Behavioral data layer · Earnings calls
Per-speaker baselines. Novel language detected.
Spoken Alpha is the behavioral data layer for earnings call research. We build per-speaker historical baselines for every public-company executive and flag when language breaks from their established pattern — at the phrase level, with exact call attribution. Built for research analysts, equity teams, and hedge fund researchers.
How it works
Same speaker. Their own history. The deviation that matters.
Generic sentiment scoring against cross-company corpora is noise. Per-speaker longitudinal deviation is information — because a CFO who suddenly hedges 47% more than they ever have isn't a “tone shift,” it's a specific break from that person's established pattern.
Transcripts ingested per-call, per-speaker
Every public-company earnings call — transcripts, IR pages, and participant rosters — pulled and structured as soon as the call goes out. Transcript infrastructure powered by EarningsCall.biz.
Per-speaker historical baselines built across quarters
Each executive is measured against their own prior calls, going back years. We surface deviations in hedge density, qualifier load, prepared-vs-Q&A divergence — patterns that cross-company sentiment misses entirely.
Novel language flagged at the phrase level
When an executive uses language they have never used before — verbatim phrases, rhetorical structures, hedging patterns — we surface it with exact call attribution. Not a score. A specific deviation you can read.
What this looks like in practice
A real example: TMO Q1 FY2023.
Spoken Alpha identified 11 verbatim phrases the CEO had never used in his prior 12 earnings calls — three weeks before guidance cuts began appearing in the following quarters.
11
Verbatim phrases the CEO had never used in prior calls
12
Prior calls in our baseline for this speaker
Q1
FY2023 call — novel language flagged at the phrase level
This is not a black-box score. A research analyst can open the call, see exactly which phrases broke from the baseline, read them in context, and make their own judgment about what they mean. That's the product.
Who it's for
Built for researchers. Available to builders.
Read the call differently.
Per-speaker behavioral data for fundamental researchers, hedge fund analysts, and IR competitive intelligence teams. Built as a research input, not a black-box score.
- Per-speaker historical baselines — each executive vs. their own prior calls
- Novel language detection flagged at the phrase level
- Hedge density, qualifier load, Q&A evasion patterns per speaker
- Call-level summaries with deviation breakdown and attribution
Earnings call data, as an API.
Machine-readable IR data and derived scoring for AI agent builders. Self-serve, free tier available.
- IR URLs, earnings call metadata, press releases
- Parsed transcripts and Q&A exchanges
- Derived scoring data (no proprietary deviation internals)
- Tiered pricing · self-serve signup
How we fit
Adjacent to AlphaSense and RavenPack. Specialized on call language.
Spoken Alpha sits in the behavioral data layer — the same category as alternative data vendors serving institutional research. Our focus is narrow by design: per-speaker longitudinal baselines on earnings calls, not a broad document or event platform.
AlphaSense
Document search + summarization across filings, calls, broker research
Broader document universe; no per-speaker longitudinal baseline
RavenPack
News and event sentiment, structured event detection
News and events layer; call language is one of many inputs
Sentieo
Collaborative research platform with transcript search and tagging
Workflow and collaboration focus; quantitative NLP is secondary
Transcript infrastructure via EarningsCall.biz. We process; they source.
The methodology
A research-grounded methodology.
The phenomena we detect — hedging, conditionality, ownership detachment, novel language — are well-established constructs in how corporate language reveals information. Our contribution is the engineering: modern language models scoring every public-company call against every speaker's own longitudinal baseline.
Request research access.
Research access is by application. We scope each engagement to what you're actually trying to answer — let us know your use case.