The research library
Every study below is computed from the same corpus: 165,000+ earnings-call transcripts our AI read and scored with an identical battery of questions, joined to a 22,449-call sample of 12-month market outcomes. We publish what the data shows — including when the answer is "no edge."
The methodology behind every page on this site: the 37-question battery, the A–F grade, the expected-move model, and the honest story of 1,600 tested hypotheses.
Call Signals
Critic Ammunition: What 165,000 Earnings Calls Show
Most earnings calls invite criticism, and most contain it: of 165,182 transcripts read by an AI, 145,051 — about 88% — show what we call critic ammunition, language that hands skeptics material to use against the company…
Skeptic Reassured: What 165,000 Earnings Calls Show
When a skeptical analyst pushes hard on an earnings call and the management team answers with substance instead of deflection, something measurable changes. We call this pattern "Skeptic Reassured." An AI read 165,182 ea…
Underused Fixed Costs: What 165,000 Earnings Calls Show
When a company says its fixed costs are "underused," it is arguing that extra volume would flow almost straight to profit. This study asks how often that framing appears in earnings calls and what tends to come with it. …
Pricing Recovering: What 165,000 Earnings Calls Show
Pricing recovering — calls where management signals that price increases are sticking or coming back after a weak stretch. It is one of the most-watched signals on any call, because pricing power is the cleanest lever on…
The Hidden Segment: What 165,000 Earnings Calls Show
Every earnings call has a moment where the numbers disappoint but the story stays rosy — and some companies live there. Across 165,182 transcripts read by AI, 34,917 calls (21.1%) fit a pattern best described as results …
Founder-Led Companies: What 165,000 Earnings Calls Show
Founder-led companies — those still run by the person who started them — account for 33,092 of the 165,182 earnings calls in Artul.ai's library, about one in five. Because founders speak with unusual authority and often …
Business Verdicts
Management Behavior
20-Year Trends
Signal Combinations
Rehearsed and Vague: When Two Signals Appear Together
When an executive leans on rehearsed talking points while staying vague on substance, the combination is easy to miss in real time but stands out when measured at scale. Across 165,182 earnings-call transcripts read by a…
Hypotheses Tested
Hypothesis Tested: Uncontested runway
Some companies spend their earnings calls describing a growth runway that nobody on the Q&A line contests — analysts don't push back, and management doesn't hedge. This study, "Uncontested runway," tests that hypothesis …
Hypothesis Tested: Coming out of the tunnel
Some companies spend a stretch of calls describing a downturn, then at some point start talking as if they have come out the other side. This study tests that pattern: an AI read 165,000 earnings-call transcripts and fla…
Hypothesis Tested: Answers go deeper than the script
When an executive stops reading the script and actually answers the question, does the call get better? To find out, we had an AI read 144,500 earnings-call transcripts from the 165,000-call corpus and score each one on …
Hypothesis Tested: Earned edge, pressed harder
Some companies claim an edge they've earned — a durable advantage they can point to — and this study asks what happens when management presses that claim harder on the call. Among 485 calls in our 165,000-transcript corp…
Hypothesis Tested: Second demand front open and funded
When a company announces that a second demand front has opened and been funded — new customers or products beyond the core, backed by real investment — how does that show up on the call? We tested this hypothesis against…
Hypothesis Tested: Paid to expand
When a CFO takes center stage on an earnings call, the message is usually "we're spending to expand" — and it shows in the numbers. Across a corpus of 165,000 AI-read transcripts, 123 calls fit this "paid to expand" patt…
Hypothesis Tested: External validators converging
When multiple outside references converge on the same claim in an earnings call — analysts citing the same third-party data, or management pointing to the same industry sources — that's a pattern worth testing. Among 497…
Hypothesis Tested: Tone of discovery
Some earnings calls are organized around a discovery: management announces something it just learned — pricing is recovering, an early product is scaling faster than expected — and the rest of the call is built on that r…
Hypothesis Tested: Still getting better as they speak
Some executives warm up as the call goes on: their answers get more confident and more specific in the second half than the first. This study isolates that pattern — 106 calls out of 487 flagged by Artul.ai's AI reader, …
Hypothesis Tested: Engine explained, runway named
When a company explains why a metric moved and names the runway ahead, calls read differently: more candor, more specificity, less stress. This study tests that pattern across 165,000 earnings-call transcripts read by an…
Hypothesis Tested: Strong facts, held-back story
Some executives load their calls with hard specifics while keeping the narrative quiet — strong facts, held-back story. We tested this pattern across 165,000 earnings-call transcripts read by an AI, isolating 101 calls t…
Hypothesis Tested: Substance without an audience
Some earnings calls deliver real substance — specific numbers, candid discussion, concrete guidance — but almost nobody seems to be listening: low analyst engagement, thin coverage, no audience. This study isolates that …
Hypothesis Tested: The feared thing keeps not happening
On earnings calls, management often names a fear — a churn risk, a cost headwind, a regulatory question — and then, quarter after quarter, the feared thing keeps not happening. We tested this pattern across 165,000 AI-re…
Hypothesis Tested: Secure base, several live doors
Some companies face pressure from multiple directions at once yet keep answering from a position of security — candid about problems, specific about numbers, and clearly working several live strategic doors rather than o…
Hypothesis Tested: Two-sided intensity
Some earnings calls read like a seesaw: strong promotion and confidence on one side, with candor and specificity on the other. This study tests whether that two-sided intensity pattern means anything. Out of 165,000 tran…
Hypothesis Tested: More where that came from
Some earnings calls leave investors expecting a follow-up — a pattern our library tags "More where that came from." An AI read 165,000 earnings-call transcripts and scored each one on the same battery of questions, letti…
Hypothesis Tested: Leaning into the storm
Some management teams respond to a bad quarter by confronting it head-on: naming the problems, quantifying the damage, and taking questions squarely. We tested whether that "leaning into the storm" posture shows up measu…
Hypothesis Tested: Acted like it's already bigger
Some executives talk as if the company is already much larger than it is — describing capacity, addressable markets, and volume as if scale were in hand rather than ahead. We tested that hypothesis on 165,000 AI-read ear…
Hypothesis Tested: Compounding evidence
Some earnings calls don't just show one good sign — they stack several at once: fast-growing early products, tiny market share, founder leadership, and volume about to step up. This study tests whether that compounding o…
Hypothesis Tested: Senior hire recruited
When a company announces a senior hire recruited from outside during an earnings call, it often signals a strategic pivot — a new growth push, a turnaround, or a bet on a capability the firm lacks. This study asks what e…
Hypothesis Tested: Demand off the charts
When management teams describe demand as "off the charts" — signaling surging orders, capacity constraints, or backlog growth — what does that actually look like on the call? Our AI read 165,000 earnings-call transcripts…
Not investment advice. Artul.ai publishes AI-generated earnings-call quality grades and expected-volatility estimates — never buy or sell recommendations. We tested over 1,600 predictive hypotheses against 165,000 transcripts; the honest result, including what failed, is documented in our
methodology.