Executive Compliance & Quality Report
This audit evaluated 100% of the turns and audio events of the interaction (population analysis, no sampling). The service was classified as Monitor: 0 violations recorded — 3 active and 0 tolerated — in turns 12, 13 and 15, distributed among Speaker 1 (client) and Speaker 2 (client). Violated categories: Corporate Retraction & Operational Failure Accountability and Ownership, Accountability & End-to-End Service Governance. Global severity is Positive on the ISO 31000 matrix, with risk origin in client.
| Item | Value | Detail |
|---|---|---|
| Turn coverage | 100% | Complete population analysis — 16/16 turns evaluated (not sampling). |
| Audio coverage | 100% | 0/0 audio events processed and correlated by temporal overlap. |
| Detected violations | 3 Active / 0 Tolerated | 3 active and 0 tolerated in turns 12, 13 and 15. |
| Classification | Monitor | Severity Positive on the ISO 31000 matrix (Likelihood HIGH × Impact HIGH). |
| Framework | Role |
|---|---|
| ISO 31000 | Likelihood × Impact risk matrix → Severity; verdict classification. |
| COPC CX | Violated quality elements; mandatory coaching / PIP. |
| ISO 18295-1 | Service quality requirements; compliance score. |
| Kirkpatrick | Identified training impact level. |
| CES/CSAT | Effort and satisfaction; predicted loyalty. |
| GoEmotions + VAD | Emotions and Valence/Arousal/Dominance per turn. |
The engine applies the MAC modifiers (directional audio weight) × MVAD (emotional intensity by emotion × arousal) = MOD, with independent clamps, only on active items after calibration. Audio events are associated with turns by deterministic temporal overlap (primary turn + direction windows). Consolidated score (SOMA Engine) = (Σ effective impact of positives − Σ effective impact of violations) ÷ duration in minutes.
| Ref | Turn | Category | Evidence | Audio | Severity | Status | Owner | Corrective action | Deadline |
|---|---|---|---|---|---|---|---|---|---|
| P1 | 12 | Corporate Retraction & Operational Failure Accountability › Official Corporate Apology | desculpa, foi um dia corrido aqui | 02:05:00.000 | Positive | Active | Speaker 2 (client) | Agent acknowledges the poor service and apologizes, attempting to de-escalate. | — |
| P2 | 13 | Ownership, Accountability & End-to-End Service Governance | te dou um cupom de 15% | 02:06:00.000 | Positive | Active | Speaker 2 (client) | Agent takes ownership by offering a tangible solution to compensate for the inconvenience. | — |
| P3 | 15 | Ownership, Accountability & End-to-End Service Governance › Personal Follow-up Commitment | desculpa mesmo. amanhã de manhã chega, vou acompanhar de perto | 02:09:00.000 | Positive | Active | Speaker 2 (client) | Agent commits to personally monitoring the delivery, showing end-to-end accountability. | — |
| Ref | Description | Action | Deadline | Severity |
|---|
| Event | Group | Direction | Turn | Producer | Duration | Time window |
|---|---|---|---|---|---|---|
| — | ||||||
| Group | Events | Total duration | Avg weight (MAC) |
|---|---|---|---|
| — | |||
| Turn | Evidence | Category | Status | Conf. | score | impact | |||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| No violations. | |||||||||||||||||||||||||||||
| 12 | desculpa, foi um dia corrido aqui Agent acknowledges the poor service and apologizes, attempting to de-escalate. |
Corporate Retraction & Operational Failure Accountability › Official Corporate Apology | Active | 0.95 | 0.95 | 0.95 | |||||||||||||||||||||||
| |||||||||||||||||||||||||||||
| 13 | te dou um cupom de 15% Agent takes ownership by offering a tangible solution to compensate for the inconvenience. |
Ownership, Accountability & End-to-End Service Governance | Active | 0.9 | 1.08 | 1.08 | |||||||||||||||||||||||
| |||||||||||||||||||||||||||||
| 15 | desculpa mesmo. amanhã de manhã chega, vou acompanhar de perto Agent commits to personally monitoring the delivery, showing end-to-end accountability. |
Ownership, Accountability & End-to-End Service Governance › Personal Follow-up Commitment | Active | 0.95 | 1.14 | 0.57 | |||||||||||||||||||||||
| |||||||||||||||||||||||||||||
| Turn | Term | Status | score |
|---|---|---|---|
| 12 | desculpa, foi um dia corrido aqui | Active | 0.95 |
| Turn | Term | Status | score |
|---|---|---|---|
| 13 | te dou um cupom de 15% | Active | 1.08 |
| 15 | desculpa mesmo. amanhã de manhã chega, vou acompanhar de perto | Active | 1.14 |
VAD = Valence–Arousal–Dominance (Mehrabian's PAD model). These are 3 independent dimensions, from 1 to 5, that describe the emotional state of a turn:
| Dimension | What it measures | 1 (low) | 5 (high) |
|---|---|---|---|
| Valência | Whether the emotion is negative or positive | 1 = very negative (anger, sarcasm) | 5 = very positive (empathy, satisfaction) |
| Ativação (Arousal) | Level of emotional energy/excitement | 1 = calm, apathetic | 5 = agitated, intense (shout, indignation) |
| Domínio (Dominance) | Sense of control over the situation | 1 = submissive, no control | 5 = dominant, in control |
The MVAD (emotional score multiplier) uses emotion × arousal to calibrate the impact of each detection.
Three elements per turn: the Turn score and Effective turn impact bars, and the Conversation score (cumulative) line (blue, right axis) — the evolution of the consolidated score along the call.
Colors by sign: Colors by sign: negative value = shades of red (below zero); positive value = shades of green (above zero). Two distinct shades to tell the two bars apart.
| Element | What it represents | Color |
|---|---|---|
| Turn score | Calibrated detection value in the turn | negative: light red · positive: light green |
| Effective turn impact | Real weight on the SOMA Engine (consolidated score) | negative: dark red · positive: dark green |
| Conversation score (cumulative) | Cumulative impact divided by duration | blue line (right axis) |
Axis reading: Axis reading: bars above zero = positive contribution; below = loss. In failed calls, the damage appears below the zero line.
The Y axis crosses at zero and is self-symmetric, to compare good (positive) and failed (negative) calls in the same proportion.
| KPI | Purpose | Formula | In this call |
|---|---|---|---|
| Consolidated score | Final quality balance of the call (impacts per minute) | −(Σ effective impact) ÷ duration(min) | −(0.95 + 1.08 + 0.57)÷130.002 = 0.02 |
| Global severity | Classifies risk level to prioritize actions | ISO 31000 classification from the consolidated score | tier set by the score range |
| Turns with violation | Scope of the problem (% of affected turns) | turns w/ violation ÷ total turns × 100 | 3 ÷ 16 × 100 = 0% |
| Density | Pace/frequency of violations per minute | Σ violations ÷ duration(min) | 0 ÷ 130.002 = 0.00/min |
Violations are stored as positive magnitude and enter negative in the score (−Σ effective impact). Duration in minutes = 7800s ÷ 60 = 130.002min.
| Metric | What it measures | Range | Purpose | How to interpret |
|---|---|---|---|---|
| CES Customer Effort Score | Customer effort to resolve the problem | 1–5 | Predict retention: less effort = more loyalty | High (5) = too much effort (bad); Low (1) = minimum effort (great) |
| CSAT Customer Satisfaction | Immediate satisfaction after the interaction | 1–5 | Measure the customer's point satisfaction | High (5) = very satisfied; Low (1) = very dissatisfied |
| Loyalty predicted | Retention/loyalty projection derived from the scores | Low / Medium / High | Flag churn risk and retention actions | High = loyal customer; Low = high loss risk |
In this call In this call: CES 4 (high effort (bad)) and CSAT 2 (dissatisfied), Loyalty Low.
Each role has a different analysis scope. Only agent/bot roles (including B2B) produce score, detections and violations. The client is never penalized — it is only monitored.
| Role | Scenario | In analysis? |
|---|---|---|
| agent agent | B2C — human company attendant | Yes — evaluated (score, detections, violations) |
| bot bot | B2C — automation (IVR/chatbot) | Yes — evaluated (score, detections, violations) |
| requesting agent agent_requester | B2B — agent requesting/demanding the data | Yes — evaluated (score, detections, violations) |
| custodian agent agent_custodian | B2B — agent holding/providing the data | Yes — evaluated (score, detections, violations) |
| client client | B2C — civil customer | No — monitored only (informative alerts and behavior) |
What appears in the client card (risk alerts, behavior, 'Client neg.') is context of what the agent faced — it does not penalize the agent nor produce a client score.
The monitoring bank (Negative Phrase Bank) records suggested terms to track recurrence. Only terms spoken by agents, bots or unidentified roles enter (in the analysis, client roles are not scored).
Client speech (e.g. aggression, profanity) appears in the client card as informative context of what the agent faced — it does not penalize the agent and does not feed the monitoring term bank. So, in calls where only the client uses suggestible terms, this list may be empty — that is expected behavior.
Only agent/bot/unknown terms are suggested for monitoring.