Why applicants go cold, and what a same-hour response changes.

Applications arrive in the evening and at the weekend, because that is when people look for work. Responses go out in office hours, because that is when recruiters exist. The gap between those two facts is where the pipeline drains. By the time a candidate is called on Thursday afternoon about a Tuesday night application, they have applied to four other places and heard back from one of them, and the advertising that produced them has bought nothing.

The original version of this guide said that from experience. This version sets out what the published research says, because the pattern is measurable, and then what it costs to fix. Two things have changed since 2024 that make it worse: applications per job have roughly doubled, and the number of recruiters per application has fallen further still. Both are in the benchmark data below.

Robert Half’s survey of jobseekers, with 1,000 UK respondents among 9,000 across several countries, found that 52 per cent named slow feedback about where they stand as their biggest frustration, and that 36 per cent were willing to withdraw an application if they did not receive a timely response. Its earlier survey of more than 200 UK HR directors found they spent 27.59 days on average recruiting for an open position, and that 54 per cent had lost a qualified candidate to another opportunity because their own process was too lengthy. Its November 2025 fieldwork with 500 UK hiring managers found 65 per cent now say recruitment delays pose the greatest risk to business continuity.

Indeed’s ghosting study of more than 4,000 jobseekers and nearly 900 employers found 83 per cent of employers had been ghosted, while only 18 per cent of jobseekers admitted to it. Of those who had, 40 per cent did so after receiving another offer, 46 per cent stopped responding to calls and emails, and 50 per cent skipped a scheduled interview. That survey is American and from 2019, and we cite it because nothing UK-specific of the same scale has been published since.

The nearest UK evidence is the CIPD’s 2024 resourcing survey of 1,016 HR professionals, which found 27 per cent said selected candidates always, mostly or sometimes failed to turn up on their first day, 27 per cent reported dropouts or counteroffers after acceptance, and 41 per cent said new recruits resigned within the first twelve weeks at least sometimes. The same survey found 31 per cent of organisations using AI or machine learning somewhere in recruitment, up from 16 per cent in 2022, and 66 per cent of those said it had improved hiring efficiency.

Read together, the pattern is consistent across a decade and two countries: candidates leave processes that go quiet, and they leave for whoever answered first. Speed is not a nicety in high-volume hiring; it is the conversion mechanism, and the surveys that measure it keep finding the same third of candidates who will not wait. That third is the advertising budget leaving by the back door, and it leaves before anyone in the office has seen the application.

Greenhouse’s 2026 benchmark, drawn from more than 800 European companies and 17 million applications between 2022 and 2025, shows applications per job in Europe rising from 99 to 183, an 85 per cent increase, while time to fill lengthened from 44.39 to 62.24 days. Applications per recruiter went from 98 to 474. The global figures from more than 6,000 companies are steeper: 116 to 244 applications per job, and 146 to 746 per recruiter. Workable’s network data for September 2024 put candidates per hire at 181.2 and time to fill at 42.8 days, and its UK and Ireland benchmark for qualified candidates per hire runs from 16 in customer service to 25 in marketing.

The UK labour market explains part of it. ONS early estimates for May to July 2026 put vacancies at 707,000, the lowest outside the pandemic since late 2014, and the REC and KPMG Report on Jobs for August 2026 recorded the sharpest rise in people seeking new employment for three months, with permanent candidate supply rising faster than temporary. More people are applying for fewer jobs, and each one is waiting behind more others.

The arithmetic for a recruiter is unforgiving. If one person handles 474 applications a year in Europe’s benchmark, and it takes ten minutes to read and respond to each, that is 79 hours of first responses alone, before any conversation. Nobody does that within the hour by hand, and the CIPD found only 21 per cent of organisations even collect time-to-hire data, let alone time to first response.

It is intuitive that more people means faster response, and at low volume it is true. At high volume it stops being true, because the constraint is not how hard the team works. It is that the hours when applications arrive and the hours when the team is available barely overlap. Doubling the team doubles cost and improves the overlap not at all. That is why offices that scale headcount with volume find their conversion rate flat and their cost per hire rising.

The benchmark data makes the same point from the other side. Applications per recruiter in Europe went up nearly fivefold in three years. No agency hired five times the recruiters. The gap was absorbed by silence, which is the thing candidates say they will not tolerate. The only way to close it is to make the first response independent of who is awake, which means a system rather than a rota, and to give the recruiters back the hours for the conversations that need a person.

A same-hour response is not a thank-you email. It is four things happening before the candidate has closed the job board, and the order matters: the acknowledgement buys a few hours of patience, the screening decides who deserves a recruiter, and the interview offer is what converts.

None of this requires the candidate to talk to a machine if they would rather not. The screening can be a short chat on their phone at 11pm, a voice call if they answer, or a form if they prefer; what matters is that the process moved while the recruiter was asleep, and that the recruiter arrives in the morning to a ranked list rather than an inbox. The four steps are these.

We build and run Onrolo, a recruitment platform of our own, and we let its AI voice screening call real applicants over a ten-week production window, with consent gathered first and the agent stating plainly that it was an AI. The outcomes are published in full in our guide on AI phone calling, and the shape is the point: an outbound call reached a human about one time in eight, about half went to voicemail, about a quarter never connected, and under one in ten hit a switchboard or a gatekeeper. When a person did answer, the typical screening call ran a little over three minutes and nineteen in twenty of those conversations produced a complete, scored assessment the recruiter could read before deciding whether to book an interview.

The lesson for anyone buying same-hour contact is that the voice is not the product. Seven calls in eight do not reach a person, so the value is decided by what happens in the same minute to the seven: a text with a link to complete the same screening by chat, a record that voicemail was reached so nobody redials in an hour, and retries spaced by hours and capped. A voice agent without a text fallback is an expensive way to reach voicemail.

The other thing production taught us is where the money goes. Language model calls were a single-digit share of the variable spend on AI features; text messages to candidates were the largest line by a wide margin, with telephony second. Any proposal for automated first contact that prices the model and forgets the channels around it will be wrong every month, and the vendor questions in our AI calling guide are the ones to ask.

Prices read on the vendors’ own pages on 14 September 2026. Most applicant tracking systems for agencies price per user; the engagement and screening tools price per role, per hire or per month. Several of the best-known AI screening products publish no price at all.

Vendors quoting in dollars do so on their pricing pages; we have not converted. Vincere and Recruit CRM pages required a proxy fetch to read; treat those two as indicative and check the page.

Take an agency that makes 20 placements a month. At Workable’s network figure of 181.2 candidates per hire, that is about 3,624 applicants a month to acknowledge, screen and sort. At Workable’s UK qualified-candidates benchmark of around 20 per hire, only about 400 of them are worth a recruiter’s conversation. The job of automation is to find those 400 within the hour and keep the other 3,200 informed.

The channel cost is small. At Twilio’s UK rates, an acknowledgement, a screening link and one reminder is three SMS segments per applicant, or $0.168, which is about $609 a month for all 3,624. If a fifth of the qualified group are called by voice at three minutes each, that is about 240 minutes at $0.0305 to mobiles, under $8. The model cost behind the screening is a single-digit share of the variable spend in our own production systems, with messaging the largest line; the point is that the whole automated first contact for 3,600 applicants costs less than a day of a recruiter’s time.

The recruiter time is the larger number. If a human first response, including the dial attempts that go to voicemail seven times in eight, costs fifteen minutes per applicant worth contacting, then screening 400 people a month is 100 hours, and doing it for the wider group before anyone knows who is qualified is far more. Price that at the ONS full-time median of £766.60 a week, taken as £20.44 an hour, and the automated first contact is worth between two and a half and three recruiter-weeks a month, or about £2,000 in gross pay, before the conversion gain from being first. Against a first agent at £4,000 to £15,000, the payback is two to eight months on time alone.

The conversion gain is the number that actually matters and the one we cannot give you from published data. If even a handful of the 36 per cent who say they would withdraw without a timely response stay in the process, at an agency fee per placement the return dwarfs the time saving. Measure it on your own pipeline for a month before and after, which is what the audit does.

Time from application to first meaningful contact is the number to track, measured in minutes rather than days, and split by when the application arrived. The evening and weekend figures are where the problem hides. Then the proportion of applicants who reach a booked appointment, which tells you more about the health of a hiring operation than any measure of recruiter activity. If you buy AI calling, measure conversations per hundred dials and what the conversations produced, never calls made.

Automation belongs on acknowledgement, qualification, scheduling and reminders. Anything requiring judgement, persuasion or reading the room stays with a person, and a supplier who tells you the whole process can be automated has not run one. We run one, and the recruiter still makes the call that matters: the conversation with the candidate who is qualified, interested and about to accept something else.

For transparency on price, since we sell this: a first screening agent, built on your ATS and covering acknowledgement, chat or voice screening, scoring and interview booking, is £4,000 to £15,000 fixed. A custom AI system that adds voice with text fallback, retry rules and reporting is £8,000 to £25,000. An audit first, if you want the numbers on your own pipeline before committing, is £2,500 to £6,000 and is credited against the build. Running costs are dominated by messaging rather than the model and are modelled before you sign.

We read every research page cited above at source on 14 September 2026 and quote the figures as published, with base and date. The Indeed ghosting study is American and from 2019; the CIPD 2024 report is the latest edition on the CIPD site. The ONS vacancies figure is from the August 2026 labour market overview because the vacancies bulletin itself was unavailable on the day. Vendor prices were read on the vendors’ pricing pages; two required a proxy and are flagged in the table. The first-party voice outcomes are those already published on our AI phone calling guide, quoted unchanged. Hours in the worked example are assumptions. Augustova builds and operates Onrolo, a recruitment platform, which is our interest in this subject.