# AI Evaluation

## Your AI agents, held to your standard.

Score every AI interaction automatically, catch errors and hallucinations before they reach customers, and use the findings to make your models sharper over time.

## Evaluations

ID #890321746102564

### Chat

**D**  
Hi! I’d like to request a refund for the two tickets I bought yesterday. Something came up and I won’t be able to attend anymore.  
**Dan**  
15:59:53 UTC  
Hi there! I'm sorry to hear that. I can help with general questions, but refund requests need to be handled by our billing team. Could you share your order number so I can forward your request to them?  
**Drake**  
16:02:20 UTC  
**D**  
Sure, my order number is #128213.  
**Dan**  
16:08:02 UTC  
Switched to email

Hi Dan,

Thanks for sharing your order number. I've forwarded your refund request to our billing team for review. A member of the team will reach out shortly regarding the status of your request.

Best,

Drake  
**Drake**  
16:15:25 UTC

### AQS

| Metric                             | Score |
|------------------------------------|-------|
| AQS                                | 90    |
| Opening                            | 80%   |
| Meeting Customer Needs              | 45%   |
| Effective Resolution                | 85%   |
| Communication Skills                | 95%   |
| Product Knowledge                   | 90%   |
| Data Capture & System Accuracy      | 95%   |

### Spot support issues before your customers do.

Quality Score Trend  
86%

Score every interaction automatically. Get visibility across channels with automated scoring for 100% of AI-generated responses.

### Escalation requested

Target hallucinations and errors immediately. High-risk AI responses are flagged instantly, giving your team immediate insight into what needs review.

#### Explanation

Meta Support Agent: Should ask for reservation number first before offering a refund for the two tickets yesterday to confirm that you are refunding the correct ticket in case there were multiple purchased.

### Improve AI models with targeted feedback

Use feedback to make refinements that help your AI agent of choice improve over time.

### Evaluations

Support team evaluation

| Metric           | Score     |
|------------------|-----------|
| Testing          | 80%       |
| Accuracy target   | 75%       |
| Current accuracy  | 951       |
| Calibrated       | 100       |

### Scale QA coverage from 1% to 100%.

No sampling. No backlog. Solidroad reviews every interaction across your human and AI agents, giving your QA team visibility into what’s actually happening across support.

### Identify high-risk interactions.

Pattern detection and scoring logic work together to surface interactions that pose customer, compliance, or brand risk, and elevate them to your QA team for immediate review.

### Build a QA engine tailored to your team.

Solidroad’s AI model is trained on your real conversations, policies, and scorecards, so that every reviewed interaction is scored according to your voice, workflows, and expectations.

We have complete visibility into quality across hundreds of thousands of conversations. We can verify agent readiness before it impacts customers.”

### Alex Dimitrov

Head of Learning and Development

18% Reduction in average handling time  
3% Increase in CSAT scores

### Case study

[Case study](/content/customers/crypto/index.html)