Dial attempt self-check

Why the tool returns a ratio, and never your rate against ours

Every human-answered count behind both sides of this comparison -- yours and ours -- comes from an automated call-progress classifier, not a person listening to the call. On our own estate that classifier was independently audited against audio it did not write, and its HUMAN class was wrong on 28 of 92 bridged calls. That error does not disappear because you run your own numbers through this tool; it is a property of the classifier, and most dialer platforms in this industry run something similar.

What survives that error is the RATIO between two buckets measured the same way, because a roughly constant misclassification rate divides out of a ratio. It does not divide out of a level. So this tool never tells you whether your attempt-1 or attempt-2 rate is higher or lower than ours -- that comparison would hand a critic the exact argument that discredits the whole family of figures, and it is precisely the misuse the underlying dataset exists to prevent. It tells you only how your ratio compares to ours, and it holds only if your own classifier is about as consistent between your two buckets as ours is between ours.

If your dialer platform uses a materially different classifier configuration between your first and second attempt buckets -- a different threshold, a different vendor, a mid-period change -- the ratio comparison above stops being a fair one, for the same reason a level comparison never was.

The SQL shape that pulls your four numbers

This is the same query shape our own generator runs against a VICIdial history replica to produce the attempt curve on our published benchmark: one row per outbound dial attempt, the numerator resolved from the campaign's own status dictionary first and the system dictionary second, grouped by attempt number. No client name and no campaign name belongs in its output.

select l.called_count as attempt,
       count(*) as dials,
       sum(case when coalesce(cs.human_answered, ss.human_answered) = 'Y'
                then 1 else 0 end) as human_answered
from vicidial_log l
left join vicidial_campaign_statuses cs
       on cs.status = l.status and cs.campaign_id = l.campaign_id
left join vicidial_statuses ss
       on ss.status = l.status
where l.call_date >= '<period start>' and l.call_date < '<period end>'
group by l.called_count;

Run it over your own reporting period, then read the attempt = 1 and attempt = 2 rows into the four fields above: attempt-1 dials and human_answered, attempt-2 dials and human_answered. If you query a live dialler directly, use a read-only replica or read-only credentials -- never a path that can write back to the log.

This is not a benchmark for your operation

The contact rate calculator already makes this point for the four ratios it computes: the only benchmark that ever applies to your operation is your own list, dialled the same way, over time. This tool operationalises that same instruction for one specific claim -- which dial attempt reaches a person more often -- rather than overturning it. Watch your own ratio across a few reporting periods before treating any single one as settled.

Our own attempt curve runs the identical spread check across ten attempts and against the hour of day: a factor of 1.994 across attempts against 1.185 across the hours we work. If your own second-dial ratio is close to flat, the more useful question is not "is our second dial better" but whether attempt number moves your answer rate at all -- and that is a question this same query shape, run out to more attempts, can also answer.

Where our numbers come from

6,288,938 outbound dial attempts across 17 campaigns over 9 complete weeks, 2026-05-25 to 2026-07-26 inclusive; 278,651 of them (4.431%) carry a human-answered disposition Source: apps/ccdocs-astro/src/data/dial-cadence-study.json (`universe`), generated by apps/ccdocs-astro/scripts/dial-cadence/dial_cadence_distribution.py from the vicidial_log history replica. The full methodology, the withheld buckets, the per-campaign robustness check and a machine-readable copy of every figure are on the benchmark page itself, published under CC BY 4.0.

Where this fits

• Outbound dial cadence benchmark -- the full study this tool's own comparison numbers come from, including the parts that do not flatter us.
• Contact rate calculator -- connect rate, right-party contact rate and list penetration from one period of your own dialer counts, with the same no-benchmark posture.
• Outbound call center -- the operation that would place the dials this tool checks.