QA → RCA
Why the score dropped
A failed item points at a cause. Root Cause Analysis tells you whether the cause is the agent, the process or the product.
Every interaction scored against your own scorecard, in every language you run, on the day it happens. The score carries the rule it applied and the passage it read.
[01 / THE PROBLEM]
A 300-seat operation handles around 150,000 interactions a month. Reviewing five per agent covers 1,500 of them. The other 148,500 are described by a number nobody built from them.
THE REVIEW FORM TODAY
WITH REVEAL QA
Sampling was a compromise made when listening was expensive. Reviewers are not the weak part of it. The sample is.
[02 / SCORECARD]
REVEAL does not arrive with a quality model and ask you to adopt it. It reads the form your team already argues over, item by item, and turns each item into a rule it has to defend.
01 · THE FORM
The scorecard goes in as it stands, including the sections nobody has reviewed since 2023. Weights and pass marks stay yours. COPC-aligned templates are there if you want a starting point, not as a replacement.
02 · THE RULE
A score with no passage attached is not written. If the model cannot point at the words that earned or lost the point, the item comes back unscored and a human gets it, which is the honest outcome.
03 · THE AUTO-FAIL
Compliance breaches, missing verification, an unauthorised promise. Each is a named rule with its own event, so it routes to an approval gate the moment it is detected rather than surfacing in a report.
[03 / CALIBRATION]
Human review does not disappear. It moves to the calls that need judgement, and the disagreement between reviewers turns into something you can watch instead of something you sit through.
Every interaction carries a machine score with its evidence attached, available the day it happens.
Reviewers score a slice blind, against the same items, without seeing the machine result first.
The variance between them is measured per item, per reviewer, per team. Drift has a shape.
Disagreements go to the item, not the call. A rule everyone reads differently gets rewritten once.
The change is versioned. Scores before and after it stay distinguishable, so a trend never quietly changes meaning.
A calibration drift event fires when reviewer variance passes the threshold you set. REVEAL Flows can turn that event into a scheduled session and the material to run it, which is the only version of calibration anyone has ever enjoyed.
[04 / THE AGENT]
An agent signing in does not get the six-tab dashboard. They get their own card: their scores, the items that moved, the passages behind them, and the two things to work on next.
Quality scoring only changes behaviour when the person being scored can see the reasoning. Every point on the card links to the moment in the conversation that produced it, so a coaching session starts from the recording rather than from a disagreement about the number.
The Agent Performance Index rolls the whole chain up per person: quality, resolution, effort, the drivers they handle most. It is the same data the executive view runs on, at the only scale that means anything to the agent.
[05 / MODULES]
The eight L1 modules share a single log, so a score does not sit on its own. The reason a call failed and the reason it happened at all are written against the same interaction.
QA → RCA
A failed item points at a cause. Root Cause Analysis tells you whether the cause is the agent, the process or the product.
QA → TCD
Scores split by contact driver. A refund query and a technical escalation stop being judged on the same average.
QA → KB
An agent marked down for a wrong answer that no article covers is a knowledge gap, not a coaching case.
QA → KPI
Quality feeds the contractual indicators directly, so the number in the monthly review drills back to the interactions behind it.
[06 / NUMBERS]
Nothing here needs a new process from your team. The scorecard is the one you use, the interactions are the ones you already record, and the coverage is the part that moves.
100%
INTERACTIONS SCORED
Voice, chat, email and social in the same pipeline, against the same items.
1%
WHAT A FIVE-CALL SAMPLE COVERS
1,500 of 150,000 at 300 seats. The rest was described, not read.
5
RECORDS PER SCORE
Input, rule, model, signature, export. Written every time, checkable later.
All four
MODELS CARRY QA
Starter to Enterprise. QA is in every model, including the smallest.
QA ships in every model, Starter to Enterprise, and in REVEAL Local from 100 seats inside your infrastructure. The full model table is on the platform page · the price is a call with your own numbers.
[07 / CONSOLE]
Open the console.Load an interaction, watch the items resolve, open the passage behind each one. Same instance our clients run. No login, no sales call.