The Unexamined Front End

The front of the contingent hiring funnel is more automated, higher in volume, and less supervised than direct hire, and almost nobody has examined it for disability.

Eight of thirteen programs have no recruiter training requirement on disabled or neurodivergent candidates; two require it and verify completion. Five of thirteen have taken no action on AI hiring-tool bias in two years; three do not know; three have audited tools or changed vendors. Five of sixteen say their interview and assessment design works for neurodivergent candidates.

Practitioners ranked their sourcing barriers and refuted the convenient assumption. Recruiter awareness and bias carried the best mean rank; no active pipeline drew the most first-place votes; screening and assessment tools came third; worksite readiness came last, ranked fifth by half the respondents. The barriers are people and pipeline, not buildings. Asked about organizational barriers, respondents put lack of understanding first and lack of leadership awareness second. Not resistance. Absence.

The exposure is no longer theoretical. In Mobley v. Workday, a federal court conditionally certified a collective of applicants over 40 across a platform used by more than 10,000 employers, and on June 22, 2026 allowed an ADA proxy-discrimination claim to proceed: screening on employment gaps, which correlate with medical leave and recovery, screens out disabled candidates by proxy. The first U.S. bias-audit mandate, New York City's Local Law 144, requires audits for sex and race/ethnicity only. The regulatory net has a disability-shaped hole in it. Protection has to be procured, not awaited.

The leading programs do four unexotic things: invite self-identification at onboarding, track accommodations somewhere, verify training instead of announcing it, and put disability on the supplier scorecard. A recruiter's account in this brief shows the latent demand: when her firm added one manual accommodation field to onboarding, requests roughly tripled in a quarter. People do not ask for what they think will get them screened out.

Start Monday by asking every screening-tool vendor one question: does your validation measure adverse impact on candidates with disabilities? Almost none do. Then issue requisition standards that strip non-essential physical requirements, and train recruiters and prove it.

Council members also receive the register behind this series: 167 documented inclusive hiring programs at 144 North American public companies, with source documents. Request access or talk with the Working Group at pathways@defiantsolutions.com.

This is Brief 3 of four drawn from The Missing Workforce: Disability and Neurodivergent Inclusion in the Contingent Workforce, a white paper of the Pathways Opportunity Council's Neurodivergence and Disabled Contingent Hiring Best Practices Working Group, published September 2026. This brief examines the front of the funnel: which sourcing barriers practitioners actually rank highest, how little recruiter training programs require, how few have asked their AI screening vendors a single question about disability, and why the courts are now asking it for them. Each brief stands alone; together the four carry the full argument. The complete report, including methodology, the survey instrument, and full response tables, is available from the Council at pathways@defiantsolutions.com.

When practitioners ranked five sourcing barriers (n = 12), the results refuted the field's most convenient assumption. By average rank: (1) recruiter awareness and bias in early screening, which eight of twelve placed second and which carried the best mean rank; (2) no active pipeline or sourcing partnerships for this population, which drew the most first-place votes (six of twelve); (3) screening and assessment tools that disadvantage neurodivergent candidates, polarized between first and fourth place; and, tied for last, (4) client worksite environment and (5) exclusionary job descriptions. Half the respondents ranked worksite readiness dead last. The barriers are people and pipeline, not buildings. Programs that have deferred inclusion pending "worksite audits" are solving the problem they do not have. The open-ended answers add the texture: "lack of knowledge and acceptance of neurodivergent workers by employers," "lack of awareness of the benefits neurodivergent workers bring," time-to-fill pressure that leaves no room for a different screen, and one program's candid note that experience requirements can rule out candidates a population has already been held back from acquiring.

Asked to rank seven organizational barriers (n = 11), respondents put the two awareness items on top: lack of understanding or interest first, and lack of leadership awareness or buy-in second; between them the two drew seven of eleven first-place votes. Unclear legal responsibility in the tri-party model ranked third, supplier readiness fourth, technology fifth, budget sixth, and competing DEI priorities last. Not resistance, but the simple absence of the issue from program governance agendas. One respondent put it plainly: with so many priorities, "the extra effort of building this out correctly is a big barrier. We wouldn't want to do it wrong, so it might be best to keep status quo." That finding pairs with the practitioner accounts in Chapter 6 of the full report: the programs that moved did so after one leader asked one question ("where do accommodation requests go?") and discovered there was no answer. It also explains why the highest-rated interventions in Section 5.6 of the full report are visibility mechanisms (scorecard metrics, tracked training, explicit contract language, certification recognition): they put the issue on agendas that currently do not contain it. Pipeline reach, which drew the most first-place votes among sourcing barriers, is the most readily solved: Disability:IN's DOBE program certifies disability-owned businesses, including staffing firms, for supplier-diversity credit, specialist neurodiversity employment intermediaries maintain screened talent networks, and programs like DXC's Dandelion have open-sourced their entire supported-employment curriculum.¹²

Eight of thirteen programs have no recruiter training requirement for disability inclusion: four require none, and four neither track nor require it of their suppliers. Two do not know; one offers training without requiring it; two require it and verify completion. Recruiters are the people who decide, candidate by candidate, whether a screening tool's output is trusted. Five of thirteen have taken no action in the past 24 months on AI hiring-tool bias, three do not know, two say they do not use AI tools that could create the risk, and three have audited tools, changed vendors, or modified scoring criteria. Conversations suggest "no action" usually means the question has never been put to the vendor. Programs' self-assessments agree: five of sixteen say they have implemented technology tools that reduce bias against disabled and neurodivergent workers, and five of sixteen say their interview and assessment design works for neurodivergent candidates. This is the unexamined front end: contingent hiring is more automated, more voluminous, and less supervised than direct hire, which makes it the highest-exposure surface for exactly the failures now entering the courts.

The exposure is no longer theoretical. In Mobley v. Workday, the first major AI-screening discrimination case, a federal court has conditionally certified a collective of applicants over 40 rejected since September 2020 across a platform used by more than 10,000 employers, and on June 22, 2026 the court allowed an ADA "proxy discrimination" claim to proceed: the allegation that screening on patterns like employment gaps, which commonly correlate with medical leave, treatment, and recovery, screens out disabled candidates by proxy.³ The ACLU's 2024 FTC complaint against Aon Assessments makes the same point about gamified cognitive tests and AI-scored video interviews, tools embedded in contingent pipelines everywhere.⁴ Meanwhile the first U.S. bias-audit mandate, New York's Local Law 144, requires audits of automated employment decision tools for sex and race/ethnicity only; disability is not a required audit category, and the audit industry that has formed around it (auditing to the four-fifths rule, NIST's AI RMF, and ISO/IEC 42001) does not measure disability screening impact as a standard offering.⁵ Colorado's AI Act was rewritten in May 2026 and its enforcement is currently stayed; the EU AI Act's employment provisions face a proposed delay to December 2027; the EEOC removed its own AI-and-ADA technical assistance from its website in early 2025.⁶ The regulatory net has a disability-shaped hole in it, which means the protection has to be procured, not awaited. Playbook Section 8.2's audit checklist starts with the only question that currently matters: does your validation measure adverse impact on candidates with disabilities, and if not, why not?

Three composite profiles illustrate how the machinery gap is experienced from inside. Each is assembled from conversations among Working Group members, who operate or oversee contingent programs themselves, and from informal exchanges with survey respondents and Council network members during the fielding period. Details are composited and anonymized so that no single organization is identifiable, and the quotations are composite renderings of what practitioners told us, not verbatim transcripts. The survey data in Chapter 5 of the full report confirm each pattern the profiles describe.

A senior recruiter at a light-industrial supplier described the accommodation process at her firm as "me." Candidates disclose to her directly (a request for a written rather than verbal instruction set, for a stool on the line, for a consistent shift because of medication schedules) and she negotiates each one informally with the site supervisor. It works because she is experienced and trusted; it would collapse if she changed jobs, and it is invisible to everyone: the client does not know the accommodations exist, the MSP sees nothing, and no record survives the assignment. When her firm joined a program that added an accommodation field to its onboarding checklist, a manual field, not a VMS feature, request volume roughly tripled in the first quarter. The demand was always there. The channel was not.

People don't ask for what they think will get them screened out. The day we made asking safe, the requests tripled. Nothing else changed; not the jobs, not the sites, not the people.

The survey's high-maturity minority, the programs that would score Operational or Embedded on the Council Inclusion Maturity Model (Chapter 7 of the full report), do four things differently, and none of them is exotic. They invite self-identification at contingent onboarding, with the same voluntary, firewalled framing used in direct hire. They track accommodations somewhere; if not in the VMS, then in a named, owned register with response times. They verify training rather than announcing it: attestation, audit, scorecard weight. They put disability on the supplier scorecard, which is what converts policy into supplier behavior. What distinguishes these programs is not budget or sophistication; it is that someone assigned each piece of the machinery an owner.

The market's evidence points the same direction. The employer programs with documented outcomes at scale (Microsoft's decade-old neurodiversity hiring program, SAP's Autism at Work, EY's Neuro-Diverse Centers of Excellence, JPMorgan Chase's Autism at Work, and DXC's Dandelion Program) share the survey leaders' DNA: structured alternative assessment instead of standard screens, trained interviewers, named support ownership, and measured retention.⁷ DXC Dandelion is the most instructive for this industry because it is structurally a staffing model: DXC employs, trains, and embeds neurodivergent teams inside client organizations, with most clients converting participants to permanent hires, and reports 92 percent retention across 350 current team members in eight countries, figures that are vendor-reported but corroborated by an independent cost-benefit evaluation.⁸ A note of sourcing discipline the Working Group applies to its own members too: spectacular productivity claims attached to some corporate programs ("teams 48 percent faster," "up to 140 percent more productive") have never been published with methodology, and this report deliberately relies on retention, which is consistently reported, plausible, and decisive against contingent churn, rather than on unauditable multipliers.⁹

Staffing suppliers

Suppliers touch the candidate first and the worker longest:

Train recruiters, and prove it. Deploy the curriculum in Playbook Section 7.1 (disclosure, accommodation basics, neurodivergent candidate engagement, screening-tool limits) and provide completion evidence to programs that ask.

Fix job descriptions and screening. Apply the requisition standards (Playbook Section 6.1): strip non-essential physical requirements, separate genuine demands from boilerplate, and offer assessment alternatives as standard practice, not special favor (Playbook Section 6.2).

Build pipeline partnerships. Register with DOBE directories and specialist neurodivergent talent networks, connect with university channels such as the USC Lancaster pipeline (Section 9.2 of the full report), and track source-of-hire for disabled and neurodivergent candidates (Playbook Section 6.3).

Pitch by sector. The employer corpus shows where neurodiversity and disability hiring infrastructure already concentrates: finance and technology run the mature cohort programs and skills-based assessment redesigns, industrials build accommodation capacity through centralized teams and technician pipelines, and healthcare has the least named-program infrastructure of any major sector, making it the largest greenfield for a supplier arriving with a ready framework.

Make asking safe. Adopt the application-stage accommodation notice and intake script (Playbook Section 6.4). The practitioner evidence is unambiguous: when asking is safe, request volume rises. That is the system working, not failing.

VMS and technology providers

The technology tier sets the market's ceiling, and the survey says the ceiling is low:

Meet and publish accessibility standards. WCAG 2.1 AA conformance with a current VPAT for every candidate- and worker-facing surface, tested with assistive-technology users, not only automated scanners.

Open the AI black box. Publish validation documentation for screening and matching models that includes disability: subgroup adverse-impact analysis, accommodation pathways inside assessments, and audit rights for customers. Local Law 144 does not require it; the market should. The AskEARN/PEAT eRecruiting checklist offers a ready framework.¹¹

Monday morning

Fix the front end. Issue the job-description and requisition standards from Playbook Section 6.1, and ask every screening-tool vendor the disability question from the AI audit checklist in Playbook Section 8.2: does your validation measure adverse impact on candidates with disabilities? Almost none do; New York's bias-audit law, the first in the nation, does not require it.¹²

The Pathways Opportunity Council is an invite-only council of workforce professionals focused on expanding access and economic mobility for underrepresented talent, including Black, Hispanic, disabled, and veteran communities, as well as other historically disadvantaged groups. The Council's mission is to elevate industry practice and the professional standing of its members by convening leaders committed to measurable improvements in access, hiring, and advancement outcomes. The Council prioritizes organizations that demonstrate sustained, authentic commitment to expanding opportunity. Research and thought leadership are sponsored by Primary Talent Partners and AgileOne and executed by Defiant Solutions as the Council's named consulting partner. For more information email us at pathways@defiantsolutions.com.