A million AI voice calls in India say the dialer, not the voice, is the bottleneck
Pickup starts at 48% on fresh numbers and decays toward 20%. Only retry scheduling gets past 70% cumulative connect, and the arithmetic says that costs about three times the dials.
The Engineer · Build desk
What happened
- Fresh outbound numbers picked up at 48% on the first attempt across their first 1,000 leads.
- Timed retry sequences took some campaigns past 70% cumulative connectivity.
- Once a call was answered, fewer than 3% of conversations ended in a hangup initiated by the person on the line.
Why it matters
- cost The connect target is met with dials, and holding 70% on aged numbers costs roughly triple the attempts per lead, which is telephony minutes and concurrency headroom rather than model spend.
- decision With unanswered calls outnumbering human hangups by more than 130 to 1, the next engineering hour is worth more in the retry scheduler than in the voice stack.
- constraint An SLO written against median latency cannot see the failure this data describes, because the pause that breaks a conversation lives in the tail.
- exposure Anyone sizing a campaign on the 70% number is planning against an unnamed subset of one vendor's campaigns, with no denominator disclosed.
The retry arithmetic is the part of this worth doing by hand. Treat each dial as an independent draw at the fresh-number rate of 48% [8], and two attempts put you at 73% cumulative contact [1]. Do the same at the aged rate of 20% [9], and five attempts reach only 67%; you need six to clear 70% [2]. Same contact target, three times the dial budget [3], and every one of those dials is telephony spend, concurrency capacity and scheduler logic rather than model quality.
Six is a floor, not an estimate. Independence is a generous assumption here: a number that does not answer at 11am on Tuesday is disproportionately a number that will not answer at 11am on Wednesday. The source makes the same point in softer language, saying a retry is not simply another dial and that its timing, frequency and relation to previous attempts affect the odds of eventually reaching the person [14]. Which means the six attempts only work if they land in different windows. Stack them in one slot and you buy the dials without buying the contacts.
Set that against what the dataset says is not broken. User-initiated drop-off came in under 3% [10]. On an aged number at 20% pickup, 100 dials produce 20 conversations and fewer than one hangup by a human who did not want to talk to a machine, against 80 calls that never connected at all: the unanswered outnumber the offended by more than 130 to 1 [4]. That ratio is the ceiling on what a more human-sounding voice can win you. It is not zero, and it is not where the losses are.
Latency is the one place the engineering advice holds without arithmetic. Median response was under a second, with p95 at roughly 2.1 seconds [11], so the tail runs at more than twice the typical case [5]. The listed causes of those spikes include uncommon phrasing, extra network hops and peak concurrency [12]. Note that the last one is downstream of the retry finding: a scheduler that pushes six attempts per lead into the same evening window is manufacturing its own concurrency peak, and paying for it in the tail where conversations break.
The provenance deserves stating plainly. These are one vendor's own production numbers from the DialNexa platform, across more than a million AI-assisted business calls in India [7]. The 20% figure describes "some call categories" [9] and the 70% describes "some campaigns" [13], with no denominators for either subset and no mechanism offered for why aged numbers decay. The 48-to-20 fall is 28 points, a 58% relative decline in first-attempt pickup [6], and a curve that steep almost certainly has a carrier-side cause the post does not name. Read the shape, not the constants: first-attempt pickup is a depleting asset, and the recovery lives in the schedule.
What to watch
- Whether DialNexa publishes per-attempt connect curves and denominators for the campaigns that reached 70%, which would turn an existence proof into a planning baseline.
- Any independent dataset on outbound number decay, and whether carrier-side spam labelling accounts for the 48-to-20 slide.
- Whether the sub-3% hangup rate survives outside structured qualification and event-reminder calls, which were the dataset's strongest use cases.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence26
- Adoption34
- Hype gap+24
- Incentives82
- Confidence47
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
At a 48% per-attempt pickup rate treated as independent draws, two attempts give 72.96% cumulative contact.
- [2]
At a 20% per-attempt pickup rate treated as independent draws, five attempts give 67.2% cumulative contact and six give 73.8%, so six attempts are the minimum to clear 70%.
- [3]
Reaching the same 70% cumulative contact target at the aged 20% rate rather than the fresh 48% rate requires about three times as many dials per lead.
- [4]
On 100 dials to an aged number at 20% pickup, 20 calls are answered, fewer than 0.6 end in a user-initiated hangup and 80 never connect, a ratio of more than 130 unanswered calls per user-initiated drop.
- [5]
The 95th percentile latency is more than 2.1 times the median, since the median is below one second and p95 is about 2.1 seconds.
- [6]
The fall from 48% to 20% first-attempt pickup is 28 percentage points, a 58% relative decline.
- [7]
The analysis covers more than one million AI-assisted business calls across India, drawn from production traffic on the DialNexa platform.
- [8]
New outbound numbers achieved a 48% first-attempt pickup rate across the first 1,000 leads.
- [9]
As numbers accumulated more calls, pickup rates declined, with some call categories falling toward 20%.
- [10]
User-initiated drop-off in the dataset was under 3%; once people answered, most did not immediately hang up.
- [11]
Median response latency in the dataset was under one second and the 95th percentile was approximately 2.1 seconds.
- [12]
Contributors to latency spikes included uncommon phrasing, additional network hops and peak concurrency.
- [13]
With appropriately timed retry sequences, some campaigns achieved more than 70% cumulative connectivity.
- [14]
The post argues a retry is not simply another dial, and that its timing, frequency and relationship to previous attempts can directly affect the probability of eventually reaching the person; it advises optimising for the entire contact cycle rather than the first call.
- [15]
The strongest use case in the dataset was pre-sales lead qualification, followed by webinar and event attendance, both of which have a single-sentence definition of success.
- [16]
English, Hindi and Hinglish were among the most consistent languages in the dataset, and speech-to-speech approaches handled mixed-language conversations more naturally than some cascade pipelines.
Get this ledger inside ChatGPT. Ask about any of our stories and the answer brings its figure, each claim’s verdict and the sources. Free, no account.
Sources
1 independent publisher whose own reporting we read for this story.
- dev.toWe Analyzed 1 Million AI Voice Calls in India. Here’s What Actually Makes Them Work
1 article · August 26, 2026
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.