1. Why this case matters

Humanitarian and development organizations increasingly encounter AI in information-intensive work such as data processing, needs assessment, monitoring and evaluation, translation, reporting, planning, and decision support.

But having access to AI tools does not necessarily mean that an organization is ready to use them responsibly. Readiness also depends on organizational capabilities, governance arrangements, data practices, workforce skills, infrastructure, and the conditions under which people are expected to use AI.

The Frontiers study is particularly useful because it separates two related but different questions: how positively people perceive AI adoption, and whether the organizational conditions needed to support responsible adoption are actually in place. The authors describe this as a layered readiness pattern rather than a simple question of whether an organization is “AI ready.” :contentReference[oaicite:2]{index=2}

2. What did the study investigate?

The study examined AI readiness among stakeholders working within Qatar’s humanitarian and development ecosystem. It used a dual conceptual approach:

• UTAUT: used to understand adoption-related perceptions, including perceived usefulness, effort expectancy, social influence, and facilitating conditions.
• ODTR: used to structure organizational readiness around technological resources, management and governance, human capability, and organizational culture.

The purpose was not to establish a causal model or conduct an independent organizational audit. Instead, the study provides an exploratory stakeholder-based snapshot of how readiness appeared across individual and organizational dimensions. :contentReference[oaicite:3]{index=3}

3. How was the study conducted?

The researchers conducted a bilingual online survey between 25 November 2025 and 30 January 2026. Recruitment used purposive and snowball sampling through an established Technical Working Group involving representatives from 13 institutions.

Because the survey was distributed through professional networks and could be forwarded, the researchers could not determine the total number of people who received the invitation. Consequently, a formal response rate could not be calculated. :contentReference[oaicite:4]{index=4}

This is important when interpreting the results: the sample should be understood as a stakeholder-based snapshot, not a statistically representative sample of every humanitarian and development organization in Qatar.

4. What did respondents see AI being useful for?

The study found that respondents associated AI primarily with practical, information-intensive functions rather than autonomous decision-making.

The particularly high reported value for M&E is relevant for evaluation professionals. It suggests that, within this sample, respondents already saw potential for AI in functions closely related to evidence generation and use.

However, perceived usefulness is not evidence of actual impact. The survey measured respondents’ perceptions; it did not establish that AI had already improved evaluation quality, programme outcomes, cost-effectiveness, or decision-making.

5. The central finding: enthusiasm was higher than organizational readiness

One of the most important findings is the difference between the two index families.

The authors interpret this as evidence of a layered readiness pattern: respondents showed strong adoption-related perceptions, while organizational readiness conditions were more moderate and uneven. Governance and safeguards were the weakest organizational domain. :contentReference[oaicite:5]{index=5}

The numbers should not be interpreted as a percentage of organizations that are “ready.” The composite indices are descriptive measures constructed for this study and are not presented as validated readiness thresholds.

6. Governance was the critical weakness

The study found substantial variation in the reported presence or awareness of organizational governance mechanisms:

• 53.0% reported standard operating procedures relating to privacy or fair data use.
• 25.3% reported data-sharing memoranda of understanding.
• 18.1% reported ethics review processes.
• 27.7% reported an AI ethics framework.

The study also notes that some respondents did not know whether particular governance mechanisms existed. That distinction matters: absence of awareness is not necessarily the same thing as absence of a policy.

The study therefore provides a useful signal about the visibility and consistency of governance arrangements, but it does not independently audit organizational policies or verify whether reported safeguards were actually implemented in practice.

7. Confidentiality, privacy and data responsibility

Confidentiality is particularly important in humanitarian and development settings because organizations may work with personal, sensitive, operational, or otherwise high-risk information.

In the study, 84.3% of respondents identified privacy or potential misuse of data as a medium or major barrier, while 83.1% identified community trust as a barrier. :contentReference[oaicite:6]{index=6}

The researchers also describe confidentiality safeguards for the study itself. Participation was voluntary, electronic consent was obtained, the survey was anonymous, and the dataset was de-identified. The de-identified dataset was stored in a password-protected Qatar University institutional OneDrive environment with access restricted to the authorized research team, with retention and disposal handled according to the relevant ethics requirements. :contentReference[oaicite:7]{index=7}

This is an important distinction: the study’s own research-data safeguards should not be confused with evidence that participating organizations had equivalent AI governance safeguards.

8. What are the major barriers?

These percentages describe respondents’ perceptions. They do not establish that each barrier causes failed AI implementation. The authors explicitly caution against treating the descriptive frequencies as causal evidence. :contentReference[oaicite:8]{index=8}

9. The capability gap is broader than “learning AI”

The study identified gaps between the perceived importance of different capabilities and respondents’ self-rated competence. Areas with notable gaps included:

• Digital humanitarianism
• Data cleaning and visualization
• Machine learning
• Generative AI
• Responsible-AI ethics
• Privacy and security
• Prompt engineering

This suggests that organizational AI readiness cannot be reduced to teaching staff how to write better prompts. Responsible implementation also requires data literacy, verification skills, privacy awareness, risk recognition, and the ability to understand when an AI output should not be trusted or used.

10. Why confidentiality matters especially for M&E

M&E professionals may handle interview transcripts, beneficiary information, survey responses, case-management data, geospatial information, partner information, evaluation reports, and other sensitive material.

Introducing an AI system into an evaluation workflow can therefore create a new data pathway. Before uploading or processing information, an evaluator needs to understand what information is being shared, where it goes, who may access it, how it is retained, what contractual or institutional protections apply, and whether the use is authorized.

This is consistent with wider humanitarian data-responsibility guidance. OCHA’s 2025 Data Responsibility Guidelines provide principles, processes, and tools for safe and ethical management of operational data in humanitarian contexts. :contentReference[oaicite:9]{index=9}

The ICRC likewise treats data protection as part of protecting people affected by humanitarian crises and has developed an institutional AI policy built around a human-centred and humanitarian-principles-based approach. :contentReference[oaicite:10]{index=10}

11. What this means for M&E professionals

EvalCommunity perspective: the study can be translated into a practical M&E readiness question: not simply “Can our staff use AI?” but “Can our organization use AI while maintaining evidence quality, confidentiality, accountability, human judgment and learning?”

A practical readiness assessment can therefore examine at least nine dimensions:

1. Purpose: Is there a clearly defined evaluation or organizational problem that AI is intended to address?
2. Data: Are the data sufficiently accurate, relevant, complete and appropriate for the intended use?
3. Capability: Do staff understand both AI use and its limitations?
4. Confidentiality: Are privacy, security and data-protection requirements understood before information is processed?
5. Governance: Are responsibilities, approvals and acceptable-use rules clear?
6. Human oversight: Can qualified people review, challenge and override AI outputs?
7. Evidence quality: Can important AI-generated claims be traced back to source evidence?
8. Equity: Has the organization considered whether AI performs differently across populations, languages or contexts?
9. Learning: Is there a mechanism for monitoring performance, documenting errors and changing the workflow?

These nine dimensions are an EvalCommunity adaptation for M&E practice, not a framework proposed or validated by the Frontiers study.

12. From AI readiness to AI assurance

For evaluators, readiness should not be treated as a one-time checklist. An organization can have policies and trained staff while still producing unreliable AI-assisted evidence.

Readiness → Implementation → Verification → Assurance → Learning

This distinction is important because an organization can be technically capable of deploying AI without having sufficient evidence that a particular AI application is reliable for a particular evaluation task.

EvalCommunity’s AI Tool Evaluation Protocol for M&E proposes a practical approach involving task specification, a benchmark dataset, a pre-specified scoring rubric, independent evaluation, model/version recording, controlled testing, error analysis, and reproducible results. This is an EvalCommunity framework and is not presented as a methodology developed or endorsed by Google DeepMind. :contentReference[oaicite:11]{index=11}

13. Connecting the case to established AI risk-management practice

The readiness issues identified in the study also connect with established external frameworks.

• ICRC: its AI policy emphasizes a responsible and human-centred approach grounded in humanitarian principles.
  ICRC AI Policy. :contentReference[oaicite:12]{index=12}
• OCHA: its Data Responsibility Guidelines provide practical guidance for managing humanitarian operational data safely and ethically.
  OCHA Data Responsibility Guidelines 2025. :contentReference[oaicite:13]{index=13}
• NIST: the AI Risk Management Framework provides a voluntary framework for incorporating trustworthiness considerations into AI design, development, use and evaluation.
  NIST AI RMF. :contentReference[oaicite:14]{index=14}

14. EvalCommunity resources for applying the findings

• Principles for AI Use in Monitoring & Evaluation
  — nine principles covering purpose, transparency, data quality, ethics, equity, human oversight, learning, accountability and appropriateness. :contentReference[oaicite:15]{index=15}
• AI Tool Evaluation Protocol for M&E
  — a practical protocol for testing and documenting AI performance. :contentReference[oaicite:16]{index=16}
• Use of AI in Evaluation
  — practical guidance covering AI-assisted evaluation workflows, data protection, evidence traceability, bias checks and human review. :contentReference[oaicite:17]{index=17}

15. What should an M&E team actually measure?

The Frontiers study does not provide an M&E indicator framework for organizational AI readiness. The following are therefore EvalCommunity-proposed evaluation questions, derived from the issues identified in the study and from responsible-AI practice.

16. What the study does not establish

This section is essential for interpreting the evidence correctly.

• It does not establish that humanitarian organizations in general are ready or unready for AI.
• It does not provide a representative estimate for the entire humanitarian sector.
• It does not independently verify organizational governance arrangements.
• It does not establish causal relationships between reported barriers and AI adoption.
• It does not demonstrate that AI has improved humanitarian outcomes.
• It does not provide a longitudinal assessment of how readiness changes over time.
• It does not establish the performance of specific AI tools in operational settings.

The authors identify several of these limitations themselves, including the relatively small stakeholder sample, purposive and snowball recruitment, self-reported data, potential self-selection and common-method bias, and the absence of longitudinal or independently verified organizational evidence. :contentReference[oaicite:18]{index=18}

17. A useful M&E interpretation

The most useful lesson for M&E professionals is not that organizations should simply “adopt AI.” It is that AI adoption creates a new object for organizational learning and evaluation.

The evaluation question therefore shifts from:

“Are people using AI?”

toward a broader set of questions:

• What problem is AI supposed to solve?
• What evidence shows that it is solving that problem?
• What errors does it introduce?
• Who is responsible for checking those errors?
• What information is being exposed to the system?
• Who could be disproportionately affected by poor performance?
• What happens when the AI is wrong?
• How does the organization learn from those failures?

This approach moves AI readiness from a technology-adoption conversation toward an evidence, governance and learning conversation.

18. Key takeaway

For humanitarian and development organizations, responsible AI readiness therefore involves much more than access to technology. It includes workforce capability, data responsibility, confidentiality, governance, human oversight, evidence verification, community trust, and mechanisms for learning from experience.

For M&E professionals, this creates an important new area of practice: evaluating not only what AI can produce, but how AI changes the evidence system around it.

Original research and further reading

Primary source:
Attia Y, Maddah D, Al-Kaabi A-A, Seifo G, Al-Sada MB, Mundodan JMA, Jadeed AK, Abu-Rmeileh NME, Kwalombota K. (2026).
Responsible AI readiness in a humanitarian and development ecosystem: an exploratory case study.
Frontiers in Public Health, 14.

Original article

|  DOI

External authoritative resources:

• ICRC — Building a Responsible Humanitarian Approach: The ICRC’s Policy on Artificial Intelligence
  
• OCHA Centre for Humanitarian Data — Data Responsibility Guidelines 2025
  
• NIST — AI Risk Management Framework
  

EvalCommunity resources:

• Principles for AI Use in Monitoring & Evaluation
  
• AI Tool Evaluation Protocol for M&E
  
• Use of AI in Evaluation