Human-Centered AI Maturity Model

• Human-Centered AI Maturity Model (HCAI-MM) is a framework that defines five maturity stages to enhance AI practices through measurable metrics and structured governance.
• It integrates both technical performance and socio-technical dynamics by employing quantitative indicators, iterative design cycles, and standardized tools.
• The model guides organizations from initial HCAI efforts to optimized practices, validated by real-world case studies in healthcare and technology.

The Human-Centered AI Maturity Model (HCAI-MM) is a staged organizational framework designed to systematically assess and advance an enterprise’s capability to design, develop, deploy, and govern AI systems that prioritize human needs, values, and experiences. It offers a roadmap from basic, ad-hoc HCAI efforts to optimized, industry-leading organizational practices, coupling technical and social dimensions through quantifiable metrics, structured governance, standardized tools, and a documented methodology that interweaves organizational design with HCAI progression (Winby et al., 17 Dec 2025).

1. Conceptual Foundations and Scope

HCAI-MM is defined as a maturity model comprising five sequential stages by which organizations can evaluate, monitor, and incrementally enhance the design and implementation of human-centered AI (HCAI) practices. The scope encompasses all elements required for robust HCAI: human-AI collaboration, explainability, fairness, and user experience. The core purposes are to (1) articulate a staged progression from novice to leader, (2) provide metrics and tools for self-assessment, and (3) institutionalize organizational mechanisms that ensure continuous, measurable enhancement of HCAI capabilities. HCAI-MM uniquely integrates organizational design perspectives directly into the progression framework, unlike prior models that treat socio-technical and technical change in isolation.

2. Maturity Stages: Structure, Criteria, and Objectives

HCAI-MM delineates five progressive stages of maturity, each defined by specific practices, metrics, governance structures, and benchmarks:

Stage	Characteristics & Capabilities	Key Objectives
Level 1: Initial	Isolated HCAI pilots, reactive AI, low awareness. Metrics: $M_{\text{entry}}$, $M_{\text{train}}$.	Executive sanction, readiness assessment, awareness.
Level 2: Developing	Emerging frameworks, basic user research/testing. Metrics: $\Delta_{\text{DP}}$, $F_{\text{feedback}}$.	Institute frameworks, social/technical analysis.
Level 3: Defined	Formal governance body, published guidelines. Metrics: $S_{\text{interp}}$, $U_{\text{success}}$.	Standardize user-input, launch multi-disciplinary training.
Level 4: Managed	HCAI embedded in KPIs, lifecycle integration. Metrics: $A_{\text{compliance}}$, social impact index.	Audit mechanisms, HCAI dashboards organization-wide.
Level 5: Optimizing	Continuous innovation, external advocacy, co-design. Metrics: $R_{\text{CI}}$, stakeholder engagement.	Shape standards, maintain user communities.

• $M_{\text{entry}} = \frac{\text{# completed entry tasks}}{\text{# defined entry tasks}}\times 100\%$
• $M_{\text{train}} = \frac{\text{# stakeholders trained}}{\text{# total stakeholders}}\times 100\%$
• $$\Delta_{\text{DP}} = \left|P(\hat y=1|A=0) - P(\hat y=1|A=1)\right|$$
• $F_{\text{feedback}} = \frac{\text{# feedback events}}{\text{time period}}$
• $$S_{\text{interp}} = \frac{1}{N}\sum_{i=1}^N s_i,\;s_i\in[1,5]$$
• $U_{\text{success}} = \frac{\text{# tasks completed successfully}}{\text{# tasks tested}}\times 100\%$
• $A_{\text{compliance}} = \frac{\text{# projects passing ethics audit}}{\text{# audited projects}}\times 100\%$
• $R_{\text{CI}} = \frac{\text{# iterative changes based on feedback}}{\text{time period}}$

Progression relies on both quantitative measures (e.g., audit scores, usability rates) and qualitative practices (e.g., establishing cross-functional governance).

3. Metrics for Human-Centered Maturity

Metrics in HCAI-MM are defined across four central dimensions:

• Human-AI Collaboration Index (HAC):

$HAC = w_h\cdot\frac{\text{# successful human-AI tasks}}{\text{# total tasks}} + w_c\cdot\frac{\text{# collaborative sessions}}{\text{time period}}$

with $w_h + w_c = 1$.

• Explainability Score (EXP):

$$EXP = \frac{1}{N} \sum_{i=1}^{N} \left[ s^{\text{local}}_i + s^{\text{global}}_i \right], \quad s \in [0,1]$$

integrating local and global model explanation ratings.

• Fairness Gap (FG):

$$FG = \max_{a \neq b} \left| P(\hat y=1|A=a) - P(\hat y=1|A=b) \right|$$

quantifying disparities across protected attributes.

• User Experience Composite (UX):

$$UX = \alpha \cdot S_{\text{sus}} + \beta \cdot U_{\text{success}} + \gamma \cdot T_{\text{task}}$$

with $S_{\text{sus}}$ as System Usability Scale, $U_{\text{success}}$ the task completion rate, $T_{\text{task}}$ normalized time, and $\alpha+\beta+\gamma=1$.

These measurements support evidence-based benchmarking and progression across maturity stages.

4. Governance Structures, Toolkits, and Best Practices

Governance mechanisms and tooling are stage-specific, scaling in complexity and organizational embedment as maturity increases:

Stage	Governance/Tools	Best Practices
Level 1	Self-assessment surveys	Assign HCAI sponsor, awareness workshops
Level 2	User feedback platforms, IBM AI Fairness 360, draft guidelines	Pilot usability/fairness tests
Level 3	Design-lab environment, LIME/SHAP, HCAI committee, published design guidelines	Stakeholder sign-off in lifecycle
Level 4	CI/CD dashboards, MS Fairness Dashboard, internal/external audits	Quarterly HCAI reviews, impact assessments
Level 5	Co-design portals, live analytics, public ethics reports	Annual summits, external research grants

Tool adoption and best practices are mapped to maturity level, with compliance and ongoing audit institutionalized from stage 4 onward.

5. Organizational Design and Socio-Technical Cycle

HCAI-MM embeds progression in a five-phase socio-technical design cycle—operationalized in LaTeX as:

$$\text{Entry} \longrightarrow \text{Research {data} Analysis} \longrightarrow \text{Design Lab} \longrightarrow \text{Implementation} \longrightarrow \text{Adaptation}$$

Phases are:

• Entry & Sanction: Secure executive buy-in and conduct readiness scan.
• Research & Analysis: Perform both technical (process mapping, variance identification) and social analyses (user research, task analysis).
• Design Lab: Iterative prototyping and multi-stakeholder deliberation, integrating ethical frameworks.
• Implementation: Pilot deployment, training, and establishment of feedback loops.
• Adaptation: Continuous monitoring, detection and correction of variances, refinement of governance mechanisms.

A simplified TikZ representation formalizes the workflow for organizational communication and planning.

6. Empirical Validation: Case Studies

Empirical case studies illustrate real-world progression across maturity levels:

• Mayo Clinic (Healthcare, Level 2 → 3): Transitioned from an NLP-based clinical scheduling pilot with co-design to institution-wide deployment by formalizing HCAI guidelines, instituting governance checkpoints, and systematic usability testing. Resulted in the publication of design principles and cross-departmental tool scaling.
• IBM HR (Technology, Level 2 → 4): Advanced from initial explainable dashboards and manager feedback (Level 2) to Level 4 by incorporating fairness audits in HR processes, forming an AI Ethics Committee, and embedding HCAI KPIs and dashboards company-wide.

These exemplars validate the staged approach and highlight the criticality of embedding governance, continuous measurement, and structured feedback at each step.

7. Significance, Utility, and Progression Pathways

HCAI-MM enables organizations to benchmark current HCAI practices, select and implement appropriate governance structures and tools at each stage, operationalize systematic socio-technical design cycles, and accelerate progress by learning from peer case studies. By institutionalizing quantitative and qualitative measurement of human-AI collaboration, explainability, fairness, and user experience, and integrating these into both technical and organizational subsystems, the model provides a foundation for cultivating human-centered, ethically grounded, and continuously evolving AI capabilities (Winby et al., 17 Dec 2025).