Skip to main content

What it does

Fetches and parses the most recent earnings call transcript to extract management’s forward-looking guidance on revenue and operating margins. When guidance is available, it provides a better starting point for DCF assumptions than extrapolating historical trends. This skill runs as Phase 5.5 — after cost of capital and before growth & profitability. It is optional: if no transcript is found or no usable guidance is extracted, the pipeline proceeds with LTM-derived defaults.

Why earnings guidance matters

Historical financials tell you where a company has been. Earnings guidance tells you where management expects it to go. For companies undergoing rapid change — high-growth tech, turnarounds, recent IPOs — the gap between past performance and forward expectations can be enormous. When management provides explicit guidance, the plugin uses the conservative end of any range as the default starting point, which the user can then adjust.

What it extracts

The skill looks for guidance on four forecast variables: Two additional variables are stored for reference but not currently used in the pipeline:
  • Sales-to-capital — CapEx guidance, useful for cross-checking reinvestment assumptions
  • Cost of capital — refinancing commentary, interest rate improvements

How it works

1

Determine the expected earnings call

From statements_required, identify the most recent filing by fiscal period end and map it to the corresponding earnings call quarter (e.g., 10-K with FY end 12/31/2025 → Q4-2025).
2

Check the cache

Look for an existing guidance-Q{N}-{YYYY}.json in the ticker cache. If found, load and return it — no fetch needed.
3

Fetch the transcript

Search for the transcript using a priority cascade: company IR website, Motley Fool, Investing.com, SEC 8-K exhibit. Fetches and filters the transcript to the most relevant 5,000–8,000 tokens (prepared remarks + forward-looking Q&A).
4

Extract structured guidance

An LLM prompt extracts guidance for each variable, including raw quotes, inferred numerical values, inference method, and confidence rating.
5

Save to cache

Two files are saved: a .json file (consumed by the pipeline) and a .md file (human-readable with raw quotes and reasoning).

Inference methods

Management rarely says “our revenue will be exactly X.” The skill handles several patterns of guidance: Each extraction includes a confidence rating (high, medium, or low) so downstream skills can weight it appropriately.

Mode behavior

Output

Two files are saved per extraction to the ticker cache:
  • guidance-Q{N}-{YYYY}.json — structured data consumed by growth-and-profitability and the lucky pipeline
  • guidance-Q{N}-{YYYY}.md — annotated markdown with raw management quotes, inference reasoning, and confidence ratings

Edge cases