Analytics [Preview]

The Analytics section of DIAL Admin shows how DIAL is used, what it costs, and what happened in individual conversations. It turns the raw request stream that flows through DIAL Core into data you can read in two ways:

• Ready-made views — Dashboards for overall usage, and Sessions for investigating single conversations.
• Your own questions — Queries over the same data, plus Tables and Pipelines to control what data is available.

Note

Analytics is a [Preview] feature. It is gated by the ANALYTICS_ENABLED environment variable and may change before general availability. All screens are backed by the Analytics Data Access Service (ADAS).

How Analytics fits together

Analytics is a small data pipeline you operate from the Admin panel:

raw events          SOURCE tables        enrichment PIPELINES      ENRICHMENT tables
(dial_usage_log, ─► (landing tables  ─►  (per-row/-group          ─► (derived columns,
 rate_analytics)     for raw data)        LLM or SQL transform)       e.g. session_insights)
                                     └─► aggregate PIPELINES ──────► rollup tables
                                                                      e.g. response_ratings
                                                                         │
                                         QUERIES, DASHBOARDS, SESSIONS ◄─┘
                                         read the tables above

1. Raw events land in source tables (for example the hop log dial_usage_log).
2. Enrichment pipelines read a source table and run a transform — an LLM call or SQL — to write derived columns into an enrichment table (for example session_insights). Aggregate pipelines group and roll up data into summary tables.
3. Queries, Dashboards, and Sessions read the resulting tables to analyze, visualize, and inspect the data.

Two source tables come pre-configured as read-only system tables:

• dial_usage_log — one row per call DIAL Core handled, ingested from blob storage.
• rate_analytics — user feedback events (rating and optional comment) for responses.

Dashboards and Sessions read these tables and the tables derived from them, so they work without any table setup. For example, the response_ratings rollup, built by the response_ratings_rollup pipeline over rate_analytics, holds one row per rated response. Tables and Pipelines are where you change what data is available. Insights shown in Sessions appear only after an enrichment pipeline has processed a session.

What do you want to do?

I want to…	Go to
See overall usage, cost, latency, and errors	Dashboards
Find a conversation that went badly and see why	Sessions
Answer a question the dashboards don't cover, and save the answer	Queries
Add a derived field, such as sentiment or topic, to conversations	Pipelines
Create a table, define its columns, or control who can read or write it	Tables

Screens

Screen	Best for	Description
Dashboards	Monitoring	System-wide usage, cost, latency, and activity dashboards, split by LLM, MCP, or Routes. Read-only.
Tables	Data setup	The catalog of analytics tables. Create source and enrichment tables, define their column schema, manage access, and get connection snippets.
Pipelines	Data setup	Enrichment and aggregation workflows that derive data from source tables. An enrich pipeline with an LLM transform is what the product elsewhere calls an evaluator.
Queries	Analysis	Author, save, and run read-only queries against analytics tables using a visual builder, JSON, SQL, or an AI assistant. Results render as a table or a chart.
Sessions	Investigation	Browse conversations with usage, cost, feedback, and LLM-derived insights, and drill into per-call traces. Read-only.

Where to start

1. Open Dashboards for the big picture.
2. Open Sessions to look into a specific conversation.
3. Use Queries when you need an answer neither view gives you.
4. Use Tables and Pipelines when you need data that isn't there yet, such as a new derived field.

Configuration

Platform engineers connect the Analytics data-access service and enable the screens through DIAL Admin analytics settings. Sessions, Dashboards, the Queries AI view, and pipeline runtime controls have additional settings there. Platform deployment, storage, and ingestion settings are in DIAL Analytics configuration. Enrichment execution requires the Enrichment Runner. Fresh installations need the recommended catalog initialization before Sessions can show derived data.