The Human Makes the Final Call, After the Machine Decides Who Gets One: Recruitics, Maki People, and the Pre-Apply Screen

Sponsor content

Talroo brings clarity to improve new-hire retention

You filled the seat, but not with someone who plans to stay. The candidate made it through your entire interview process while quietly interviewing with other companies that pay 25% more.

When expectations aren't clear up front, retention suffers.

Get clarity up front with Talroo's SmartQualify.

Get a demo today

The reliability of disposition data is, I am quite sure, going to be an ongoing topic for the rest of 2026.

Today, I want to talk about the other end of the spectrum: AI screening.

Recruitics is bringing AI voice interviews into the “pre-apply” (how is it pre-apply if it is a screening interview?) experience through an integration with Maki People. The pitch is efficiency with a human safety net. The product pages tell a more complicated story.

Here’s what we’re covering this week:

• Recruitics and Maki People partnership
• Who is Maki People
• What Mochi actually scores
• The audit list and the three missing demographic categories 

THE ANNOUNCEMENT

Last week, Recruitics announced an integration that makes Maki’s voice AI agent, Mochi, part of the application flow through AdaptiveApply, its candidate experience technology. I want to be precise about what this is: it “shifts candidate assessment from the post-apply process to an earlier stage of the hiring journey.” Screening moved earlier. That’s it. Screening candidates is not groundbreaking. It is the oldest activity in recruiting. What’s new is who conducts it, what gets scored, and when the scoring starts.

Here is the release’s own description of the product:

“With a human always making the final call.” That last clause is carrying a huge amount of the lift here, and we have to go down this path together.

WHO IS MAKI PEOPLE?

Maki (Maki SAS) is a Paris-based AI hiring platform founded by Maxime Legardez, Paul-Louis Caylar, and Benjamin Chino. In January 2025 the company raised a $28.6 million Series A led by Blossom Capital. The platform is five AI agents covering intake through hire recommendation, and the client roster is legitimately impressive: Capgemini, Deloitte, PwC, BNP Paribas, Nespresso, and Volkswagen.

Of note, HireAIScore, a compliance directory operated by the consultancy Casework, scored Maki 45 out of 100, an F.

Casework has its own commercial interests, so weigh the letter grade however you like. However, for a vendor whose pitch leans hard on the word “defensible” the audit trail should be available.

(Updated on July 17, 2026: the bias audits exist, and the auditor has a name. My piece, relying on an independent compliance directory and on Maki’s own product page as it existed on July 10, said Maki published no downloadable bias audit and named no independent auditor. That was wrong. Maki’s screening agents are audited annually under NYC Local Law 144 (NYC LL144) by Holistic AI, with the most recent audits of Mochi and Shiro completed March 23, 2026)

(Updated on July 17, 2026, confirmed rather than corrected: ISO 42001. Maki People confirms it does not hold the certification, describing it as a deliberate choice to prioritize EU AI Act conformity, with ISO 27001 in hand.)

KNOCKOUT QUESTIONS OR SELECTION PROCEDURES?

There has always been a compliant and standard way to screen early in the apply process: objective knockout questions. Do you have a CDL? Can you work nights? Are you authorized to work in the US? These map cleanly onto what employment law calls basic qualifications, which the regulatory framework required to be objective, non-comparative, and job-related, established before anyone answers.

Mochi is more than knockouts, and we don’t have to infer it. Maki’s product architecture says it. Knockout questions belong to Shiro, Maki’s skills screening agent. The Mochi page describes its own eligibility screening as “complementing Shiro’s knockout logic.” Maki built a separate product for knockouts. Mochi exists to do something else.

Making a jobseeker talk to a robot for yes-or-no questions would be a waste of everyone’s time, and vendors do not raise $28.6 million to waste time.

So what does Mochi score? The product page says it plainly: “Real-time evaluation of behavioral, situational, and linguistic performance,” scored against a fixed behaviorally anchored rating scale, with the system evaluating “eligibility, motivation, availability, communication, and basic job-fit signals through structured conversational scoring.” The Recruitics CEO’s quote in the release says candidates are assessed on “their skills, communication, and fit right inside the apply flow.”

Selection procedures with adverse impact require validation under the Uniform Guidelines on Employee Selection Procedures, which remain EEOC’s rules regardless of what happened to the federal contractor compliance (more on that in a moment).

And linguistic scoring in particular is a legal minefield with its own zip code. Scoring spoken communication implicates national origin discrimination through accent, and disability discrimination for candidates with speech differences, hearing loss, or autism spectrum profiles. EEOC’s 2022 guidance on algorithmic tools under the ADA addressed exactly this category of risk, and made a second point the employer is liable for the vendor’s tool and cannot rely on the vendor’s own assessment of adverse impact.

WHAT IS AN APPLICANT, AND WHO JUST MADE them at scale?

In 2005, OFCCP’s Internet Applicant Rule answered the question the internet had made urgent: who counts as an applicant for recordkeeping purposes?

Four criteria define an applicant:

• An expression of interest through electronic means,
• Consideration by the employer for a particular position,
• Meeting the basic qualifications;
• And not withdrawing before a selection decision

Now the part your compliance attorney will email me about if I skip it. The rule was promulgated under Executive Order 11246, which was revoked in January 2025. OFCCP halted enforcement, and in July 2025 the Department of Labor proposed formally rescinding the regulations, calling them null and void. Per Littler’s analysis, the internet applicant rule is no longer an enforceable EO 11246 obligation. So no, you cannot wave 41 CFR 60-1.3 at a vendor in 2026 and call it a day.

But the rule’s logic did not die. It relocated, and it left behind an inconvenient insight for this pitch.

The old rule contained a genuine safe harbor: data management techniques. Random sampling, first-N cutoffs. Submissions nobody reviewed never became applicants, because nobody considered them.

An AI interview at the start of the process is the exact opposite of not-reviewing. It is individualized, recorded, scored consideration of every person who completes it. The press release says so twice. Mochi “evaluates every candidate against validated, science-backed criteria.”

And my favorite line in the entire release, from the benefits list, promises candidates “a more streamlined path to consideration.”

And the consideration question still has teeth in three venues.

1. Section 503 and VEVRAA are statutes, not executive orders; they survived the rescission intact, OFCCP is funded at roughly $101 million for fiscal 2026, and contractors must still maintain applicant flow data by disability and veteran status.
2. Title VII disparate impact claims remain available to private plaintiffs and state enforcers even after the administration deprioritized federal enforcement, because an executive order cannot amend a statute, and Littler’s own footnote advises employers to consult counsel before dismantling applicant tracking systems for exactly this reason.
3. State AI hiring laws, Illinois, NYC Local Law 144, Colorado’s SB 26-189, all assume tiy can show what your automated tools evaluated and how.

Which is where Maki’s defense comes in. The company’s help center states plainly that Mochi makes no hiring decisions, no AI decides whether you are selected or excluded, and humans decide everything. Taken on its own terms, fine. But Mochi’s own workflow diagram ends with this step: “Candidates who pass the threshold flow to the next stage.”

A threshold the machine applies determines who a human ever sees, and NYC’s AEDT law covers tools that substantially assist discretionary decisions, not just ones that replace them. Mobley v. Workday, the pending class action over AI screening tools, is live proof that “the human made the final call” is not functioning as a liability shield.

The human makes the final call after the machine decides who gets one.

THE AUDIT LIST

Here is the detail I just couldn’t leave out. The headline on Mochi’s product page is “Voice screening, 24/7, bias-audited.” The fairness testing behind that claim, from the same page: “Adverse impact testing across gender, age, and nationality before any test ships.”

Not race.

Not disability.

Not veteran status.

Now, precision. Maki claims NYC Local Law 144 bias audits elsewhere, and LL144 audits cover sex and race/ethnicity by definition, so I will not tell you they have never tested for race. What I can tell you is the product page’s own fairness list omits race entirely.

(Updated on July 17, 2026: race and ethnicity are tested. The “gender, age, and nationality” list I quoted described Maki People’s internal pre-ship testing only. The independent Holistic AI audit covers gender and race/ethnicity at both standalone and intersectional levels as required under the NYC LL144 regs.)

Disability and veteran status are a different story. Those are the two categories OFCCP still actively enforces. They are the two categories for which federal contractors must still produce applicant flow data. And they are nowhere in the published testing scope of a product whose own feature list says it scores “linguistic performance.”

A tool that grades how people speak, with no published evidence it was ever tested against the populations likely to speak differently, positioned at the very front of the hiring process for employers who answer to federal agencies who I promise are still watching. (Just like each one before, trust me, this administration will not last forever.)

So employers evaluating this integration get two questions to ask yourselves and two to ask at the demo.

For yourselves:

A scored AI interview means every person who completes it was individually considered. Your considered applicant pool just got larger, not smaller. Is a bigger pool of scored, rejected candidates what you actually wanted? How does that pool look when someone runs the adverse impact math against it? Hi ya, standard deviation!

For the demo:

• Show me adverse impact testing for disability status specifically.
• Walk me through how a candidate who is deaf, or stutters, or is on the spectrum, requests an accommodation before your agent scores their linguistic performance.
• Or better yet, walk me through how a candidate who is {insert disability here} is scored by your agents.

So what do you think? Different perspective, different information, share it with me. We are always stronger together.

Until Next Time,

Julie “The Doc” Sowash

Standard Disclaimer: Julie is not a lawyer. She has never been a lawyer. She does not play one on TV, at parties, or in this newsletter. Nothing here constitutes legal advice, and citing “but the Job Board Doctor said…” will impress exactly zero judges. Hire a real attorney for real legal questions.

Got a tip, a press release that deserves a closer read, or a vendor demo that made your compliance spidey-sense tingle? Send it my way. The tip line is always open.

[Want to get Job Board Doctor posts via email? Subscribe here.]