Merivex at a read-only database, or upload an export of recent conversations.
02
Set the scorecard
Enter your criteria and weights, or start from the default QA framework and adjust it.
03
Calibrate
Compare Merivex's scores with your reviewers' on the same conversations, criterion by criterion.
04
Run it at volume
Evaluate at plan volume, let the risk-based review queue route what needs a person, and coach from the patterns.
Questions
What BPO quality teams ask first
Can each client program have its own scorecard?
A workspace scores against one active, weighted rubric at a time, and the rubric is versioned so changing it never rewrites earlier scores. If your programs are scored on different scorecards, talk to us before you start so the setup matches how you report to each client.
How is Merivex priced for a BPO?
By conversation volume, not by seat. Every paid plan includes unlimited users, so supervisors, coaches and quality analysts across programs can all work in it. Plans start at $299 a month.
Does Merivex analyze call audio?
No. Merivex works on text: chat and email conversations, and transcripts of voice calls. It does not transcribe audio and does not measure tone, silence or talk ratio. If your calls are already transcribed, those transcripts can be evaluated.
Which languages does Merivex support?
English and French. Neither has been independently validated for accuracy, which is why calibration against your own reviewers comes before relying on the scores.
Where is conversation data stored?
Production data is stored in the EU (Frankfurt). Conversation content is sent to a US-based AI inference provider to be analyzed. That provider is named in the trust centre, alongside every other subprocessor. Merivex does not train or fine-tune models on your data.
Keep reading
Try it on one program first
Start with an export of last month's conversations and your current scorecard, then compare what Merivex finds with what your sample told you.