How Many Times Should You Call a Lead? We Measured 7.69 Million Dials

The short answer: four to nine attempts

Nearly every article on call attempts cites the same handful of vendor surveys, and almost none of them show their data. This one does.

The numbers below come from 7,690,451 outbound dial records across 5,960,061 leads in a production VICIdial installation. They are first-party: our own dialer, our own logs, aggregated across campaigns. No client names, no campaign identifiers, no individual records.

The headline is that the attempt band almost everyone abandons is the one that pays:

Attempts made	Share of bookings	Share of the lead file	Return
1–3	47.4%	44.6%	1.06x
4–9	28.4%	13.6%	2.09x
10–19	16.2%	10.4%	1.56x
20+	7.9%	11.7%	0.68x

“Return” is the share of bookings a band produces divided by the share of the file it occupies. Above 1.0 the band pays for itself; below 1.0 it does not.

Two things jump out. Attempts 4–9 do double duty — they carry 28.4% of all bookings on 13.6% of the file. And the 1–3 band, where nearly half of all leads sit, is essentially break-even at 1.06x. Most operations are concentrating their effort in the least efficient part of the curve.

The funnel nobody publishes

Before the attempt curve makes sense, the base rates have to be on the table. Here is what 7.69 million dials actually produced:

Stage	Count	Rate
Dials placed	7,690,451	—
Reached a human	312,746	4.1% of dials
Booked an appointment	6,570	2.1% of conversations

That is one human conversation per 24 dials, and one booked appointment per 1,170 dials.

If those numbers look brutal, they are — and they are the honest denominator that most “connect rate” benchmarks quietly exclude. A “35% contact rate” almost always means 35% of answered calls, not 35% of dials.

“Reached a human” here is not a guess. VICIdial carries a human_answered flag on each disposition status, and the figure above counts only statuses where that flag is set. This matters more than it sounds: the status code A is Answering Machine, not a human answer, and counting it as a conversation inflates the number by roughly 120,000.

Where the dials actually go

The single largest fact in the dataset is not about attempts at all:

Outcome	Dials	Share
Answering machine (auto-detected)	6,163,727	80.1%
No answer	730,603	9.5%
Human answered	312,746	4.1%
Disconnected / invalid number	147,516	1.9%

Four out of five outbound dials hit an answering machine. Not “most” — 80.1%.

This reframes the attempt question entirely. When you dial a lead three times and stop, you have most likely never spoken to that person; you have left three voicemails. The 4–9 band pays because it is the range in which you finally catch someone live, not because persistence wears people down.

It also reframes what a dialer is worth. If 80% of your dials terminate in a machine, the accuracy of your answering-machine detection and how fast it releases the agent is not a minor setting — it is the majority of your call volume.

Where to stop

The 20+ band returns 0.68x — below break-even. On its face, that says stop at twenty.

We are going to argue against our own number here, because it deserves a caveat.

The attempt count for a lead freezes when that lead books, but keeps climbing for leads that never book. So the 20+ band is enriched with permanent non-converters by construction: a lead that booked on attempt 5 is recorded at 5 attempts forever, while a lead that will never book keeps accruing dials until someone stops calling it. Some of the decline past 20 is therefore an artifact of how the counter works, not purely a fact about diminishing returns.

What survives that caveat is the comparison between bands that are all subject to the same bias, and there the 4–9 result is the robust one: it outperforms the 1–3 band by roughly 2:1 despite the bias running in the opposite direction (booked leads are systematically pushed toward lower attempt counts, which if anything understates the 4–9 lift).

So the defensible reading is: the case for attempts 4 through 9 is strong; the case for a hard stop at exactly 20 is directional. Treat 20 as the point where you should be actively re-qualifying the lead rather than dialling it again.

What this means for how you set up a campaign

Four practical consequences, in order of how much they change:

1. Set your minimum attempt count to at least four. If your dial plan retires a lead after three tries, you are cutting it off immediately before the band that produces the best return. This is a configuration change, not a strategy change.

2. Budget for the answering-machine rate, not the connect rate. At 80.1% machine and 4.1% human, an agent’s talk time is a small fraction of the campaign’s call volume. Staffing and cost-per-lead models built on connect rate rather than dial volume will be wrong by an order of magnitude.

3. Space the attempts. This dataset does not measure timing, so we are not going to claim an optimal interval — but nine attempts compressed into two days is a different thing from nine attempts across three weeks, and the 4–9 result should not be read as licence for the former.

4. Re-qualify at twenty, do not just keep dialling. Whatever the exact shape of the tail, nothing in this data supports unlimited persistence.

Method, and what would change these numbers

• Source: production VICIdial vicidial_log and vicidial_list tables, read through a SELECT-only proxy. Aggregate only.
• Corpus: 7,690,451 dial records, 5,960,061 leads, all campaigns, lifetime.
• Bookings: 6,570 leads reaching the APPTBK (appointment booked) disposition. Note that VICIdial’s built-in sale flag is not a reliable booking marker in a real installation — in this one it matches a single administrative reset status and would return a meaningless answer.
• Human contact: dispositions where vicidial_statuses.human_answered is set. 30,289 dials (0.4%) carry a status not present in the status table and are excluded from the human-contact rate.
• Attempt bands: the lead’s recorded call count, banded. See the caveat above on how this counter behaves for booked versus unbooked leads.

What would change these numbers: a different vertical, a different list source, or a different dial ratio would all move the base rates. The attempt-band shape is the part we would expect to generalise; the absolute connect and booking rates are specific to this mix of campaigns.

If you are running outbound and your own numbers disagree with these, the most common reasons are that your connect rate is measured against answered calls rather than dials, or that your booking status is not the one you think it is. Both are worth checking before concluding your operation is an outlier.

---

Running an outbound operation and want the same analysis on your own dialer data? Talk to us about a call center audit — or read our cost per lead benchmarks for the economics that sit on top of these numbers.