Queue time costs you CSAT twice. Once at second six, again at sixty.
A correlation study of 750,000 inbound calls across nine industry verticals, testing whether queue time drives customer frustration. It does, and it does so twice: once at second six, again past sixty.
What the data says, in plain terms.
One hypothesis: does queue time cost you CSAT?
Three months of inbound calls, the first half of 2026, grouped into four Average Speed of Answer tiers and scored for sentiment as a CSAT proxy. At this scale the pattern is not a sample; it is the population.
Queue time is a CSAT variable you can measure on the call itself, without a survey. Two things fall out of the data.
1. The cliff is at zero, not at sixty. The single largest movement anywhere in the dataset is the step from an immediate answer to a short wait: negative sentiment climbs 14.8 points and positive sentiment falls the same 14.8 points, in one step. Every service-level target built around a 20 or 30 second threshold is measuring the wrong edge.
2. There are two failure modes, not one. Frustration spikes in the 6 to 19 second tier (uncertainty), recovers 18.2 points at 20 to 59 seconds as customers recalibrate, then slides again past 60 seconds (endurance). Recalibration buys tolerance, not immunity, so the short unexplained wait and the long tail are separate problems with separate fixes.
The practical read: a single service-level threshold set at 20 or 30 seconds manages neither mode well. Eliminate or explain the short wait, and cap the long tail through capacity and routing.
Four findings, one threshold.
Waiting costs CSAT, and the bill arrives in the first few seconds.
The core hypothesis holds: waiting costs CSAT. But the data does not support a simple longer-wait, worse-sentiment straight line. The damage is concentrated early, in the first 6 to 19 seconds of queue time, which is exactly where waiting-psychology research would predict it.
Sentiment splits cleanly along ASA tiers.
Two things stand out: the crossover point at five seconds, and the size of the swing immediately after it.
| ASA tier | Negative | Positive | Calls | % Positive | % Negative | CSAT (net) |
|---|---|---|---|---|---|---|
| 0-5s (Immediate) | 251,508 | 279,099 | 530,607 | 52.6% | 47.4% | +5.2% |
| 6-19s (Short wait) | 80,726 | 49,058 | 129,784 | 37.8% | 62.2% | -24.4% |
| 20-59s (Moderate wait) | 11,325 | 10,003 | 21,328 | 46.9% | 53.1% | -6.2% |
| 60s+ (Long wait) | 37,691 | 30,590 | 68,281 | 44.8% | 55.2% | -10.4% |
Calls = negative + positive. CSAT (net sentiment) = % positive minus % negative. Source: inbound calls, 2026 H1.
Immediate answers score 11.4 points better on net sentiment than the next-best tier, 15.6 points better than the long-wait tier, and 29.6 points better than the worst. Every tier that involves waiting scores worse than the tier that does not. That is the headline proof, and it is the clearest signal in three months of data.
The shape of the curve is a behavioral result.
A sharp early drop rather than a smooth decline is exactly what decades of service-operations research on waiting would predict. Four principles apply directly.
None of this contradicts the hypothesis; it explains it. Queue time does drive frustration and negative sentiment. The behavioral literature predicts precisely this shape: a spike in the earliest, most uncertain seconds, then partial stabilization as customers adjust. The implication is not that waits stop mattering past 20 seconds; the 60 second plus tier turns down again, 4.2 points below the moderate tier. Recalibration has a ceiling, and both ends of the curve need managing.
How many of your calls are in the danger tier?
Enter your own monthly inbound volume and the share of it that waits between 6 and 19 seconds. The study's measured sentiment rates do the rest.
Defaults reflect this study's distribution: 70.7% immediate, 17.3% short wait. Adjust to your own queue mix.
Get the complete report, free.
The full PDF carries the complete tier tables, the behavioral analysis, the correlation tests, and the methodology behind every figure on this page.
Email me the study
One email, with the PDF attached. No sequence, no follow-up cadence.
The study is headed to your inbox. You can also grab it directly right now:
Download the PDFHow the study was built.
Population-scale observational analysis on production scoring data. Every figure is computed on the full call set.
The dataset covers 750,000 inbound calls handled across nine industry verticals over three months, the first half of 2026. Each call carries a positive or negative sentiment score derived from speech analytics, used throughout as a proxy for CSAT in the absence of a post-call survey instrument. Calls were grouped by queue time at the point of connection to an agent, into four tiers consistent with prior passes of this study: 0 to 5 seconds (immediate), 6 to 19 seconds (short wait), 20 to 59 seconds (moderate wait), and 60 seconds or more (long wait). Each tier's net sentiment is % positive minus % negative, using mutually exclusive counts that sum to the tier's total volume.
| Step | Result |
|---|---|
| Negative-sentiment calls | 80,726 |
| Positive-sentiment calls | 49,058 |
| Total calls in the 6 to 19 second tier | 129,784 |
| % Positive (49,058 / 129,784) | 37.8% |
| % Negative (80,726 / 129,784) | 62.2% |
| Net sentiment (37.8% minus 62.2%) | -24.4% |
Queue-time context and sentiment arcs are Layer 1 scoring signals. Read against staffing and queue design they feed Operational Intelligence (Layer 4), and read against retention risk they feed Customer Intelligence (Layer 2). See the Six Layers of Intelligence.
Produced by the QEval® Speech Analytics team from production scoring data. Sentiment scoring and queue-time classification were performed by the QEval® platform, built by ETS Labs, an Etech Global Services company.