Designing “Contestability” in Automated HR Decisions: Appeal Mechanisms, Evidence, and Outcomes
Open accessAbstract
More and more, automated decision systems are being used in Human Resource Management (HRM) to screen resumes, rank candidates, decide who gets promoted, and predict who is likely to leave the company. Even if these tools make things more efficient, they often make it harder for people to question, fix, or challenge decisions, which might be unfair, unclear, and legally accountable. This study examines contestability in automated HR decisions by developing and assessing appeal systems that facilitate substantive dispute, evidence submission, and review. Contestability is defined as a socio-technical capability comprising three dimensions: informational access to decision factors, evidentiary access for submitting corrections or contextual information, and revision authority that delineates who reviews appeals and the standards under which decisions may be revised. Phase 1 uses a mixed-methods designscience approach to find out what kinds of evidence are admissible, what kinds of operational limits there are, and what kinds of contestation needs there are by interviewing applicants, employees, recruiters, and HR compliance officers. Phase 2 analyzes different appeal designs by running controlled trials that look at procedural fairness, trust, privacy concerns, and how well people accept the results. Phase 3 tests good designs in fake recruiting and promotion instances to see how well they fix mistakes, how long it takes to get back to work, and how outcomes change for different groups. Keywords: contestability, automated HR decisions, algorithmic recourse, procedural justice, Explainable AI (XAI), algorithmic governance
AI snapshot
Written from the paper’s full text.
TL;DR
The paper conceptualizes and tests contestability mechanisms for automated HR decisions, arguing structured appeals with explanations, evidence submission, and review authority can improve procedural justice, trust, and error correction. It reports experimental/simulation evidence supporting structured and evidence-based appeals, with trade-offs in workload and residual inequities.
Key findings
- Structured appeal mechanisms (clear explanation + guided evidence submission + defined review timeline) produced higher perceived procedural justice and trust than unstructured or no-appeal conditions; H1 supported. Reported ANOVA: F(2, 12) = 33.90, p < .001, η² = 0.85; regression: structured X1 B = 0.375, p = 0.0194; clarity X2 B = 0.335, p = 0.00160; R² = 0.991.
- Evidence-based appeal designs with structured, verifiable evidence produced higher error-correction and decision-revision rates than narrative-only appeals; H2 supported. HR reviewers showed higher confidence when evidence was standardized and easier to validate.
- Support features (plain language, examples of acceptable evidence, accessibility options) reduced disparities across participant subgroups in successful appeals and completion rates; H3 partially/fully supported, though some gap may remain due to unequal access to documentation, digital skills, or time.
- A trade-off was identified: stronger contestability can increase review time and workload unless triage rules and escalation paths are implemented.
Methods
- Mixed-methods Design Science Research approach: semi-structured interviews with candidates/employees, HR professionals, and compliance stakeholders to identify requirements for meaningful appeals, acceptable evidence, and review constraints.
- Multiple appeal prototypes varied explanation type (factor-based vs. counterfactual), evidence submission (structured vs. open), and review workflow (single vs. panel); a controlled experiment tested effects on procedural justice, trust, understanding, and privacy concerns.
- An HR-reviewer simulation or pilot measured error-correction, revision rates, turnaround time, workload, and subgroup equity impacts; analysis included ANOVA and linear regression as reported.
Limitations
- The paper does not include a dedicated limitations section. It states that some gap may remain due to unequal access to documentation, digital skills, or time, and that broader structural inequalities may still influence appeal success.
- Evident from the text: the reported ANOVA has total df = 14 (implying N = 15), and the regression reports F(2, 3) = 166.61, p = 0.00084, which is inconsistent with the ANOVA degrees of freedom; post-hoc tests are noted as something that “should be conducted,” not as completed.
- Evident from the text: the study relies on prototypes, a controlled experiment, and an HR-reviewer simulation or pilot, which may limit generalizability; the provided abstract is empty apart from author names.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
2 of 7 cited sources checked · 2 supported · 0 flagged
- Premisecitation
If algorithmic systems are treated as authoritative, employees may feel pressure to conform to model-driven expectations even when those expectations conflict with their judgment, identity, or context.
“If algorithmic systems are treated as authoritative, employees may feel pressure to conform to model-driven expectations, even when those expectations conflict with their judgment, identity, or context.”
Quote in the paperSubstantial claim1 of 1 sources checked- cites Leicht-Deobald et al. (2019) source
- Findingown data
Appeals that allowed structured, verifiable evidence (document upload, correction fields, and standardized categories) produced higher error-correction and decision-revision rates than narrative-only appeals.
“evidence (document upload, correction fields, and standardized categories) produced higher error-correction and decision-revision rates than narrativeonly appeals.”
- Also stated in: § Conclusion · plain
“evidence-based pathways improve correction and revision outcomes, supporting decision quality.”
Quote in the paperSubstantial claim - Also stated in: § Conclusion · plain
- Findingown data
Holding other variables constant, each 1-unit increase in explanation clarity increases perceived procedural justice by 0.335 points (p = 0.0016).
“Explanation clarity (X2) is also positive and significant: each 1-unit increase in clarity increases procedural justice by 0.335 points (p = 0.0016).”
Quote in the paperSubstantial claim - Findingown data
Holding explanation clarity constant, switching from a non-structured to a structured appeal increases perceived procedural justice by 0.375 points (p = 0.019).
“Structured appeal (X1) has a positive significant effect: holding clarity constant, switching from non-structured to structured increases procedural justice by 0.375 points (p = 0.019).”
Quote in the paperSubstantial claim - Findingown data
The effect size for the group difference in perceived procedural justice was large, η2 = 0.85.
“The effect size was large (η² = 0.85)”
Quote in the paperNot a substantial claim
- Findingown data
Participants exposed to a structured appeal mechanism (clear explanation + guided evidence submission + defined review timeline) reported significantly higher perceived procedural justice and trust than participants in unstructured or no-appeal conditions.
“Participants exposed to a structured appeal mechanism (clear explanation + guided evidence submission + defined review timeline) reported significantly higher perceived procedural justice and trust compared with unstructured or no-appeal conditions.”
- Also stated in: § Conclusion · plain
“Structured appeal mechanisms increase perceived fairness and trust”
Quote in the paperSubstantial claim - Also stated in: § Conclusion · plain
- Interpretationreasoning
Model-centric interventions such as bias audits, feature selection, explainability tools, or human-in-the-loop review are valuable but insufficient for contestability.
“Many organizational initiatives focus on model-centric interventions such as bias audits, feature selection, explainability tools, or human-in-the-loop review.These are valuable but insufficient.”
Quote in the paperSubstantial claim - Interpretationown data
Contestability is not merely an ethical principle but an operational requirement for responsible HR automation.
“This study demonstrates that contestability is not merely an ethical principle but an operational requirement for responsible HR automation.”
Quote in the paperSubstantial claim - Interpretationown data
Process clarity and review transparency are central to legitimacy in AI-supported HR decisions.
“This indicates that process clarity and review transparency are central to legitimacy in AI-supported HR decisions.”
Quote in the paperSubstantial claim
- Classified as probably not a substantial claim. 10
- May combine two claims. 7
- The quote only partly states this claim. 1
Classified as probably not a substantial claim.
PremisecitationAlgorithmic HR systems can promote an efficiency-driven logic that changes the balance between employee personal integrity and organizational compliance.
“they argue that algorithmic HR systems can promote an efficiencydriven logic that changes the balance between employee personal integrity and organizational compliance.”
Quote in the paperSubstantial claim1 of 1 sources checked- cites Leicht-Deobald et al. (2019) source
Classified as probably not a substantial claim.
FindingreasoningGuidance features (plain language, examples of acceptable evidence, accessibility options) reduced disparities across participant subgroups in successful appeals and completion rates.
“Guidance features (plain language, examples of acceptable evidence, accessibility options) reduced disparities across participant subgroups in successful appeals and completion rates.”
Quote in the paperSubstantial claimMay combine two claims.
Findingown dataThe regression model predicting perceived procedural justice had very strong fit, R2 = 0.991, and was overall significant, F(2,3)=166.61, p=0.00084.
“Model fit is very strong (R² = 0.991), and the overall regression is significant (F(2,3)=166.61,p=0.00084).”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Findingnone givenHR reviewers showed higher confidence when evidence was standardized and easier to validate.
“HR reviewers also showed higher confidence when evidence was standardized and easier to validate.”
Quote in the paperNot a substantial claimThe quote only partly states this claim.
Findingown dataA one-way ANOVA showed a significant difference in perceived procedural justice among the three appeal groups, F(2, 12) = 33.90, p < .001.
“A one-way ANOVA showed a significant difference among the three groups, F(2, 12) = 33.90,p < .001.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S1 (asserted).
FindingreasoningStructured appeals can increase administrative load and turnaround time unless triage rules and escalation paths are implemented.
“While structured appeals improved fairness outcomes, they can increase administrative load and turnaround time unless triage rules and escalation paths are implemented.”
Quote in the paperNot a substantial claimMay combine two claims.
PremisecitationAlgorithmic hiring tools often rely on complex pipelines (parsing, scoring, ranking, filtering), and fairness claims may be based on limited evaluations, narrow metrics, or assumptions that do not hold across employers and job roles.
“The paper highlights that hiring tools often rely on complex pipelines (parsing, scoring, ranking, filtering) and that fairness claims may be based on limited evaluations, narrow metrics, or assumptions that do not hold across employers and job roles.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites Raghavan et al. (2020) source
May combine two claims.
InterpretationreasoningEffective contestability requires both technical design (explanations, evidence intake) and governance design (review authority, timelines, accountability).
“Overall, effective contestability requires both technical design (explanations, evidence intake) and governance design (review authority, timelines, accountability).”
- Also stated in: § Introduction · boosted
“Contestability requires an explicit institutional and technical infrastructure: clearly defined appeal rights, accessible channels for submitting evidence, structured review processes, and accountability for decision revision.”
Quote in the paperSubstantial claim- Also stated in: § Introduction · boosted
May combine two claims.
PremisecitationIn many settings, inputs are causally connected; changing one attribute may be impossible, may require changing others, or may not lead to the desired outcome due to external constraints.
“The paper emphasizes that in many settings, inputs are causally connected; changing one attribute may be impossible, may require changing others, or may not lead to the desired outcome due to external constraints.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites Karimi et al. (2021) source
May combine two claims.
Interpretationnone givenProviding an explanation does not guarantee a pathway for redress; individuals may understand a decision yet still be unable to correct it.
“Moreover, providing an explanation does not guarantee a pathway for redress; individuals may understand a decision yet still be unable to correct it.”
Quote in the paperNot a substantial claimClassified as probably not a substantial claim.
InterpretationreasoningSome gap in successful appeals may remain due to unequal access to documentation, digital skills, or time.
“However, some gap may remain due to unequal access to documentation, digital skills, or time, suggesting contestability must be designed for inclusion.”
Quote in the paperSubstantial claimClassified as probably not a substantial claim. May combine two claims.
PremisecitationRecourse is not the same as explanation, and naïve counterfactuals can be misleading when interpreted as recommended actions.
“Karimi et al. (2021) deepen the contestability discussion by arguing that recourse is not the same as explanation, and that naïve counterfactuals can be misleading when interpreted as recommended actions.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites Karimi et al. (2021) source
Probably not a substantial claim: may not pass S2 (checkable).
InterpretationreasoningBy translating contestability into implementable workflows and measurable outcomes, organizations can align AI-enabled HR decisions with procedural justice, transparency, and human dignity while maintaining operational feasibility.
“By translating contestability into implementable workflows and measurable outcomes, organizations can align AI-enabled HR decisions with procedural justice, transparency, and human dignity while maintaining operational feasibility.”
Quote in the paperNot a substantial claimClassified as probably not a substantial claim.
PremisecitationCounterfactual explanations-statements describing the smallest changes to an input that would flip an automated decision-are a practical route to meaningful information when opening the model is constrained by trade secrets, complexity, or security.
“The authors position counterfactuals as a practical route to meaningful information in settings where opening the model is constrained by trade secrets, complexity, or security.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites Wachter et al. (2018)
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
Probably not a substantial claim: may not pass S4 (based).
Interpretationnone givenA model can be audited and still produce harmful outcomes for individuals due to data errors, context omission, or borderline cases where reasonable people disagree.
“A model can be audited and still produce harmful outcomes for individuals due to data errors, context omission, or borderline cases where reasonable people disagree.”
Quote in the paperNot a substantial claimClassified as probably not a substantial claim. May combine two claims.
PremisecitationContestability is often invoked in ethics guidelines but remains under-specified, making it difficult to implement in real decision pipelines.
“They argue that contestability is often invoked in ethics guidelines, yet remains under-specified in ways that make it difficult to implement in real decision pipelines.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites Lyons et al. (2021) source
The paper relies on this without support.
Automated HR decisions are prone to data quality issues, historical bias, proxies correlated with protected characteristics, and hybrid human-algorithm harms.
“Contestability is especially important because automated HR decisions are prone to several wellknown failure modes.First, data quality issues are common: resumes may be parsed incorrectly; employment histories may be incomplete; performance data may reflect measurement noise rather than true contribution; and records may contain errors that individuals cannot easily detect.”
The paper relies on this without support.
The practical design of contestability mechanisms in HR remains underdeveloped.
“Despite growing interest in responsible AI, the practical design of contestability mechanisms in HR remains underdeveloped.”
Statistics recomputed
Every reported test result, recomputed from its test statistic and degrees of freedom. A result shows as not adding up only once an independent check has confirmed it.
- ConsistentF(2, 12) = 33.9reported p < .001 · recomputed p < .001
“F(2, 12) = 33.90,p < .001”
- ConsistentF(2, 3) = 166.61reported p = .00084 · recomputed p < .001
“F(2,3)=166.61,p = .00084”
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Self-check results appear here once an independent check has confirmed them.
Cites
The publisher hasn’t deposited this paper’s reference list.
- Authors
- Professor, S. O. C., Radha, P., Chandraseka, V., Associate Professor, S. O. C., Vennila, R., & Professor, S. O. C.
- Affiliations
- Jain University (IN)
- ORCID
- Venue
- GS WOW: Wisdom of Worthy Research Journal
- Publisher
- Gaurishankar Campus
- Published
- 2025
- Volume
- 4
- Issue
- 2
- Pages
- 9-15
- ISSN
- 3021-9590 (ISSN-L)
- OpenAlex
- W7125647628
- Semantic Scholar
- 7c9e9d7ede30c1fbb5de220daeca4d7f985953d5
- Cited by
- 0
- References
- 0
- FWCI
- 0.00
- Type
- journal article
- Language
- en
- Open access
- bronze
- doi.org
- Locations
- GS WOW Wisdom of Worthy Research Journal (publishedVersion)
- Topics
- Social Sciences › Safety Research › Ethics and Social Impacts of AI
- Social Sciences › Business, Management and Accounting › Organizational Behavior and Human Resource Management › AI and HR Technologies
- Physical Sciences › Computer Science › Artificial Intelligence › Explainable Artificial Intelligence (XAI)
- Keywords
- contestability, algorithmic recourse, appeal mechanisms, procedural justice, explainable AI, algorithmic governance, algorithmic fairness, evidence submission, privacy concerns, resume screening, promotion decisions, employee attrition prediction, design science
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-30.
Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-30.
Cited sources are checked by gpt-6-sol.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
