For a recruiter facing hundreds of applications, a polished CV is a starting point. The next task is to understand what sits behind it: what a candidate actually did, how they approached a problem and whether their experience fits the role.
HireWithLumi’s proposed answer combines AI assessment of a CV with a live, two-way screening conversation, followed by recruiter review and human interviews. An enterprise graduate recruitment example supplied to Onrec gives a picture of how that process is being used at volume. It also exposes an important distinction: processing applications is measurable, but demonstrating better hiring requires different evidence.
From 2,163 CVs to a human interview shortlist
In figures supplied on 28 September, HireWithLumi’s Dane Kostic said Lumi had processed 2,163 CVs across four roles in a graduate intake for a US-based enterprise customer. He described a programme using Lumi across multiple countries. The client is unnamed because the example is covered by an NDA.
The largest role, Digital Project Analyst, received 837 applications. Of those, 296 candidates were invited to the company’s Enhanced Assessment, 155 completed a full live screening with Lumi, and 34 were shortlisted for human interviews.
These are distinct stages. The 2,163 figure covers CV processing across all four roles; it is not a count of live screenings or necessarily unique people. Within the Digital Project Analyst role, an invitation did not mean a completed screening, and a place on the shortlist did not mean a job offer.
Kostic clarified that the 141 invitees who had not completed screening included candidates who never started, started but did not finish, or declined the invitation. The supplied account does not break down those categories or establish whether any assessments remained outstanding at the reporting date. The gap is not a count of rejections. Nor does the account explain what happened to applications outside the invited group.
Kostic also said the team had noticed apparently implausible figures, inconsistent work histories or unusually strong role claims in many CVs within the non-completing group. He suggested that questions probing those claims might have discouraged some candidates from proceeding. These are vendor observations and a possible explanation, not verified findings of fraud or evidence of why individual candidates did not complete screening.
All of these numbers come from HireWithLumi. Onrec has not independently audited the underlying records or obtained confirmation from the anonymous customer. They describe reported activity, rather than establishing the accuracy or fairness of the selection process.
What the live screening adds
In its written explanation to Onrec, HireWithLumi says the initial CV assessment identifies points that merit closer examination. Screening then gives candidates an opportunity to explain gaps, substantiate claims and describe experience that may not be clear from the application alone.
The company says questions can be personalised, standardised or a mixture of both, and that recruiters can edit them. This customer uses a hybrid approach: every candidate receives a fixed set of core questions, with personalised follow-up questions generated from their answers. Candidates can ask questions or request a rephrase, and their responses can affect their assessment score.
HireWithLumi describes this as a screening stage, separate from the human interview that follows. It says candidates are told that no preparation is required and that the assessment takes no more than 20 minutes.
The intended benefit is an opportunity to explore an application before a recruiter commits interview time. A claim about managing a project, for example, can become a discussion of responsibilities, decisions and results. A missing detail can prompt a question rather than an assumption. Follow-ups are intended to probe beyond an initial, potentially rehearsed answer.
The shared core questions introduce an element of structure. Wider research provides a reason to take structured interviews seriously: a 2022 meta-analysis by Sackett and colleagues in the Journal of Applied Psychology ranked structured interviews highest for predicting job performance among the selection procedures it reviewed.
That finding does not validate Lumi’s AI screening, its generated follow-ups or its scoring. A common question set alone does not establish that this implementation is equivalent to the structured interviews studied. Employers still need evidence that the questions, follow-ups and assessment criteria consistently capture what matters for the job.
A person at the decision point
“AI should do the reading. People should do the deciding.”
Robert Hodds, founder and CEO, HireWithLumi
Hodds supplied the statement to Onrec.
Hodds says the platform was built so that candidates are not rejected by automated means alone, with reasoning and supporting evidence attached to scores and the ability to override them. That is the company’s stated principle; the graduate programme figures do not independently demonstrate how it was applied to every application.
HireWithLumi says recruiters can review assessment summaries and scores, examine findings and watch recordings before deciding how to proceed. It also says recruiters may make a shortlist or rejection decision from the summary and score alone.
This makes the design of the employer’s review process central. The ability to override a score needs to be backed by time, access to the underlying evidence and permission to disagree. Review should reach the earlier points where applicants leave the process, as well as the final interview shortlist.
A useful check would be to examine a sample of applications that did not progress, including cases near a threshold, and record why the reviewer agreed or disagreed with the assessment. Otherwise, a human interview at the end says little about candidates who never reached it.
Access and integrity flags
The candidate’s route through screening also deserves attention. Kostic clarified that candidates can request a human screening through the platform only when the recruiter enables that option. If it is unavailable, candidates can complete the AI screening or decline it, with further consideration left to the recruiter.
Employers considering this approach should establish how candidates can raise access needs, report technical problems or seek another way to demonstrate their suitability. The enterprise example does not establish which alternatives this customer offered.
HireWithLumi also describes a separate integrity check using audio and video signals to flag moments for recruiter review. It says these flags do not change the candidate’s assessment score and acknowledges that ordinary behaviour, such as looking away while thinking, can produce false flags.
According to the company, each flag links to a timestamped moment in the recording, allowing the recruiter to watch the relevant passage in context before acting.
A flag is therefore a prompt to examine evidence, rather than proof of cheating. Recruiters need to consider context and give candidates an opportunity to explain a concern before relying on it.
What candidates think
Separately, HireWithLumi supplied anonymised screenshots of candidate correspondence which it says came from its own hiring process. These concern a separate Sales Director recruitment exercise, not the anonymous customer’s graduate intake.
The selected replies include praise for the screening technology and appreciation for feedback following rejection.
“A glorious rejection” that “actually cheered me up”.
Selected candidate feedback supplied by HireWithLumi, from its separate hiring process
Another respondent said that in 20 years they had never seen technology like the platform used for their interview.
Other replies add a different perspective. Candidates explained experience they felt their CV had not fully conveyed, asked the employer to reconsider transferable skills or expressed interest in alternative roles. Such responses show why an opportunity to clarify an application can matter.
These are vendor-supplied anecdotes, not independently authenticated correspondence or a representative sample of candidates. They do not establish overall satisfaction, assessment accuracy or how candidates in the graduate programme experienced screening. Appreciation for a rejection email should also be distinguished from an endorsement of the assessment itself.
Measuring the benefit
Kostic says the enterprise customer previously conducted screening manually, recorded it and had each screening reviewed by a two-person panel. He says moving to Lumi saved thousands of staff hours.
That remains a vendor claim. The supplied material does not provide a quantified methodology that would allow Onrec to verify it, including the measurement period, comparable workloads, screening and review durations, or time spent configuring the system and resolving exceptions.
There is a plausible operational case for reducing repetitive screening work. Establishing the size of that benefit would require a comparison that also counts the work retained or added around the technology.
The same distinction applies to quality and fairness. HireWithLumi has described external testing and checks within its assessments, but Onrec has not reviewed a testing report or subgroup results. Application volumes and shortlist counts cannot establish those outcomes.
There is also a timing limit on what this graduate example can show. Kostic says the recruitment programme went live in mid-September and the graduates had not yet started work at the time of his clarification. No subsequent job-performance data therefore exists for this group under any screening method. That absence reflects the stage of the intake; it is not a finding about the effectiveness of AI or manual screening.
For recruitment teams, this example makes the workflow more tangible: CV processing, selective invitations to live screening, recruiter decisions and a human interview stage. Its value as a case study lies in showing those stages clearly.
Whether the approach improves hiring depends on what happens within and between them: relevant assessment criteria, accessible participation, scrutiny of the evidence and people able to challenge the system before an applicant loses an opportunity. Those are the conditions against which screening at scale should be judged.
About this feature
Reported and edited by Onrec, with information and factual clarifications supplied by Dane Kostic and Robert Hodds, founder and CEO of HireWithLumi. The company supplied the anonymised customer figures and selected candidate correspondence; Onrec has not independently audited those records.
Illustrative stock photograph: KOBU Agency / Unsplash. The people pictured are not identified as participants in the recruitment programme.
