Finance
Track earnings guidance against actuals
A research agent reconstructs each company's guidance from archived filings and transcripts, grades it against reported results, and stores every cited claim in a Snowflake ledger.
Quick Start
Inputs
- Ticker watchlist The companies to track, for example "NVDA, MSFT, JPM, JNJ", each with a one-line note on its guidance style.
- Quarters to backfill How far back to reconstruct, for example 8 quarters, built in one batch on day one.
- Follow-up questions Natural language, for example "Which company beat revenue guidance most often?" answered as SQL over the ledger, or new research questions sent to the live agent.
Outputs
- Verdict grid: beat, miss, or inline per ticker and quarter
- Guidance ledger in Snowflake with verbatim guidance strings and parsed ranges
- Citation audit trail: source URL, confidence, and excerpt for every value
- Slack scorecard posts per ticker
How it works
A 6-phase pipeline.
- Configure the analyst One Web Search Agent with a sell-side analyst persona, acceptance-criteria goals, and prioritized sources: SEC EDGAR, investor relations pages, call transcripts, and financial press.
- Backfill the history The agent researches each ticker's past guidance and reported results from archived sources, 4 quarters per run, in 24 chunked, concurrent, resumable runs.
- Grade deterministically The app parses guided ranges like "$26.0B to $26.5B" in code and computes beat, miss, or inline per metric, refusing scale-mismatched comparisons instead of fabricating verdicts.
- Store with provenance Three Snowflake tables hold raw runs, graded metrics, and field-level claims; headline views cover formal ranges, segment guidance, and outlook-only companies.
- Post to Slack Each ticker's scorecard lands in the channel: quarter-by-quarter verdicts with the track record.
- Answer questions Aggregation questions become SQL over the ledger, shown transparently; new questions go back to the live agent with the ticker's research context.
Stack
Nimble primitives plus the full runtime stack.
Nimble APIs
What it does
- Web Search Agents Research each ticker's guidance and reported results from archived filings, releases, and transcripts, returning schema-conforming output.
- Trust framework Field-level provenance: every guided and actual value maps to a claim with confidence, source URL, and excerpt.
3rd Party Tools
Role
- Snowflake The guidance ledger: runs, graded metrics, and claims tables plus headline views.
- LlamaIndex Natural-language-to-SQL query engine over the ledger.
- claude-sonnet-5 Writes the SQL for Ask-the-Desk questions.
- Slack incoming webhooks Scorecard delivery after each ticker lands.
- Streamlit Scorecard grid, company drill-down, and Ask the Desk pages.
Reach out if you have any questions.
Talk to an Expert