Airgentic Help
This module covers how to use Airgentic's analytics to understand performance and continuously improve your service. Read this when you want to move beyond basic operation into data-driven optimisation.
Customer Service Insights shows how conversations are performing:
| Section | What it tells you |
|---|---|
| Summary Stats | High-level counts: total outcomes, fully answered, partially answered, unanswered |
| Answer Completeness | Proportion of conversations in each completeness category |
| Expert Agents | How conversations are distributed across agents |
| User Feedback | Positive vs negative ratings |
| Non-Neutral Sentiment | Emotional tone of conversations |
| Products Mentioned | Which products generate questions |
| Top Citation URLs | Which content pages are used most |
Are users getting answers?
- Check answer completeness — what percentage are fully answered?
- Unanswered and partially answered conversations indicate problems
Are users satisfied?
- Check user feedback — what's the positive/negative ratio?
- Check sentiment — are conversations trending positive or negative?
What's driving volume?
- Check products mentioned — which products generate the most questions?
- Check top citations — which content is working hardest?
Filters let you drill into specific subsets:
| Filter | When to use it |
|---|---|
| Date range | Compare periods, track trends |
| Answer Completeness | Focus on problem conversations |
| Website | Separate deployments or sections |
| User Rating | Find dissatisfied users |
| Sentiment | Identify emotional conversations |
| Needs Review | Prioritise flagged items |
Related: Customer Service Insights
Search Insights shows how users search and what they find:
| Section | What it tells you |
|---|---|
| All Searches | Chronological log of every search |
| Top Search Queries | Most frequently searched terms |
| Top Clicked Results | Most frequently clicked pages |
| Timeline | Search and click trends over time |
| Locations | Geographic distribution of searchers |
What do users want to find?
- Check top search queries — what are users looking for?
- Are these topics well-covered in your content?
Are users finding what they need?
- Check top clicked results — which content is working?
- Compare search volume to click volume — large gaps indicate problems
What's missing?
- Look for high-volume searches with low clicks
- Look for searches that don't match your content terminology
Related: Search Insights
| Metric | Target | What low scores indicate |
|---|---|---|
| Fully Answered % | High (70%+) | Content gaps, prompt issues, poor search |
| Unanswered % | Low (<10%) | Missing content, questions outside scope |
| Partial Answer % | Monitor | May be acceptable or need investigation |
| Metric | Target | What to look for |
|---|---|---|
| Positive ratings | High | Indicates users are satisfied |
| Negative ratings | Low | Each one is worth investigating |
| Negative sentiment | Low | May indicate frustrated users |
| Metric | What it indicates |
|---|---|
| Conversation volume | Demand for the service |
| Questions by agent | Which agents are working hardest |
| Needs attention count | Workload for human review |
| Usage vs allowance | Capacity and planning |
High unanswered rate:
1. Review unanswered conversations — what topics come up?
2. Check if content exists for those topics
3. If yes → Check search configuration, prompts, categorisation
4. If no → Create content or add curated answers
High negative ratings:
1. Review conversations with negative ratings
2. Look for patterns — same topic, same agent, same type of error?
3. Address root cause (content, prompt, search, or legitimate limitation)
High escalation requests:
1. Review conversations where users asked for humans
2. Was the AI handling it poorly, or do users just prefer humans?
3. If AI issue → Fix the underlying problem
4. If preference → Consider whether AI-only is appropriate for this service
| Finding | Action |
|---|---|
| Users ask about topics not in content | Add content or curated answers |
| AI gives wrong answers for specific questions | Add curated answers |
| Search returns irrelevant results | Adjust search settings, synonyms, or boosts |
| Users phrase things differently than content | Add synonym rules |
| AI tone is wrong for certain situations | Refine prompts |
| Certain agent routing is broken | Adjust agent role descriptions |
Not everything needs immediate action:
| Priority | Criteria |
|---|---|
| High | High volume + negative outcome + fixable |
| Medium | Moderate volume + clear fix |
| Low | Low volume or unclear fix |
| Watch | May not need action, just monitoring |
Focus energy where it will have the most impact.
Spend 30-60 minutes weekly:
Spend 1-2 hours monthly:
Spend a few hours quarterly:
| Metric | Time period |
|---|---|
| Total conversations | Monthly |
| Fully answered % | Monthly |
| Positive rating % | Monthly |
| Top 5 topics | Monthly |
| Key improvements made | Monthly |
Export data from Insights for analysis in spreadsheets or BI tools.
Problem: AI can't answer common questions
Solution: Add content to your website, then sync
Measure: Unanswered rate for those topics
Problem: AI answers incorrectly or inconsistently
Solution: Add curated answers for specific intents
Measure: Negative ratings for those questions
Problem: Users search for terms not in your content
Solution: Add synonym rules
Measure: Search click-through rates
Problem: AI tone or behaviour is wrong
Solution: Refine prompts
Measure: Sentiment and ratings for affected conversations
Not every negative rating is actionable. Some users are frustrated before they arrive. Focus on patterns, not isolated cases.
Make incremental changes so you can measure impact. If you change everything, you won't know what worked.
It's easy to assume things are fine. Regular review catches problems before they accumulate.
A curated answer is a quick fix, but if the underlying content is wrong, fix the content. Curated answers should be exceptions, not the primary solution.
Back to: Optional Deep-Dives