Search earnings calls
Use the Risk Tool to run a query over earnings-call transcripts and create a dataset with call-level measures.
Before you search
Prepare:
- A draft or validated query string. The Keyword Tool helps build and validate one.
- The date range and filters that complete the query.
- A clear target sample for later interpretation.
NL Analytics searches English earnings-call transcripts at the sentence level.
Run the search
- Open the Risk Tool.
- Enter the query string using keywords, phrases, or supported query syntax.
- Complete the query by selecting the date range, filters, and adjacent-sentence behavior.
- Select measures and export options separately from the query.
- Name the dataset.
- Run the search. This creates a dataset and counts as one search against plan limits.
- Review the dataset results overview.
- Open matched sentences in the Snippet Tool to validate the query.
- Download the CSV output when the query is ready for downstream work.
Query and dataset settings
The Risk Tool separates settings that define the query from settings that control the dataset and its exports:
| Setting | Part of | What it does |
|---|---|---|
| Section of earnings call | Query | Restrict matching to the management presentation or to the questions-and-answers part of the call. |
| Speaker affiliation | Query | Restrict matching to corporate participants or to conference-call participants such as analysts. |
| Speaker title | Query | Restrict matching to preset speaker roles, including CEO, CFO, COO, CTO, Analyst, or Operator, or to custom title patterns you define. |
| Economic sector | Query | Restrict matching to selected economic sectors. |
| Headquarters country | Query | Restrict matching to companies headquartered in selected countries. |
| Include adjacent sentences | Query | Detect risk and sentiment words in a three-sentence window around each matched sentence: one sentence before and one after, by the same speaker, where available. Query-string matching itself stays at the single-sentence level. |
| Include per-keyword breakdown file | Export | Export per-keyword counts so each keyword's contribution can be analyzed separately. See keyword_breakdown.csv. |
| Dataset name | Dataset | Label the dataset so it is easier to find in My Datasets. |
Specific companies cannot be pre-selected before a search. To focus on particular firms, filter the dataset export or use the Snippet Tool's company filters after the search.
Record the query's section, speaker, country, sector, and adjacent-sentence settings because they change what the measures count. See How query settings change the counts.
Very broad queries
A search is rejected when its estimated matches exceed a share of all sentences in the corpus (the default threshold is 3 percent). If that happens, narrow the query with more specific phrases or exclusions, or contact support if the broad search is essential to the project.
The results overview
After a search completes, the dataset page summarizes what matched:
- A time series of the selected measure, shown quarterly or weekly, with optional smoothing and a headquarters-country filter.
- A bar chart breaking the measure down for a selected period.
- Three tables: the keywords with the most matches, the companies with the most matched sentences, and breakdowns by sector (economic, business, and SIC).
- Totals for matched sentences and for the sentences, transcripts, and firms searched.
Use the overview as a first sample availability and variation check: if a single keyword, firm, or sector dominates the matches, or the time series is flat where the design needs variation, inspect those sentences before relying on the measure.
Output
The Risk Tool dataset export is call-level: one row per earnings call, identified by earningscallID, including calls with zero matches in the selected date range. The measure columns are exposure, risk, positive, negative, and sentiment.
See Export a research panel for the available files, their columns, and join guidance.