New Evidence-cited QA for contact centers and BPOs
Merivex is an AI-powered quality intelligence platform for contact centers, BPOs and customer operations teams. Most quality programmes review a sample and hope it represents the rest. Merivex evaluates every connected interaction, finds the patterns behind the scores, and turns them into coaching your team can act on, with an evidence trail the broad platforms skip.
Definition
Merivex is an AI-powered quality intelligence platform for contact centers, BPOs and customer operations teams. It evaluates every connected call, chat and email against your own criteria, cites the turn behind every score, and turns recurring patterns into targeted coaching.
The problem
Sampling was a reasonable answer when reviewing a conversation meant a person sitting through it. It is still how most quality programmes run, and it means the number on the board describes the sample, not the operation.
| Sampling | With Merivex |
|---|---|
| A reviewer listens to a handful of conversations a week. | Every connected conversation is classified and searchable. |
| Findings arrive weeks after the conversation happened. | Evaluations cite the turn they came from, so the score is arguable on facts. |
| Scores are defended by seniority, because the working is not written down. | A verifier challenges any claim the evidence does not support before it is published. |
| The problems that repeat quietly are the ones nobody sampled. | A behaviour is only recorded as a pattern once several interactions agree. |
Why a focused layer
Broad customer-experience platforms treat quality as one module. Merivex treats it as the entire job, which is what it takes to produce a score an operations lead will actually defend in a room.
| A broad CX platform | A focused quality layer |
|---|---|
| Quality is one module among many, and usually the one that shipped last. | Quality evaluation is the whole product. The evidence trail is the point, not a feature. |
| You take the model's word for the score. | A separate verifier challenges the score, and you calibrate it against your own reviewers. |
| Patterns are surfaced quickly, because responsiveness demos well. | A pattern has to clear a threshold. One striking interaction is never enough. |
| Seats are metered, so the findings stay with a few licence holders. | Unlimited users on every plan. Priced on conversations analysed, not on who may look. |
The analysis engine
Four things have to be true before quality data changes anyone's behaviour: it has to be complete, evidenced, generalizable, and it has to reach the right person.
Every connected interaction is classified and scored, so the number on the board describes the operation and not a sample of it.
Each score cites the turn it came from, and a verifier flags claims the evidence does not support before anyone sees them.
A behaviour becomes a pattern only once several interactions corroborate it, so you act on trends rather than the last thing read.
Findings land on a named person as a specific thing to change this cycle, with the interactions that show it attached.
of connected conversations analysed, not a weekly sample
conclusions published without the evidence behind them
access to your systems: Merivex reads, and never writes back
users on every plan, so nobody is locked out of the findings
Evaluate
Every evaluation cites the interaction and the turn it came from, and a separate verifier agent challenges claims the evidence does not support.
Evaluation · int-4821
Scored against your rubric · 4 criteria
Coaching recommendation
workforce agent
Closing turns drop empathy scoring while resolution stays high. Three interactions this week end without acknowledging the customer's frustration.
Focus for this cycle
evidence · 3 interactions cited
Discover
Recurring behaviours are only recorded as patterns when multiple interactions back them up, so you act on trends rather than on the last thing someone happened to read.
Discovered patterns
last 7 days
Recorded only when multiple interactions support them.
Data sources
None of the ways in ask you to move your data first. Merivex pulls from Zendesk, Salesforce Service Cloud, HubSpot, Intercom, Genesys Cloud, Talkdesk, Five9, Aircall and Twilio, reads the database your conversations already sit in, imports CSV, JSON, Excel or SQLite files, accepts a signed webhook (n8n included), pulls from your own API on Growth and Enterprise, and transcribes call recordings you upload: up to 50 at once, then evaluated like any chat or email.
Every connector reads only. Your helpdesk, telephony and CRM stay exactly where they are; nothing has to be migrated for you to start, and nothing is decommissioned when you do. The full list, with what each one imports, is on the integrations page, and call recordings have their own page.
See what it costsData sources
Ingestion health · read-only
See it work
Seeded sample data, no sign-up. Score the interaction, then follow any score to the exact turn it was read from.
int-4821
chat · Duplicate charge
Customer
I was charged twice for the same order this month and I need it sorted today.
Agent
I can see both charges on the account. I'll refund the duplicate now, and it lands back on your card in three business days.
Customer
Three days is a long time when it's your money that's missing.
Agent
That's just how the refund timing works, there's nothing I can do about it. Is there anything else?
Refund of the duplicate charge confirmed. Ticket closed by agent.
Rubric
Evaluation result
sample outputRun the evaluation to score this interaction against the rubric, with every score linked to the turn it came from.
Resolution scored high: the duplicate was refunded in turn 2. Empathy scored low: turn 4 dismisses the customer's concern instead of acknowledging it. Cited turns 2 and 4. Verifier: evidence supports both findings.
Changelog
What landed in the product most recently. Every item is live on every plan unless the card says otherwise.
Upload MP3, WAV, M4A, OGG, FLAC or WebM recordings, up to 50 files at once, 50 MB and 2 hours per call. Each call is transcribed, then evaluated like any chat or email.
Read about call recording transcriptionFills in customer, agent, topic, summary and contact reason for every transcribed call.
Scoped tools and scoped memory for each job, from evaluation to the verifier that challenges it.
Explore the AI agentsYour business, protocol, non-negotiables and tone of voice, read by every AI agent, the evaluation and the coaching.
Helpdesk, contact-center, database and webhook connectors in one place, each one read-only.
See every integrationPick a range once and every figure, chart and list follows it, back as far as your plan's history.
How it works
Connect the interactions you already record. What comes back is traceable all the way to its source.
Input
Outcome
The workflow
Connect your helpdesk or contact-center platform, a database, a file export or a webhook, or upload call recordings (up to 50 at once) to be transcribed. Nothing has to move to a new platform first.
Interactions are validated, normalized and classified against your intent taxonomy, with data quality problems surfaced rather than silently absorbed.
Every evaluation cites the interaction it came from. A separate verifier challenges conclusions, so unsupported claims are flagged instead of published.
Recurring behaviours are recorded only when multiple interactions support them, turning one-off observations into a defensible quality signal.
Findings become targeted coaching recommendations for individual workforce agents and teams, tied back to the evidence that produced them.
Inside the product
Intelligence is about your people. Platform is about the AI doing the analysis. They stay separate throughout the product.
Interactions, evaluations, the human review queue and coaching in one place. Reviewers see why an evaluation reached its conclusion, not just the score.
Your workforce agents and teams: performance over time, individual histories and the coaching plans attached to each person.
The Merivex AI agents, discovered patterns, organizational memory and the intent taxonomy that classifies every conversation.
Connected sources, ingestion health, executive metrics and drill-downs that trace any number back to its source interactions.
Pricing
Bring your whole team of supervisors, coaches and analysts without paying more per person. You pay for the conversations you analyse.
Replace sampling with evidence.
$299/mo
billed monthly
Run quality as a programme.
$899/mo
billed monthly
Governed intelligence at scale.
Custom
Annual contract
All prices in USD. Start your 14-day free trial: a valid payment method is required, and you won’t be charged until your trial ends.
Included on every plan
These are not upsells. A quality programme does not work without them, so they are in every plan.
Every conclusion cites the interaction and the turn it came from.
Behaviours are recorded only when multiple interactions support them.
Findings become targeted plans for specific people, tied to the evidence.
Unlimited users. No per-seat charge for the people who act on findings.
Every plan also includes call recording transcription and every helpdesk, contact-center and webhook connector. The Custom API connector starts on Growth.
Questions
Direct answers about Merivex, AI quality assurance, pricing and data. Still unsure? Talk to us.
Merivex is an AI-powered quality intelligence platform for contact centers, BPOs and customer operations teams. It evaluates every connected call, chat and email against your own quality criteria, cites the turn behind every score, finds recurring patterns, and turns them into targeted coaching. It analyses conversations after they happen; it is not a contact center platform, a telephony system or a real-time agent-assist tool.
Individual evaluations are useful from the first interaction you connect. Patterns need volume, because a behaviour is only recorded as a pattern once several interactions support it. In practice a few weeks of a real queue is enough to see them form.
No. It removes the part of their job that was never worth their time: listening to conversations at random to build a sample. Reviewers still confirm, overrule and calibrate, and Merivex tracks how often it agrees with them so you can see whether the scoring can be trusted.
Two things. Every evaluation has to cite the interaction and the turn it came from, and a separate verifier agent reviews conclusions specifically to challenge claims the evidence does not support. Anything unsupported is flagged rather than published as fact.
No. Merivex pulls from Zendesk, Salesforce Service Cloud, HubSpot, Intercom, Genesys Cloud, Talkdesk, Five9, Aircall and Twilio, reads a database or a file export, accepts a signed webhook, and transcribes call recordings you upload. Your helpdesk, telephony and CRM stay exactly where they are, and nothing is written back to them.
You define the criteria and the intent taxonomy, and evaluations are scored against them. You also write a workspace brief: your business, the people you serve, your working protocol, what great looks like, non-negotiables, escalation rules and tone of voice. Every AI agent, the evaluation and the coaching read it, so findings reflect your standards rather than generic ones.
That depends on what you do with it, and it is a fair worry. Merivex is built for the coaching direction: findings arrive attached to the evidence and to a specific, actionable recommendation, so a person can see why something was flagged and what to do differently. Scores without evidence are what feel arbitrary.
Pricing questions
From $299 a month, with unlimited users.
Priced on the conversations you analyse, not on how many people are allowed to look at the findings. Start with a 14-day free trial. Back to pricing
Because quality work only changes anything when supervisors, coaches and analysts are all in the product acting on what it finds. Charging per seat gives you a reason to keep them out. Every plan includes unlimited users; you pay for the conversations you analyse.
One customer conversation ingested into Merivex: a chat, an email thread, a call transcript or a message exchange, from any connected source. You are only counted once per conversation, no matter how many turns it contains.
A full quality assessment of one interaction against your criteria, with the evidence cited back to the conversation it came from. Your plan includes an allowance, and you choose which conversations get evaluated. Everything else is still ingested, classified and searchable.
We tell you first. You get a notification at 80% and again at 100%, then a grace band above your allowance so nothing stops mid-month. If you consistently need more, you can add capacity or move up a plan. Your analytics never stop without warning.
Every plan includes a rolling window of history, and the reporting period can reach back that far: 12 months on Starter, 24 months on Growth, unlimited on Enterprise. Recent conversations stay instantly queryable. Older ones move to durable archive storage and still count toward your plan's storage allowance; they remain retrievable, just not in the hot path. Enterprise can set its own retention and data residency terms.
Every plan includes call recording transcription, the Zendesk, Salesforce Service Cloud, HubSpot, Intercom, Genesys Cloud, Talkdesk, Five9, Aircall and Twilio connectors, n8n and the signed Webhook, plus compliance scoring in every evaluation, with breaches sent to human review. Growth and Enterprise add the Custom API connector and compliance and risk monitoring: the org-wide compliance trend, below-threshold alerts and the compliance drill-down.
Yes, in either direction. Moving up applies immediately and is prorated. Moving down takes effect at the end of your billing cycle, and your existing data stays intact.
Yes. A 14-day free trial of Starter or Growth, with every feature of the plan you pick. A valid payment method is required to start it, and you are not charged until the trial ends. Cancel any time before then and nothing is billed. Fourteen days is long enough to connect a real data source and let patterns accumulate, which is when you can actually judge the product.
Contact
The most useful first message describes how quality is measured in your operation today, roughly how many interactions you handle, and where the data lives. That is enough for us to say something concrete rather than generic.
Questions about what Merivex does, whether it fits your operation, or how it would connect to your data.
contact@merivex.aiTell us how your quality programme runs today and what you would want to see proven. We will set up an evaluation against your own interactions.
sales@merivex.aiHelp with a connected data source, an evaluation that looks wrong, or anything not working the way the docs say it should.
support@merivex.aiInvoices, plan changes, usage questions and anything about what you are being charged.
billing@merivex.aiFound a security issue? Tell us before you tell anyone else, at security@merivex.ai. Include the steps to reproduce it and what you were able to access. We will acknowledge your report and keep you updated while we fix it.
The fastest way to judge Merivex is to point it at real interactions and look at what comes back. Or run the seeded demo first: no sign-up, same pipeline.
