top of page

LATEST INSIGHTS

Expert insights across your SaaS environment

Blogs and papers from the SAASiQ team on enterprise SaaS, cloud and AI.

AI in HR: Workable's Recruiting Agent Screens 1.2 Million Applications, Employers Report Less Trust in CVs, and Workers Replaced by AI Say They Were Asked Back

Writer: SAASiQ.ai
SAASiQ.ai
4 days ago
10 min read

Title: AI in HR: Workable's Recruiting Agent Screens 1.2 Million Applications, Employers Report Less Trust in CVs, and Workers Replaced by AI Say They Were Asked Back

Date: 12 November 2026

Type: Paper

Author: SAASiQ (contact@saasiq.ai)

Word count: 2678 words

Reading time: 10 min

Published: 12-11-2026


Workable said on 10 November that its AI recruiting agent had assessed 1.2 million job applications. Of more than 350,000 candidates the agent went on to contact, 222 asked to speak to a person instead. Built In published a survey of job seekers on 10 November. In it, 88 per cent had used AI to research employers. Of those, 53 per cent had decided not to apply to an employer because of what the AI told them. Challenger, Gray and Christmas reported on 5 November that US employers cited AI for 1,265 job cuts in October, the lowest monthly figure of 2026. Fortune reported on 9 November an SHL survey in which 16 per cent of US and UK workers said they had been replaced by AI and then asked back. This paper sets out how each use works, what is known about cost, and where it breaks.


What happened this week

This is the seventh of SAASiQ's Thursday papers on how AI is used in practice, sector by sector. HR came second on 8 October, after finance. The rotation returned to finance on 5 November, and this week it reaches HR for the second time.


The 8 October paper covered skills data, outside agents reading HR systems, AI interviews and managers using AI to write reviews. This paper picks up three other parts of the work: screening applications at volume, judging whether an application can be trusted, and deciding which jobs AI can take over.


The week from 5 to 11 November brought figures on all three. Most of them come from companies that sell HR software or services, and this paper says so where it matters. None of the surveys below is a representative sample of all employers.


Use case one: an agent that screens every application

Workable sells an applicant tracking system, the software employers use to post jobs and manage candidates. It announced Workable Agent on 13 March 2026. Workable describes the agent as covering sourcing, screening and candidate engagement inside the tracking system. A product listing says it became generally available in July, priced on usage-based credits, with controls to pause, inspect or override what it does.


Workable's release of 10 November gave figures that its chief revenue officer, Sarfraz Ali, presented at the HR Tech conference in Las Vegas, which ran from 20 to 22 October. The agent assessed all 1.2 million applications as they arrived. It picked out more than 350,000 candidates for further contact. Every one of them was offered the option of speaking to a person instead of the agent. 222 took it. Workable gives that as 0.1 per cent; on the figures in the release it is under 0.1 per cent. Nearly 200,000 candidates went on to conversations with recruiters or hiring managers.


Ali said the result should not be read as candidates preferring AI, according to coverage of the release. His point was that the alternative before was not a conversation with a person either.


In practice, an agent of this kind does three jobs a recruiter used to do in sequence. It reads each application against the job's requirements and gives it a score. It sends questions to the candidates it rates highly, by message or chat, and records the answers. Then it passes a shortlist to the recruiter. The recruiter's work moves from reading applications to reading the agent's shortlist and its reasons.


Where screening agents break

The first weak point is the 0.1 per cent itself. A low rate of requests for a person measures what candidates did when offered the choice inside a chat with the agent. It does not show that the candidates who were scored out were scored fairly. Those candidates never reached the conversation where the choice was offered.


The second is the human review test. The UK Information Commissioner's Office set it out in its report on automated decisions in recruitment on 31 March 2026, which SAASiQ covered on 8 October. A person who approves the tool's output without the authority and knowledge to change it does not count as human review. If the agent rejects most applicants and no recruiter looks at the rejections, that is a solely automated decision. Since 5 February 2026, under the Data (Use and Access) Act 2025, the candidate must then be told, be able to contest it and be able to ask for a person to look again.


The third is liability. In Mobley v. Workday, in the Northern District of California, the plaintiffs have asked the court to certify a wide class of applicants over AI screening. The court had not ruled on that motion by 11 November, as far as SAASiQ can find. A separate case filed in California in January 2026 against Eightfold AI argues that AI scores built from data about job seekers are consumer reports under the Fair Credit Reporting Act. If that view holds, candidates would get notice and a right to dispute before an adverse decision.


The fourth is the credit meter. Usage-based pricing means the cost rises with the number of applications. The agent assesses every one, so a vacancy with many applicants costs more to run. In SAASiQ's view, a recruiting team should know what one assessed application costs before it switches an agent on for every vacancy.


Use case two: judging whether an application is real

The other side of the same process is the candidate's own AI. A survey of 998 hiring leaders by Criteria Corp and Lighthouse Research and Advisory, published in May 2026, found that 92 per cent believe AI-written CVs are now commonplace. Only a third were very confident that CVs show a candidate's real skills and experience. About two-thirds still use a CV screen, by a person or by software, as the first step. Only 2 per cent called the CV their most trusted signal.


First Advantage, a background screening company, published figures from its Collaborate EMEA 2026 event in a release carried by ResponseSource. More than a third of HR professionals, 36 per cent, said AI-generated candidate content had reduced their confidence in hiring decisions. The same 36 per cent figure also appears in an Equifax survey of HR professionals at the SHRM conference earlier in 2026, so the two sources may share data.


Rolf Bezemer, executive vice president and general manager at First Advantage, said AI is not making the CV obsolete, but it is changing what employers can reasonably expect a CV to tell them. He added that employers should not assume a candidate who used AI is being dishonest. First Advantage sells screening, so it has an interest in the subject.


Built In's survey of 10 November looked at the step before the application. It asked 119 job seekers, of whom 105, or 88 per cent, had used AI to find or research employers in the past year. Among those AI users, 66 per cent had applied to an employer they had not considered before because of what the AI told them. And 53 per cent had decided not to apply to an employer for the same reason. Built In sells employer branding, and the sample is small.


Where trust in applications breaks

The first problem is that both sides now use the same tools. A candidate's AI writes a CV that matches the job advert closely. The employer's AI then scores that CV highly for matching the job advert. Each step works as designed, and the employer learns less about the candidate.


The second is the response. Ben Eubanks of Lighthouse said organisations are moving towards structured interviews, assessments and work simulations. Those are slower and cost more per candidate than a CV screen. They also need a fixed set of questions, a scoring guide and trained interviewers, or they bring their own bias back in.


The third is identity. A CV that overstates a skill is one problem. A candidate who is not who they say they are is a bigger one. The FBI warned in June 2022 about fraudulent applicants for remote jobs using deepfake video in interviews. For roles that get system access soon after joining, the identity check belongs before the first day, not after it.


The fourth is what the candidate's AI says about the employer. If a job seeker's AI assistant gives an out-of-date account of pay, benefits or culture, the employer loses applicants it never sees. Built In's figures suggest that happens. In SAASiQ's view, an HR team should ask the main AI assistants about its own organisation once a quarter and correct its public pages where the answers are wrong.


Use case three: deciding which jobs AI can take over

The third use is the one with the largest consequences for staff: workforce planning, and the decision to cut or not fill a role because AI will do the work.


Challenger, Gray and Christmas, an outplacement firm that counts announced US job cuts each month, published its October report on 5 November. US employers announced 43,994 cuts in October, up 2 per cent from 43,281 in September and down 71 per cent from 153,074 in October 2025. AI was cited for 1,265 of them, the lowest monthly total this year. Market and economic conditions are now the leading reason for 2026, at 130,531 cuts. Announced cuts for the year to October total 617,189, down 44 per cent on the same period of 2025.


An analysis published on 5 November by Ryan Doser reviewed 392 tech layoff events from 1 January to 4 November 2026. The share linked to AI peaked in May. In October, 3 of 31 events were linked to AI. The analysis is one person's count from public reports.


Gartner reported on 26 October that 77 per cent of board members expect AI to deliver productivity and cost savings. Only 3 per cent of organisations exceed their cost savings goals through productivity, according to the same research, which drew on a July 2026 survey of 297 chief HR officers.


Workers asked to come back

SHL, which sells workplace assessments, released a report on 29 October on what it calls AI rebounding. Fortune reported its findings on 9 November. The survey was run by Propeller Insights in June and July 2026 among 2,036 employees in the US and the UK.


In all, 16 per cent said they had been replaced by AI and later asked to return by the same employer. The figure was 19.1 per cent in the US and 13 per cent in the UK. Among workers aged 18 to 24 it was 24 per cent. More than half said they had seen it happen, to themselves or to someone they knew.


The reasons given for the rehiring were institutional knowledge that AI could not replace, cited by 40 per cent, and the need to manage relationships, cited by 39 per cent. As a condition of returning, 51 per cent said they would want a substantial pay rise. Reports of the share wanting written guarantees of job security vary between 42 and 43 per cent. And 40 per cent wanted to know how AI would be used in their role.


The figures are self-reported. They depend on each worker's view of whether AI replaced them, and the actual number of reversed AI job cuts is not known. HR Dive, reporting the SHL survey, cited a Gartner prediction that by 2029, 30 per cent of workers terminated or replaced by AI will be rehired, often at a higher cost.


Where AI workforce decisions break

The first weak point is the task list. A role is cut on the basis of the tasks someone thinks it contains. The tasks nobody wrote down, such as knowing which supplier to call or which manager signs off what, leave with the person. SHL's top two reasons for rehiring, institutional knowledge and relationships, are both of that kind.


The second is timing. The cost saving from a cut is counted when the person leaves. The cost of the AI system that replaces them, and of the errors it makes while it beds in, comes later and usually lands in a different budget. Gartner's 3 per cent figure suggests most organisations do not get the savings they planned.


The third is the rehire. A worker asked back after a cut knows they are needed. SHL's survey suggests about half would ask for a substantial rise. That cost belongs in the business case for the original cut.


The fourth is the record. If a cut is made because AI will do the work, and the work then comes back, the reason given at the time may be tested. In the UK, under section 139 of the Employment Rights Act 1996, a redundancy rests on a reduction in the employer's need for employees to do work of a particular kind. In SAASiQ's view, an employer that cites AI as the reason for cutting a role should keep the evidence that the AI was doing that work.


What it costs and who pays

Few prices were published this week. Workable prices its agent on usage-based credits and does not publish a rate per application in the material SAASiQ has seen. First Advantage and Built In sell services and did not give prices in their releases.


The costs that are visible point the same way as in earlier papers in this series. Screening at volume is priced by use, so the total depends on how many applications the agent assesses. Better assessment of candidates, through structured interviews and work tests, costs more per candidate and needs trained staff. An AI-driven job cut saves salary at once and can cost more later, if the role comes back at a higher rate.


Then there are the costs that sit outside any licence. Someone reviews the candidates the agent rejected. Someone keeps the scoring guide for interviews. Someone writes down what a role actually does before it is cut. A business case that counts the recruiter hours saved but none of these will understate the cost.


A framework for HR teams

The steps below apply whichever applicant tracking system, HR suite and screening supplier an organisation uses.


Sample the rejections. For any agent or tool that screens applications, have a recruiter review a random sample of the candidates it scored out, every month. Record how often the recruiter would have progressed them. If that rate is high, the scoring needs changing before more vacancies run through it.


Apply the ICO's test. Ask whether a person with the authority and knowledge to change the outcome looks at each rejection before it takes effect. If not, treat it as a solely automated decision. Tell candidates, let them contest it and offer review by a person.


Price it per application. Find out what one assessed application costs under the contract, and multiply it by the number of applications last year. Set a limit per vacancy if the contract allows one.


Move trust to the work. For roles where the CV no longer tells you much, add a short structured task or work sample with a fixed scoring guide. Check identity before system access is granted, not after.


Check what AI says about you. Ask the main AI assistants about your organisation as an employer, and correct the public pages they draw on where the answers are wrong.


Write down the role before cutting it. Before any role is cut on the grounds that AI will do the work, list the tasks the person does, including the informal ones, and test the AI against that list. Keep the record. If the work comes back, the record shows what was missed.


What happens next

The ICO's final guidance on automated decisions is listed in its forward plan for winter 2026. The court in Mobley v. Workday has yet to rule on class certification. California's notice rules for automated decision tools apply from 1 January 2027, and EU AI Act obligations for high-risk recruitment tools apply from 2 December 2027. Challenger's report on November job cuts is due in early December.

SAASiQ - Intelligent Solutions for SaaS ©

Optimise your SaaS licences and software subscriptions with SAASiQ

bottom of page