From messages to decisions
Build the picture, one level at a time.
The standard conversation-analysis path builds context in five stages, from individual human messages to patterns across analyzed customers.
Find the human messages
Separate customer messages from automated calls, then group the human messages into sessions. Editable rules let you define which messages count as human.
Identify signals in each turn
Read messages for sentiment, stated goals, frustration, confusion, success, and delight. Quoted signals give the next stage context.
Follow the session’s story
Bring the signals together: what was the customer trying to achieve, what happened along the way, and did they make progress? Session summaries connect friction, recovery, and features adopted.
Build the customer report
Combine sessions in the selected analysis window into a customer story, health score, churn-risk tier, behavioral profile, and recommended actions. Use these to decide which accounts need a closer look.
Find patterns across customers
The Cohort Report brings completed customer reports together into shared themes, recurring friction, positive signals, and product priorities for your team to investigate.
Follow an analysis in progress
The analysis console shows stage progress and session status during a run. See which sessions are queued, running, or complete while the report takes shape.
Built for AI products
Focus on the person behind the agent.
One customer message can trigger planning, tool calls, and retries. LiteLedger separates those automated calls from human messages before conversation analysis, so sentiment is grounded in what the person expressed.
Review and refine the split
Inspect classifications and correct individual turns. Editable rules determine which messages count as human; unrecognized traffic defaults to machine. Optional AI suggestions help you draft rules from corrections, ready for your review.
Keep the cost context
Interaction views connect human messages with related automated calls, helping you understand the cost of serving a customer request.
Evidence and interpretation
A score you can investigate.
AI identifies qualitative signals and writes the narrative. Code calculates the health score and churn-risk tier from structured analysis inputs, including session health, engagement trend, and activity recency.
A churn-risk tier helps prioritize investigation; it is not a measured probability that the customer will leave. Read it alongside the session story and your team’s knowledge of the account.
See the score, customer story, and recommended actionAcross your customer base
Turn individual experiences into product priorities.
Bring usage, spend, and customer-analysis results into a configurable dashboard. Read health and risk alongside activity to prioritize accounts with context.
Hear the customer’s words
Quote views bring together frustration, confusion, success, and delight signals from analyzed conversations.
Investigate shared patterns
Use the Cohort Report’s themes and recommendations to frame a product question, then explore the customer reports behind it.
Qualitative views reflect completed analyses. Choose the customers and analysis scope that fit your question.
Explore a question
Ask a question. Follow the evidence.
Ask retrieves workspace metrics, Cohort Reports, customer reports, session summaries, and synced conversation text to answer questions in plain English. Start broadly, then follow up on a customer or session.
- “What friction appears across the analyzed customers?”
- “Summarize this customer’s recent sessions and suggested actions.”
- “Find conversations mentioning the export workflow.”
Add product documentation to Primer to give the analysis and Ask your product terminology and context.
Getting started
Plan your first analysis.
What do I need to get started?
A LiteLLM proxy with conversation logs, a URL LiteLedger can reach, and a valid proxy credential. Connect and test it before syncing, then configure your supported AI provider.
See setup requirementsWhich customers and history are analyzed?
Choose customers and an analysis window from your synced history. Recognized account-key families are grouped into one customer. Reports describe the selected scope; the Cohort Report brings together available completed customer reports.
How do analysis costs work?
Configure your own provider credentials. Reports record an estimated analysis cost so you can review usage alongside the output.
Explore provider choices and costHow is the data handled?
Workspace members share access to workspace data, with membership and settings managed by admins.
Review data handling and customer controlsFollow one customer from friction to progress.
See the customer story, health signals, and recommended action together in a sample report.