The criteria for AI candidate qualification software

Receiving applications has never been easier; making use of them has never been harder. Between the volumes generated by multichannel distribution and applications written with AI, manual screening is reaching its limits. AI qualification software promises to clear this bottleneck. You still need to know how to assess it: here are the six criteria that separate a serious tool from a black box.

What is AI candidate qualification software?

It is a tool that automatically analyses each incoming application, enriches it with additional data, compares it with the criteria of the role and produces a justified score. It prioritises the recruiter's work; it does not decide in their place to reject a candidate.

AI qualification sits between the application and the interview:

1. Reception: the application arrives from a job board, an advertising campaign or the careers site.
2. Enrichment: the profile is completed beyond the self-declared CV.
3. Scoring: the profile is assessed against the criteria of the role, with a justification for each criterion.
4. Prioritisation: applications arrive ranked in the ATS, and the recruiter handles the most relevant ones first.

The need has long been quantified: an eye-tracking study by The Ladders measured that a recruiter spends on average 7.4 seconds on a CV during the first screening. At that speed, errors and bias are inevitable. Automatic qualification aims for a first screening that is faster, more consistent and documented.

Sources: The Ladders, eye-tracking study, 2018.

Why automate candidate qualification?

Because manual screening does not scale: it is slow, costly, inconsistent from one recruiter to another and impossible to verify afterwards. Automation processes 100% of applications with the same criteria, in minutes, and frees up time for interviews.

The framing data:

• It took on average 44 days to fill a position, a record, according to the Time to Hire Factbook by the Josh Bersin Company and AMS (June 2023); screening applications is one of the most compressible steps in that timeline.
• In the United States, the average cost of a hire was close to 4,700 dollars according to SHRM benchmarking data (April 2022), before even counting the cost of a vacant position.
• AI use has become widespread: 99% of hiring managers surveyed use AI in some form in their process, according to the Insight Global AI in Hiring 2025 survey (published in December 2024).

Automated screening has therefore become a market standard. The real selection question is no longer "do we need an AI qualification tool" but "which one meets the following six criteria".

Sources: The Josh Bersin Company and AMS, Time to Hire Factbook, June 2023; SHRM, April 2022; Insight Global, AI in Hiring Survey 2025, December 2024.

What does good application enrichment look like?

Good enrichment completes the self-declared CV with verifiable data relevant to the role: reconstructed career path, inferred skills, consistency of experience, geographic availability. It must be lawful in its collection and transparent about its sources.

The questions to ask the vendor:

• What data do you add beyond the CV, and where does it come from?
• Is collection GDPR-compliant: documented legal bases, data that is relevant and proportionate to the role?
• Is the enriched data visible to the recruiter, separately from the data declared by the candidate?
• Is the candidate informed of the processing, as the regulation requires?
• How fresh is it: is the data collected at the time of application or taken from ageing static databases?

Opaque enrichment is a double risk: a compliance risk for the company, and a risk of error if scoring relies on false or outdated data. The rule: no enrichment without traceable sources.

Why must scoring be explainable?

Because an unexplained score is unusable for the recruiter, indefensible towards the candidate and out of step with the European regulatory trajectory. Each score must break down into readable criteria: which are met, which are missing, and with what weight.

What distinguishes explainable scoring:

The legal framework pushes in the same direction. Article 22 of the GDPR governs fully automated decisions with a significant effect on individuals, which includes rejecting an application. And the European AI Act classifies AI recruitment systems as high-risk systems, with transparency, documentation and human oversight obligations applying to standalone systems from 2 December 2027 (timetable revised by the AI Omnibus, which came into force on 27 July 2026). A vendor unable to explain its scores will not be able to become compliant.

This is the approach taken by the new generation of agentic tools: at Seeqle, for example, the Match Agent combines enrichment with a score out of 5 explained criterion by criterion, sent back into the ATS depending on what your ATS allows; the decision remains human.

Sources: Regulation (EU) 2016/679, Article 22; Regulation (EU) 2024/1689 and AI Omnibus, European Commission, 2026.

Which compliance requirements should you check?

Check four blocks: GDPR compliance of the processing (information, rights, retention periods), readiness for the AI Act, effective human oversight of decisions and auditability. Ask for documents, not promises.

The compliance checklist:

1. GDPR: data processing agreement (Article 28), record of processing activities, data hosting specified, configurable retention periods, procedure for answering candidates' rights requests.
2. AI Act: a clear position on the high-risk classification, a compliance roadmap to December 2027, ready-to-use transparency notices for your candidate journeys (the transparency obligations of Article 50 have applied since 2 August 2026, a date the AI Omnibus did not postpone).
3. Human oversight: the tool must allow the recruiter to view, correct and override any score; no rejection should be irreversible or invisible.
4. Auditability: logging of scores, export of justifications, the ability to reconstruct a decision months later, especially in case of challenge.

Note for employers: the AI Act's AI literacy obligation has applied since February 2025. The AI Omnibus has softened it: the employer must take measures so that the teams using the tool reach a sufficient level of AI literacy, including its limits.

Sources: Regulation (EU) 2024/1689; AI Omnibus, in force since 27 July 2026.

Which integrations should the software offer?

A two-way, real-time integration with your ATS: scored applications, their justification and their enriched data must appear directly in your pipeline, with no re-keying or manual export. Without that, the tool creates yet another silo.

Checkpoints:

• Native connectors with the main ATS on the market; mature platforms cover dozens of them (60+ ATS and CRM at Seeqle, for example).
• Content of the flow: the score and its explanation must travel with the application, not stay in the tool.
• Latency: qualification must appear in the ATS within minutes, so the candidate can be answered quickly.
• Upstream: does the tool also connect to the sources of applications (careers site, campaigns, job boards) so everything is qualified in one place?
• Synchronised statuses: a candidate already handled in the ATS must not be followed up again by the tool.

Test the integration on your own ATS instance during the trial: it is the most frequent point of friction during deployment.

What qualification speed should you expect?

Qualifying an application should be almost immediate: from a few seconds to a few minutes between reception and the score being available in the ATS. This speed determines how quickly you can respond to the best candidates, who leave the market first.

Why speed is an economic criterion and not a comfort:

• The best profiles, especially passive ones, disengage quickly; answering in hours rather than weeks changes the conversion rate to interview.
• With an average of 44 days to fill a position (Josh Bersin Company and AMS, June 2023), every day saved on screening carries through the whole process.
• Instant qualification allows short feedback loops: if a campaign produces poorly targeted applications, you see it within days and correct the targeting, not at the end of the campaign.

In practice, automating the upstream part (distribution plus qualification) frees up screening time, which is reinvested in interviews and the candidate experience.

Sources: The Josh Bersin Company and AMS, Time to Hire Factbook, June 2023.

How do you measure the qualification rate?

The qualification rate is the share of applications received that meet the essential criteria of the role. It is the central indicator of the distribution and qualification pair: measure it by channel and by campaign, and steer the budget with the cost per application relative to that rate.

Measurement method:

1. Define the essential criteria for each role (degree or certification, experience, area, availability) before launch, not after.
2. Measure by source: the rate varies widely between a general job board and a targeted campaign; this comparison is what guides the trade-offs.
3. Track over time: a declining rate signals targeting that needs adjusting or criteria that have become unrealistic for the market.
4. Cross-check with the downstream stages: the interview rate and hires confirm that the scoring predicts reality well.

Orders of magnitude: on targeted campaigns with automatic qualification, rates of 40 to 50% are documented (53% on average across Seeqle campaigns, 45% at Rothelec, 42% at Stanley Field), whereas untargeted distribution typically produces a majority of applications that do not meet the criteria. Good qualification software therefore does more than sort: it gives you the figures to improve the upstream part, campaign after campaign.