Twelve weeks of outbound dialling, one outcome per call, measured in the prospect’s own timezone. We tried to break each finding before publishing it, and we show you the two that partly broke.
Every number on this page comes from one dataset, frozen at a stated cutoff and reconciled in full below. Where the data is thin, or a result looks like an artefact of how we work rather than a fact about cold calling, we say so.
The single largest effect we found, and the one that held up to every test we ran. Connect rate declines from 8am through the afternoon, with a clear recovery over lunch and a weaker one at 5pm. Every dial is placed in the prospect’s local hour, derived from area code, not the caller’s clock. Most published versions of this statistic never say whose clock they used.
The obvious objection: dialling floors start fresh lists in the morning, and first dials answer far better than repeat dials. So is 8am really the first-dial effect wearing an hour-of-day costume?
We stratified by attempt number to find out. Within first dials only, 8am returns 38.4% against 10.9% at 4pm, a factor of 3.5. The effect holds independently inside repeat dials too, at 17.6% against 4.0%. It also survives per caller: six of our nine callers had meaningful volume in both windows, and five of those six reach more people in the morning.
The hour effect is not the attempt effect.
The 1pm recovery is the second most interesting bar on the chart. Connect rate falls through the late morning, jumps back to 21.1% over lunch, then collapses through the afternoon to a low of 7.8% at 4pm. If you dial a single hour a day, dial 8am. If you dial two, add 1pm, not 4pm.
Across all dials, reaching a human falls from 22.1% on the first attempt to 10.8% on the second and 5.2% on the third. That looks like a clean case against persistence, and taken at face value it would be wrong, so we checked it before believing it.
Numbers that answer leave the redial pool. Numbers that never answer accumulate attempts. So the population at attempt five is, by construction, the population that has already proved hard to reach. The curve partly measures who is left, not what persistence does.
To separate them we fixed the cohort: only numbers that received six or more dials, where every number appears at every attempt and survivorship cannot operate. Inside that cohort the shape changes.
Two things survive. The first dial really is the best dial: 5.58% against 1.46%, a 3.8× advantage, on identical numbers with no survivorship available to explain it. And attempts two and three really are a dead zone, bottoming at 0.70%.
But the monotonic collapse does not survive. Inside the fixed cohort the curve turns back up from the fourth attempt: 1.14%, then 1.33%, then 3.36% at the sixth, close to three times the average rate across attempts two to five. Persistence does not fail. It stops working for a while and then starts again.
The operationally useful version, and the one we would stand behind: call once at the right hour, then either move on or commit properly. In the fixed cohort, the half-hearted second and third attempt is the worst use of a dial.
Day of week produced a spread nearly as wide as time of day. Because we already suspected Monday was an artefact of our own scheduling, we tested that too.
Our hypothesis was that Monday looks bad because we dial it later: list preparation and weekly planning push Monday calling into worse hours. The data supports the premise. Monday’s mean local dial hour is 12.6 against 12.1 on Wednesday, and only 20.2% of Monday dials land between 8am and 10am against 26.2% on Wednesday.
But it does not explain the gap. Holding both days to the same 9am to 2pm local window, Monday returns 6.3% and Wednesday 27.3%. The hour mix accounts for very little of it. Monday is worse for reasons that are not hour mix, and that this dataset cannot identify. Note that the controlled gap, 4.3 times, is wider than the headline 3.9.
Stated in enough detail to be checked or reproduced.
| Source | Alba Talent production dialler records |
|---|---|
| Window | 18 May 2026 to 11 August 2026, 86 days inclusive |
| Records in window | 109,049 call attempts, one outcome each |
| Analysed | 98,518 dials after exclusions |
| Reached a human | 16,429 (16.7%) |
| Distinct numbers | 60,054 |
| Callers | 9 |
| Market | United States, business-to-business |
The window is frozen deliberately. The dialler is live and the underlying table grows daily, so the study is cut at the last complete day before publication. Anyone re-running these queries later will see larger totals; the figures here describe the frozen window only.
“Reached a human” means the call ended in an outcome recording a person on the line: a conversation, a gatekeeper, a wrong contact, a refusal. It is deliberately broader than “a good call”. It measures whether the phone was answered, not whether the pitch worked.
Exclusions reconcile exactly, and the chain starts from every record we hold. At the moment of the freeze the dialler table held 132,570 attempts. 22,316 carried no usable timestamp and cannot be assigned to any window, so they are removed first; a further 1,205 fall outside the window. That leaves 109,049 in-window records, from which 9,911 marked cancelled and 620 with no recorded outcome are excluded, giving 98,518. Note that only 1,205 timestamped records fall outside the window at all: this is very nearly our entire dialling history to date, not a period selected after seeing the results. Cancelled records are dials abandoned before connection rather than attempts: they average 0.8 seconds. Including them would cut every rate by about a tenth, from 16.7% to 15.1%, without describing anything real.
Attempt numbering is within-window. A number first dialled before 18 May appears here as a first dial, because the sequence is computed inside the frozen window rather than over the table’s full history. Distinct numbers and first dials are therefore equal by construction, both 60,054.
Timezones. Local hour is derived from the North American area code and a fixed UTC offset per zone. The window sits entirely within daylight saving, so offsets are constant throughout. 91,477 dials mapped, 92.9%, using a 284 area-code table. The remainder are excluded from the time-of-day chart only.
Free to quote or reproduce with attribution. Suggested form:
Alba Talent, “The 98,518 Dial Study” (August 2026). Analysis of 98,518 US B2B cold calls placed 18 May to 11 August 2026.
If you replicate any of this on your own dialler data, particularly the 8am result or the Monday gap, we would like to hear what you found, whether or not it agrees.