Measure the quality of the experience your users have with your AI agents.
Stop guessing what your AI assistants are worth. Analyze what's said in conversations, spot what the agent can't do, and act on the users concerned without switching tools.
- Via SDK or APIin-house or third-party agent
- Redacted datain the analyzed conversations
- Hosted in FranceAnalysis by Mistral
See whether your agent's answers are actually useful, not just fast.
Your agent works. Engineering teams see whether it answers quickly and without errors. Product and CX teams don't see whether the answer was useful, or what the user does next. You need both views.
What you see in your conversations.
What your teams don't see in the logs: what the user was looking for, what they got, and how it left them feeling.
Satisfaction and dissatisfaction
What annoys, what delights
A satisfaction score per agent, rage prompts and frustration, and the topics where dissatisfaction concentrates.
Unsupported requests
Material for the roadmap
What your users expect from the agent that it can't do yet, with their volume and resolution rate.
Identified issues
Where to fix first
The issues detected in conversations, the users affected, and what they cost in retention and conversion.
Safeguards
What the agent shouldn't say
Conversations that violate your safeguards, by severity, and the agent's risk level.
Themes and intents
Without reading thousands of lines
Conversation topics and their share, period after period.
Incremental effect
What the conversation brings
Users who talked to the agent compared with similar users without a conversation, topic by topic: resolution, retention, conversion.
Every conversation becomes a criterion.
Filter conversations on what the analysis found in them: intent, outcome, satisfaction, topics, safeguards. Then save their users to a segment, in one click.
- Sixteen criteriadrawn from conversation content, from satisfaction to resolution reached.
- A segment in one gowith the users of the listed conversations.
- Reusable everywhereto target a survey or an in-app message.
Connect, analyze, act.
A single connection, then continuous monitoring from Screeb.
- 1
Connect
Via SDK or API
Whether the agent is built in-house or provided by a third party.
- 2
Analyze
By intent and by theme
Conversations are grouped, and friction is spotted automatically.
- 3
Act
On the users concerned
Replays, segments, messages and product tours can be targeted based on what the analysis found.
A measurement that stays with you.
Your agents' conversations carry your customers' most sensitive information. Analyzing them shouldn't create one more risk.
- 100% hosted in FranceScaleway and OVHcloud, no data leaves the country.
- Trusted AIMistral, a French company.
- Outside the CLOUD ActNo foreign authority can demand access to your conversations.
- GDPR complianceBy design, not as an option.
Personal data redacted in analyzed conversations, through pattern detection followed by server-side verification.
Your questions.
How is Agent Analytics different from an observability tool?
They're complementary. Your observability tools monitor the technical health of the AI: latency, costs, traces, errors. Agent Analytics analyzes what your users ask for, what they get and what they do next.
How does it connect to our agent?
Via SDK or API, whether the agent is built in-house or provided by a third party.
What happens to unsupported requests?
They're ranked by volume and by cost in retention. You can show a product tour to the users concerned, and get back to them on the day the use case is covered.
Can we find users based on their conversations?
Yes: a conversation's classification is added to the user's profile. It works as a filter in your replays, a segment criterion and a targeting rule for your messages.
How does it relate to the TRUE Score™?
Agent Analytics analyzes how conversations unfold. The TRUE Score™ is the method for measuring perceived quality, valid for any AI experience, conversational or not. Details are on the TRUE Score™ page.
Where are analyzed conversations hosted?
In France, with Scaleway and OVHcloud.
Who owns AI agent analysis?
The Voice of Customer lead and product teams, depending on whether the issue is satisfaction or the agent's capabilities. The day-to-day of this role is covered on the Head of Voice of Customer page.
The other building blocks of Voice of Customer
- Feedback & Surveys In-app surveys, email and CRM surveys
- User Interviews Recruit and interview your users
- AI-Assisted Surveys Describe your goal, the survey is ready
- TRUE Score™ The perceived-quality score of your AI experiences
- Public Reviews Google, Trustpilot, App Store, Google Play and G2
- AI Perception What ChatGPT, Google and Claude say about your brand
Know what your agents bring to your users.
Bring a use case, and we'll show you what Screeb does with it.




