Financial trend index
Illustrative indexed values; not reported results for a real company.
Earnings-call intelligence
Turn transcripts and performance data into evidence-backed company intelligence.
Use coordinated AI workflows to extract reported facts, compare management narrative with performance, track quarter-over-quarter themes, surface risks, and produce summaries that can be evaluated before they reach investors or downstream systems.
This page uses synthetic example data to demonstrate the workflow. It is not investment advice and does not contain a recommendation for a real security.
Agent workflow
Break the analysis into auditable responsibilities.
Specialized steps make it easier to test errors, preserve provenance, and improve one part of the pipeline without changing everything else.
Source & normalization
Collect transcript, reported financials, performance JSON, prior-quarter outputs, and metadata. Preserve source identity, period, timestamp, and document lineage.
Fact & theme extraction
Extract reported metrics, guidance, operating drivers, management claims, risks, questions from analysts, and evidence spans from the transcript.
Quarter-over-quarter comparison
Compare the current quarter with historical performance and prior management narrative. Separate measured changes from narrative interpretation.
Evaluation
Score factuality, numerical consistency, material coverage, risk identification, evidence support, readability, and agreement with structured performance data.
Delivery
Produce a company summary, portfolio/watchlist brief, newsletter section, or analyst-assist output with citations, confidence, and clear distinction between facts and inference.
Illustrative company
Example: how evidence and narrative can move together.
The charts below are synthetic. Their purpose is to show the kind of comparative view an earnings-intelligence product can generate.
Narrative signal trend
Illustrative qualitative scores derived from hypothetical transcript themes.
Output design
Separate facts, interpretation, and decision support.
This prevents a fluent model narrative from being mistaken for a sourced financial fact.
Reported facts
Metrics, guidance, segment changes, dates, and management statements with source evidence.
Derived intelligence
Performance scores, momentum, narrative change, risk intensity, peer comparison, and confidence.
User decision layer
Questions to investigate, watchlist criteria, scenario exploration, and portfolio context—without presenting synthetic examples as real recommendations.
Prompt experimentation
Compare summary prompts on the same evidence.
When Prompt A and Prompt B summarize the same earnings call, evaluate them against the transcript and structured performance numbers—not against style preference alone.
| Dimension | What to judge | Failure example |
|---|---|---|
| Factuality | Claims supported by transcript or structured data | Invented driver or unsupported causal statement |
| Numerical consistency | Correct periods, units, direction, and magnitude | Confusing sequential and year-over-year change |
| Materiality | Coverage of changes important to the company | Overweighting minor commentary |
| Risk coverage | Meaningful headwinds and uncertainty | Positive summary that omits a guidance cut |
| Evidence | Traceable source spans or fields | Cannot explain why a conclusion was made |
| Usefulness | Helps the intended reader decide what to inspect next | Accurate but generic recap |
FAQ
Earnings intelligence questions.
Does this page provide investment recommendations?
No. The page illustrates an earnings-analysis workflow. Example company names, scores, charts, and outputs are synthetic and are not investment advice.
What should an earnings-analysis system retain as evidence?
It should retain transcript citations or spans, reported metrics, period labels, source timestamps, model and prompt versions, and the calculations used for derived scores.
How should two summary prompts be compared?
Use the same transcript and performance data for both variants, then evaluate factuality, coverage, materiality, numerical consistency, risk identification, readability, and downstream usefulness with human or governed LLM judging.
Build an earnings workflow that can be evaluated, not just prompted.
DataKnobs can connect transcripts, structured company performance, prompt variants, LLM judging, human review, and downstream publishing into one governed evaluation loop.
Discuss the Workflow