Loud Backlogs and Busy Phones: Calls Where the Model Saw a Volume Step-Up
This study examines 23,140 earnings calls—14.01% of the 165,182-call Artul.ai corpus spanning 1990 to 2026—where the model answered YES to "Volume About to Step Up" and the backlog was read as growing. Compared with the corpus baseline, these calls score higher on promotion (5.60 vs 5.05) and confidence (7.63 vs 7.21), with stress lower (2.21 vs 2.43). Guidance was raised on 32.36% of these calls versus 21.05% in the base. Language lifts are led by "Deferred Revenue Growing" at 2.17x. Among 2,598 calls with return data, the median next-window return was -7.76%, versus -7.16% for the 22,449-call base, and the beat rate was 39.53% versus 39.47%—no directional edge.
- The pattern covers 23,140 calls, 14.01% of the 165,182-call corpus (95% CI: 13.84% to 14.18%).
- Confidence averages 7.63 vs 7.21 in the base, promotion 5.60 vs 5.05, and stress 2.21 vs 2.43.
- Guidance was raised on 32.36% of these calls versus 21.05% of base calls, and lowered on 7.52% versus 11.56%.
- Median next-window return among 2,598 eligible calls was -7.76% versus -7.16% for the base, with beat rates of 39.53% and 39.47% respectively.
1Introduction
Few signals on an earnings call sound as promising as a growing backlog paired with a claim that volume is about to step up. Management teams using this language talk a particular talk: more confident, more promotional, less stressed. Whether that tone accompanies better outcomes is a separate question from whether it appears. For anyone who parses calls for a living, the combination is common enough—about one call in seven in the Artul.ai library—that its behavioral fingerprint deserves description. This study examines the profile, guidance behavior, language lifts, and subsequent-return statistics of those calls.
2Data & methodology
The corpus comprises 165,182 earnings-call transcripts published between 1990 and 2026, each scored independently by a large language model on an identical 37-field battery: seven categorical business verdicts, eight 0–9 behavioral meters, and twenty yes/no judgments. The study group is defined as calls where the model answered YES to "Volume About to Step Up" AND backlog was read as growing (n = 23,140; 14.0% of the reference set, 95% Wilson interval 13.8%–14.2%). Baseline figures use all scored calls. Market outcomes join a fixed sample of 22,449 calls with twelve-month total returns in excess of SPY, measured from the first close after each call; this sample skews toward liquid U.S. names and is reported as descriptive history only.
3Results
Calls in this set run higher on promotion (5.60 vs 5.05) and confidence (7.63 vs 7.21) than the corpus baseline, with lower stress (2.21 vs 2.43) and lower evasion (2.58 vs 2.70). They lean optimistic on guidance: 32.36% raised versus 21.05% in the base, while only 7.52% lowered versus 11.56%. Language lifts are led by "Deferred Revenue Growing" at 2.17x and "Scale-Dependent Advantage Claims" at 1.42x; "When the CFO Dominates" appears underrepresented at 0.64x. The pattern peaks in 2021 at 21.72% of calls. Subsequent returns show a median of -7.76% versus -7.16% in the base—a descriptive gap, not an edge.
| Meter | Study group | Baseline | Δ |
|---|---|---|---|
| Candor | 6.83 | 6.86 | -0.04 |
| Evasion | 2.58 | 2.70 | -0.11 |
| Specificity | 7.66 | 7.56 | +0.10 |
| Stress | 2.21 | 2.43 | -0.22 |
| Promotion | 5.60 | 5.05 | +0.55 |
| Confidence | 7.63 | 7.21 | +0.42 |
| Action | Study group | Baseline |
|---|---|---|
| Raised | 32.4% | 21.1% |
| Maintained | 46.0% | 48.8% |
| Lowered | 7.5% | 11.6% |
| Withdrawn | 1.7% | 2.7% |
| Signal | Lift | In group | Baseline |
|---|---|---|---|
| Deferred Revenue Growing | 2.17× | 19.2% | 8.9% |
| Early Products Growing Fast | 1.56× | 60.0% | 38.5% |
| Underused Fixed Costs | 1.44× | 60.1% | 41.6% |
| Scale-Dependent Advantage Claims | 1.42× | 15.7% | 11.1% |
| A Tiny Fraction of the Market | 1.36× | 40.8% | 30.0% |
| When the CFO Dominates | 0.64× | 9.1% | 14.1% |
| The Finished-Story Tell | 0.69× | 3.0% | 4.4% |
| Statistic | Study group | Returns sample |
|---|---|---|
| Median excess return | -7.8% | -7.2% |
| Interquartile range | -28.0% to +13.0% | — |
| Share beating SPY | 39.5% (95% CI 38%–41%) | 39.5% |
| Observations | 2,598 | 22,449 |
| Ticker | Quarter | Call date | Call grade |
|---|---|---|---|
| FLG | Q2 2025 | 2025-07-25 | B |
| FRST | Q2 2025 | 2025-07-25 | A |
| FFBC | Q2 2025 | 2025-07-25 | B+ |
| DBOEY | Q2 2025 | 2025-07-25 | B+ |
| SSB | Q2 2025 | 2025-07-25 | B+ |
| STEL | Q2 2025 | 2025-07-25 | B |
| BAH | Q1 2026 | 2025-07-25 | C+ |
| VLOUF | Q2 2025 | 2025-07-25 | C |
4Discussion
A careful reader should conclude that this language pattern is common, tonally distinctive, and associated with more guidance raises. They should not conclude that it causes better outcomes or predicts them. The beat rate (39.53% vs 39.47%) and median returns (-7.76% vs -7.16%) are statistically indistinguishable in practical terms from the base. Every figure here is a description of co-occurrence in transcript data, not a trading signal, and the AI-read fields that define the pattern carry their own measurement error.
5Limitations
All fields—the YES/NO volume answer, backlog reading, and behavioral scores—are AI-generated and noisy; misclassification attenuates every comparison. The returns sample covers only 2,598 of 23,140 calls within a base of 22,449 calls skewed toward liquid names, so return statistics may not generalize. Our own forward tests falsified directional prediction on this pattern. Additionally, LLMs partially remember famous stocks' histories, which can contaminate any backtest of transcript-derived signals, including the comparisons reported here. See the full methodology, including the C1 pattern’s forward-test failure and the LLM-memorization finding.