Get Knowledge Assistant
Do you want access to Knowledge Assistant? Contact your Zingtree Account Manager or our Support Team.
This guide explains every metric across the Knowledge Assistant Analytics dashboards—what it measures, why it matters, and what good performance looks like.
How to Read This Guide
This guide is designed for business stakeholders. Each dashboard section explains:
- What the dashboard is used for
- The key question it answers
- A breakdown of each metric (KPI), including targets and filters
Accessing the Knowledge Assistant Dashboards
If you have configured Knowledge Assistant and they have been used, you can access the dashboards from Knowledge Assistant > Overview:
Data Freshness
| Freshness Label | Meaning |
|---|---|
| Real-time | Reflects events within seconds of occurring |
| Updated hourly | Refreshed approximately once per hour |
Filters
| Filter | Description | Default |
|---|---|---|
| Date range | Restrict results to a calendar date range | Last 7 days |
| Intent | Focus on a specific user intent/topic | All intents |
| Assistant | Focus on a specific Knowledge Assistant | All assistants |
| AI Model | Filter by AI provider/model | All models |
Note: Date filters operate at the day level (not time-of-day). Any time selection is ignored.
Downloading the Filtered Data
Knowledge Assistant Analytics can be downloaded to a PDF file from the top-right of the Dashboards:
Dashboard 1: Executive Summary
Purpose: High-level operating health
Key Question: How is the AI platform performing overall?
| Card | KPI(s) | What It Measures | Target / Benchmark | Data Freshness | Filters |
|---|---|---|---|---|---|
| Total Conversations | Total CAI Sessions | Total AI-assisted customer conversations handled. Indicates adoption and volume trends. | >5% MoM growth | Updated hourly | Date, Org |
| Total KA Sessions | Total Knowledge Assistant Sessions | Number of self-service answers delivered. Proxy for deflection (no human agent needed). | ~30% of conversations | Updated hourly | Date, Org |
| Authenticated Users | Authenticated Users | Number of identified (logged-in) users interacting with the AI. | Growing over time | Updated hourly | Date, Org |
| Anonymous Sessions | Anonymous Sessions | Sessions with no identifiable user. Used alongside authenticated users to estimate reach. | Declining proportion over time | Updated hourly | Date, Org |
| Daily Sessions Trend | CAI, KA, Contained sessions | Day-by-day trends across volume, Knowledge Assistant usage, and containment. | Upward trend; ≥70% containment | Updated hourly | Date, Org |
| Containment Rate | Overall, KA-assisted, non-KA | % of conversations fully resolved by AI without human handoff. | >70% | Updated hourly | Date, Org |
| Organization Performance | Sessions, containment rate | Breakdown by Organization to identify high and low performers. | >70% across all Organizations | Updated hourly | Date, Org |
| AI Error Rate Trend | Error rate % by channel | % of AI calls that returned errors (e.g. throttling or outages). | <1% per channel | Updated hourly | Date, Org |
| KA Containment Impact | KA-assisted vs baseline | Measures whether Knowledge Assistant improves containment vs no-KA baseline. | KA-assisted > non-KA | Updated hourly | Date, Org |
| Session KPIs by Date | Full KPI table | Detailed table for daily analysis and root-cause investigation. | N/A | Updated hourly | Date, Org |
| Weekly KPI Comparison | WoW % change | Compares the last 7 days with the previous 7 days to identify trends. | Positive growth; declining anonymous share | Updated hourly | Org |
Dashboard 2: Intent Performance
Purpose: Evaluates the AI’s ability to understand and classify user requests
Key Question: Does the AI understand what customers are asking?
| Card | KPI(s) | What It Measures | Target / Benchmark | Data Freshness | Filters |
|---|---|---|---|---|---|
| Top Intents by Volume | Recognition rate, satisfaction, low-confidence count | Most common requests and how well they are handled. | >85% recognition | Updated hourly | Date, Org |
| Recognition Rate Trend | % over time | Tracks whether understanding is improving or declining. | ≥85% stable | Updated hourly | Date, Intent, Org |
| Confidence Score Distribution | HIGH / CLARIFY / LOW | How confident the AI is when classifying requests. | HIGH >60%; LOW <10% | Real-time | Date, Intent, Org |
| Response Time by Intent | Avg and p95 time | Speed of classification per intent. | p95 < 2,500 ms | Updated hourly | Date, Intent, Org |
| Intent Satisfaction Rate | Positive feedback % | User-perceived answer quality per intent. | 90% positive | Updated hourly | Date, Intent, Org |
| Hourly Activity Pattern | Volume by hour | When users are most active (capacity planning). | N/A | Real-time | Date, Intent, Org |
| Model Performance Comparison | Accuracy, confidence, response time | Compares AI models to identify best performer. | Highest accuracy; lowest latency | Real-time | Date, Intent, Org |
| Active Intent Count | Distinct intents used | How many configured intents are actively used. | ~80% active | Updated hourly | Date, Org |
| Intent Category Distribution | % by category | Distribution of traffic across business domains. | N/A | Updated hourly | Date, Org |
| Intent Confirmation Funnel | Auto-confirm vs explicit confirm | Measures friction in interactions. | Auto >50%; explicit <30% | Updated hourly | Date, Intent, Org |
| AI Error Rate Over Time | Error % | Tracks classification stability. | <1% | Updated hourly | Date, Org |
Dashboard 3: Knowledge Assistant
Purpose: Measures quality, speed, and reliability of AI-powered self-service answers
Key Question: Is the Knowledge Assistant providing good, fast answers?
| Card | KPI(s) | What It Measures | Target / Benchmark | Data Freshness | Filters |
|---|---|---|---|---|---|
| Assistant Performance Overview | Queries, answer rate, error rate, response time, satisfaction | Overall health and performance of each assistant. | >99% answer; <1% error; >85% satisfaction | Updated hourly | Date, Org, Assistant |
| Daily Assistant Trends | Volume, answered vs failed | Reliability trends over time. | ≥99% answered; failures near zero | Updated hourly | Date, Org, Assistant |
| KA Session Detail | Session ID, turns, duration, status | Detailed session-level troubleshooting and QA view. | N/A | Real-time | Date, Org, Assistant |
| Response Time by Assistant | Avg and p95 time | Speed of answer generation. | p95 < 3,500 ms | Updated hourly | Date, Org, Assistant |
| Model Usage in KA | Volume and latency per model | Which models are used and how they perform. | p95 < 3,500 ms | Real-time | Date, Org, Assistant |
| Top Queries by Assistant | Most frequent queries | Identifies user demand and content gaps. | N/A | Real-time | Date, Org, Assistant |
| Avg Turns per Session | Average turns | Interaction efficiency. | <3 turns | Real-time | Date, Org, Assistant |
| p95 Turns per Session | 95th percentile | Worst-case interaction complexity. | <6 turns | Real-time | Date, Org, Assistant |
| Avg Session Duration | Seconds | Total time to resolution. | <120 seconds | Real-time | Date, Org, Assistant |
| Distinct Assistants Used | Count | How many assistants receive traffic. | All configured assistants active | Real-time | Date, Org |
| AI Error Breakdown | Error type, endpoint, model | Breakdown of failures for troubleshooting. | <1% | Real-time | Date, Org |
Glossary
| Term | Definition |
|---|---|
| Session | A complete interaction (conversation or Knowledge Assistant) |
| Turn | One user message and one AI response |
| Channel | Where AI is used (Conversations, Knowledge Assistant, Shared) |
| Containment | Resolution without a human agent |
| Containment Rate | % resolved by AI (>70% target) |
| KA Containment Lift | Improvement vs baseline without KA |
| Authenticated Users | Logged-in users |
| Anonymous Sessions | Unidentified sessions |
| Total Reach | Authenticated users + anonymous sessions |
| Decision Band | Confidence level (HIGH, CLARIFY, LOW, FALLBACK) |
| Auto-confirm | AI acts without user confirmation |
| Explicit Confirm | User confirmation required before action |
| Answer Rate | % of queries answered |
| KA Error Rate | % of failed Knowledge Assistant queries |
| AI Error Rate | % of failed AI API calls |
| p95 Response Time | Time experienced by 95% of users |
| Week-over-Week (WoW) | Comparison of current vs previous 7 days |
| Recognition Rate | % correctly classified requests |
| Satisfaction Rate | % positive feedback |