Stop sampling. Score every conversation.
Pazau Insights QA-scores 100% of conversations, AI and human, against scorecards you write. It finds what's driving contacts, predicts CSAT where nobody answered the survey, and flags compliance misses the same day.
- Where is my order31%−12%
- Returns & exchanges22%+4%
- Damaged on arrival9%+41%
- Promo codes7%−3%
Your rubric, applied to every conversation, with evidence.
Write criteria in plain language, weight them, and mark some as compliance gates that fail a conversation outright. Insights scores each one with a quote from the conversation and a line from the Trace, so agents and QA leads can see exactly why.
- Plain-language criteria. “Set accurate expectations on the delivery date” is a valid rule. No keyword lists.
- Versioned rubrics. Re-score up to 90 days of conversations against a draft before you publish it, and keep trend lines intact.
- Calibration built in. Your QA team grades a sample; Insights reports agreement per criterion and flags drift.
- Coaching in the workflow. Scorecards appear in Inbox with comments, and agents acknowledge them there.
- Verified identity before account changesCompliance · gatePass
- Followed Procedure: Delivery exception v12Process · 20%18 / 20
- Offered the correct remedy for the SLA breachAccuracy · 20%20 / 20
- Set accurate expectations on new delivery dateAccuracy · 15%9 / 15
- Empathy and toneSoft skills · 15%14 / 15
- Resolved without unnecessary transferEfficiency · 15%15 / 15
- Clear, correct wrap-up and dispositionProcess · 15%13 / 15
“Your sofa will definitely arrive Thursday.”
Carrier ETA from get_shipment was a window, Thu–Sat. Procedure step 4 says never promise a single date.
Quote the carrier window and offer the delivery-tracking link instead of a date.
Know why customers are contacting you before the queue tells you.
Insights clusters every conversation into drivers automatically, across languages and channels. When a driver spikes more than 3σ above its 28-day baseline, you get an alert in Slack, Teams, or email with example conversations attached.
Auto-clustering
No taxonomy to maintain. Drivers merge across English, Spanish, and 40 other languages.
Spike alerts
Hourly detection, routed to the team that owns the fix, not just support.
Cluster to Procedure
Turn a driver into a draft Procedure in one click, then simulate it.
Why it escalated
Every handoff is summarized and grouped by root cause.
- Where is my order18,420−12%
- Delivery window changes9,310+6%
- Assembly service booking4,870+2%
- Damaged on arrival · Harlow sofa1,260+214%
- Promo code errors980−31%
Lumen & Oak spotted a seam defect on one sofa model from a driver spike three days before returns data would have shown it, and automated delivery-exception handling on the way to a 38% lower cost per contact.
Customer storiesHear from the 89% who never answer the survey. Catch the miss before a regulator does.
CSAT prediction on every conversation
A model trained on your own survey history predicts a 1–5 score for every conversation, with a confidence band. Low predicted scores open a follow-up task in Inbox while the customer can still be won back.
- Call-recording notice (two-party states)100% checkedClean
- Reg E disclosure on card disputes100% checked3 flagged
- No guarantee of claim outcome100% checked1 flagged
- Mini-Miranda on collections calls100% checkedClean
- Cancellation offered within 2 turns (FTC click-to-cancel)100% checkedClean
Compliance monitoring, not compliance sampling
Required disclosures, prohibited promises, and consent rules are checked on every conversation and call. Flags route to your compliance queue with evidence, and exports are ready for examiners. Meridian Mutual scores 100% of claims calls against state claims-handling rules.
One scorecard for AI and people. Same bar, same evidence.
Because Pazau Agent and your human team are scored on the same rubric, you can see where the AI is ready to take more volume and where people still do better. Split any metric by driver, channel, language, or Procedure version.
- Decide which drivers to hand to Pazau Agent next, with the numbers to back it.
- Compare resolution rate before and after every Procedure publish.
- Spot where handoffs happen too late, or too early, by root cause.
- Report cost per resolution for AI and human side by side to finance.
- QA score93.888.4
- Policy adherence99.2%96.1%
- Predicted CSAT4.524.61
- Median handle time1m 52s9m 40s
- Reopened within 7 days3.1%4.4%
- Cost per resolution$0.75$6.90
Built for the people who run support, not just watch it.
Calibration sessions
Run blind calibration with your QA team and see agreement per reviewer and per criterion.
Rubric diffs
Compare two rubric versions side by side, broken down by criterion, before you roll one out.
Agent and team views
Scores, trends, and coaching notes per agent, with team leads seeing only their teams.
Fairness checks
Score distributions by language and channel, so no customer group is served worse.
Warehouse export
Scores, drivers, and Traces sync to Snowflake, BigQuery, Databricks, or S3 every 15 minutes.
Custom dashboards
Build views by any field, compare custom date ranges, and schedule reports in the owner's time zone.
Works on conversations that never touched Pazau Agent.
Insights reads from Pazau Inbox and Voice natively, and from Zendesk, Salesforce Service Cloud, Genesys, and Amazon Connect through our connectors. Many teams start here, then add Pazau Agent once they can see which drivers are ready.
See integrations- Coverage
- 100% of chat, email, messaging, and voice
- Time to score
- Under 4 minutes after close (p95)
- Judge agreement
- 91% with expert reviewers
- Languages
- 45+ for text, 30+ for voice
- Retention
- Configurable, 30 days to 7 years
- Training
- Never trains foundation models on your data
Insights questions.
How accurate is automatic QA compared with our human QA team?+
Can we use our existing QA scorecard?+
Do we need Pazau Agent or Pazau Inbox to use Insights?+
How does CSAT prediction work?+
What does “compliance monitoring” cover?+
How is Insights priced?+
See what your team could resolve.
Bring 1,000 of your real conversations. We'll show you, in a 30-minute call, how many Pazau Agent resolves end-to-end.