Quality Control: How Every Message is Reviewed
The safety net that ensures every SI-generated message meets your standards before a contact sees it
What It Is
Every email and SMS generated by SI goes through a quality control (QC) gate before delivery. A separate AI review process scores each message on a 100-point scale against six criteria. Messages that meet the minimum score are approved and sent. Messages that fall below are quarantined for human review or rejected outright.
QC is what makes SI autonomy safe. It means you can give SI permission to communicate on your behalf — knowing that every message will be checked for accuracy, brand consistency, mandate compliance, and content freshness before any contact sees it.
You configure QC in Settings > Segmentation Intelligence > Quality Control.
Why It Matters
Autonomy without quality control is reckless. SI generates hundreds of unique messages — each personalized, each different. Without a review process, a single mistake (wrong name, repeated asset, off-brand tone, mandate violation) reaches the contact before anyone notices.
QC catches these issues automatically. When SI accidentally references an asset that was already shared with a contact (an anti-repetition violation), QC detects it and blocks the email — sending you a HIGH PRIORITY notification with the exact issue. When a message does not match your brand voice or violates a mandate, QC flags it.
The QC system also provides organizational metrics: approval rate, average QC score, and score distribution. These tell you whether SI's output is consistently high quality or drifting. A dropping approval rate is an early warning that your brand configuration, mandates, or RK library may need updating.
How It Works
QC Settings
The Quality Control tab has four controls:
Global Autonomy Toggle — "Should Zyntro Automatically Communicate if Min. Quality Met?" When on, messages that pass the minimum score are sent without human review. When off, every message is quarantined regardless of score.
Per-Phase Autonomy — Control whether autonomy applies differently per pipeline phase. Toggle autonomy on or off for each phase: FIND, WIN, KEEP, GROW. This lets you run autonomous in FIND (high-volume, lower-stakes) while requiring human approval in WIN (high-value, higher-stakes). For example: Find = Auto, Win = Manual, Keep = Auto, Grow = Auto.
Minimum Pass Score — A slider from 0-100 (default: 80). Messages scoring above this threshold are approved. Messages below are handled according to the Below Minimum setting.
Below Minimum Action — What happens when a message fails QC: Quarantine (held for human review — you receive a notification and can approve, edit, or discard) or Reject (blocked entirely with a high-priority alert).
Approver — The team member who receives quarantine notifications and reviews flagged messages.
The Six QC Criteria
Each message is evaluated against:
1. Personalization accuracy — Is the contact's name correct? Are referenced details (company, role, previous interactions) accurate?
2. Brand voice consistency — Does the message match your brand's tone, formality, and personality?
3. Content quality — Is it clear, valuable, and appropriately sized? Does it deliver standalone value?
4. Mandate compliance — Does it respect all active mandates (prohibitions, hard facts, comms flow directives)?
5. Technical quality — Are all merge tags resolved? Are URLs valid? Is formatting correct?
6. Repetition check — Has this content, asset, or approach been used with this contact before? Cross-referenced against the Outcome Ledger.
QC Outcomes
- Approve → Message proceeds to the send queue silently
- Quarantine → Flagged for human review. A notification is sent to the designated approver with the message content, QC score, and the specific issues identified
- Reject → Blocked from sending. A HIGH PRIORITY alert is created with the exact violation
Phone calls bypass QC because the call flow itself defines the AI agent's behavior — the quality is built into the call flow configuration rather than reviewed per-message.
Examples
QC rejects an email for an anti-repetition violation
SI composed an email sharing a video that was previously sent to this contact 3 weeks ago. QC detected the asset in the Outcome Ledger's previously-used list, scored the email 42/100, and rejected it. The notification reads: "QC Score: 42/100 (threshold: 80). Issue: The primary asset in this email appears directly in the SI Outcome History under assets_previously_used. This is a direct anti-repetition violation." SI will compose a different email on its next pass.
QC quarantines an email for a WIN-phase contact (manual autonomy)
SI composed a follow-up email for a contact in the WIN phase — where you have set QC Autonomy to Manual. Even though the email scores 94/100, it is quarantined for your review because WIN-phase messages always require approval. You receive a notification, review the email, and click Approve to send it — or edit it first.
Approval rate drops from 92% to 78% over two weeks
The Command Center's Quality Insights shows the decline. This suggests SI's output quality is dropping — likely because brand facts are outdated (a new feature launched but facts were not updated), a mandate conflict is causing inconsistent output, or the RK library has been exhausted for active contacts. Review your brand facts, check for conflicting mandates, and add fresh RK items.
