Almost everything written about cold calling benchmarks is either a vendor estimate or a survey of what people believe they do. We had the option of checking, so we did.
Between 18 May and 7 September 2026, callers on our floor handled 123,426 calls on live real estate campaigns, motivated seller lists, US markets, working the afternoon and evening. What follows is what those calls actually did. Where a number is easy to misread, we say so rather than quoting the flattering version.
Read the methodology before quoting anything
This matters more than usual here, because the most commonly quoted cold calling statistics are wrong in a specific and predictable way.
- The population is 123,426 calls that reached a live agent between 18 May and 7 September 2026, across 31 callers and multiple US real estate campaigns
- These are agent-handled calls, not raw dials. On a predictive dialer the system connects a caller only once the phone is picked up, so the underlying dial count is far larger and is not the denominator used here
- For that reason we do not publish a "contact rate" from this data. A ratio computed on an already-filtered population would look extraordinary and mean nothing, which is exactly how vendors end up advertising contact rates of 80% and higher
- Where we say "answered", we mean a disposition indicating a human being spoke to the caller: connected, not interested, callback scheduled, appointment set, sale, do-not-call request, or complaint
- Wrong numbers are counted separately. Somebody answered, but not the person on the list, which is a data problem rather than a conversation
The median cold call lasts 22 seconds
Across 57,618 answered calls, the median talk time was 22 seconds and the mean 28. The 90th percentile was 44 seconds. Only 1,077 calls, 1.9%, ran past two minutes.
That is the single most useful number here, and it reframes what a caller is actually doing. The job is not persuasion. Ninety-eight times in a hundred the conversation is over inside two minutes, and the entire skill is what happens in the first fifteen seconds: confirming you have the right person, naming the property, and asking one answerable question.
It also explains why training that focuses on objection scripts produces so little. There is rarely a long enough conversation for the script to matter. The calls that convert are the small minority that get past the first half-minute, and what gets them there is the opening.
It takes about 99 conversations to book one appointment
Of the 57,633 calls where a human spoke to a caller, 580 produced a booked appointment. That is one appointment per 99.4 conversations, and remember, that is per conversation, not per dial. The dial figure is several times larger again.
A caller having twenty real conversations in a day is therefore having a normal day and booking nothing, and will book roughly one appointment a week. Any staffing plan or commission structure built on a caller producing an appointment a day is built on a number this data does not support.
It is also why paying callers purely per appointment corrupts the pipeline. When the honest rate is one in ninety-nine, the pressure to relabel a lukewarm conversation as an appointment is enormous, and the resulting calendar converts at nothing.
One in six answered calls is the wrong person
Wrong numbers accounted for 16.19% of agent-handled calls, 19,981 of them. Roughly one in every six times a caller got a human on the line, it was not the person on the list.
That is a data cost, not a calling cost, and it is invisible in every report that only measures connects and appointments. In caller time it is substantial: at a median of 22 seconds a call, those 19,981 conversations are over 120 hours of paid calling spent reaching people who were never prospects.
The fix is upstream: better skip tracing and more aggressive list hygiene, and it produces a bigger improvement than any amount of script coaching, because it works on the first day rather than the fourth week.
What actually happens on an answered call
The full distribution across agent-handled calls:
- Not interested: 24.12%
- No answer: 18.29%
- Wrong number: 16.19%
- Connected and qualified: 14.79%
- Voicemail: 9.48%
- Failed or dropped: 9.35%
- Do not call request: 6.81%
- Appointment set: 0.47%
- Callback scheduled: 0.46%
- Sale: 0.04%
The number worth sitting with is the 6.81% asking not to be called again. On this volume that is 8,403 suppression requests in under four months. An operation without a reliable one-press suppression path, applying across every list, is generating a compliance problem at that rate whether or not anybody is tracking it.
Evening calls book more appointments, but not because more people answer
This is the finding that most contradicts the usual advice, and it only shows up if you separate the two questions.
The proportion of calls where somebody answered barely moved across the working day, staying between 45.7% and 47.8% from midday to 8pm Eastern. By that measure every hour looks the same.
The appointment rate did not:
- 12:00: 0.460% of calls booked an appointment
- 13:00: 0.414%
- 14:00: 0.420%
- 15:00: 0.381% (the low point of the day)
- 16:00: 0.530%
- 17:00: 0.513%
- 18:00: 0.552% (the high point)
- 19:00: 0.494%
The 6pm hour books appointments at roughly 45% above the 3pm hour, on an identical answer rate. Same number of conversations, materially different outcomes.
The likely explanation is who is available rather than how many: an owner-occupier home from work in the evening is a different prospect from whoever is reachable at three in the afternoon. It is not a scheduling trick. It is that the mid-afternoon population answering the phone contains fewer decision-makers.
The practical consequence is that an operation calling 9am to 5pm is spending its best hours on its worst window and stopping before its best one starts.
Day of the week barely matters
Contact rate ran 45.68% to 47.52% Monday through Friday, with Monday marginally highest and Thursday marginally lowest. A spread of under two points across the week is not something to build a strategy on.
We have too little weekend volume to publish a figure, and we would rather say that than quote a number computed on a handful of calls.
What we would change on the strength of this
- Judge callers on conversations, not appointments, at one in ninety-nine, appointment counts are too noisy to manage a person by week to week
- Spend on data before spending on coaching, while one in six answered calls is the wrong person
- Move the shift later. The 15:00 hour is the worst of the day for appointments and 18:00 the best, and most operations have gone home by then
- Train the first fifteen seconds, not the objection tree. Ninety-eight percent of these conversations end inside two minutes
- Give every caller one-press suppression. At 6.81% of answered calls, do-not-call handling is not an edge case, it is a daily volume
None of this is universal. It is one agency, four months, US real estate seller lists, worked afternoons and evenings. Your list quality and market will move every figure here. But it is measured rather than estimated, which is more than can be said for most of the benchmarks in circulation, and if it disagrees with a number you have been quoting, it is worth asking where that number came from.