NL Analytics documentation

Convert English earnings-call transcripts into auditable call-level measures for topic exposure, risk, and sentiment.

Key workflows

• From research idea to panel: Follow the main workflow from a topic idea to an exported call-level panel.
• Worked example: supply chain disruption: See every step applied to one concrete construct.
• Corpus scope and coverage: Check the coverage snapshot and learn how to verify coverage for your sample.
• Query syntax: Write keywords, phrases, Boolean logic, exclusions, and wildcard searches.
• Metric definitions: Understand Exposure, Risk, Positive Sentiment, Negative Sentiment, and Sentiment.
• Export a research panel: The output files, their columns, and identifiers for downstream joins.
• Known limitations: What the product and the measures do not do.

What NL Analytics does

NL Analytics turns matched earnings-call sentences into transparent call-level measures. The corpus is English earnings-call transcripts; you cannot upload your own text.

The core workflow is:

1. Start with a research question or topic.
2. Translate that idea into keywords, phrases, synonyms, and exclusions with the Keyword Tool.
3. Search earnings-call transcripts with the Risk Tool.
4. Inspect matched sentences in the Snippet Tool.
5. Refine the query until the matches support the intended construct.
6. Export call-level CSV output for downstream work.

What the measures are

The metrics are raw integer sentence counts calculated at the earnings-call level:

• Exposure counts topic-matched sentences.
• Risk counts topic-matched sentences that also contain risk or uncertainty language.
• Positive Sentiment and Negative Sentiment count topic-matched sentences that also contain financial sentiment words.
• Sentiment is positive - negative.

See Metric definitions for the exact definitions and interpretation caveats. The counts are built from explicit queries and curated word lists, so every value can be traced back to inspectable matched sentences.

Before using results

Results depend on query design and corpus coverage. Before treating an output as a measure, inspect matched sentences, check coverage and variation in the target sample, and record the search per the reproducibility checklist.

Where to go next

Start with From research idea to panel if you are designing a new measure, or read the worked example first to see the full path once. Use Query syntax when you need exact search rules, Manage searches when you need a prior search, and Get help when something does not behave as documented.