AI-native vs AI-skinned
Applying AI within collection or analysis can change what a team can do. Adding an AI summary to an unchanged process is mostly a coat of paint. Ask which one a proposed use actually is.
AI for M&E: A Practitioner's Resource
Guides for M&E teams using AI, from finding where it fits to doing a task and governing use across your organization. Everything here is free.
Where AI Fits Across the M&E Lifecycle
AI is a different decision at each stage. Suitability describes how much it can safely take on; a person remains accountable for every decision.
The Augmentation Gradient
AI can assist, automate a first pass with human review of every item, or run at scale with human spot-checking. The appropriate point depends on risk and how reliably output can be checked. Decisions still require a person in the loop.
Quality-assurance guideDesign
Theory of change, indicators, evaluation questions, instruments
AI suitabilityMedium to high
What AI does well
Draft items, menus, and variants; check coverage gaps against a framework you supply.
Where a human is mandatory
Context and political sensitivity, and the soundness of the underlying theory.
Related resources
11 guides and playbooks
- How to Build a Theory of Change with AI (guide)
- How to Use AI to Design an M&E Framework (guide)
- How to Use AI for Indicator Development (guide)
- How to Write a Logframe with AI (guide)
- How to Use AI to Write a MEL Plan (guide)
- How to Build Better Surveys with AI (guide)
- Build a Theory of Change with AI (playbook)
- Build a Results Framework with AI (playbook)
- Develop Indicators with AI (playbook)
- Build a MEL Plan with AI (playbook)
- Build an Evaluation Plan with AI (playbook)
Data Collection
Fieldwork, instruments, consent
AI suitabilityLow to medium
What AI does well
Translation, transcription, formatting instruments.
Where a human is mandatory
Consent, sampling integrity, field ethics. AI does not make these decisions.
Related resources
2 guides and playbooks
Data Cleaning
Cleaning and harmonization
AI suitabilityHigh
What AI does well
De-duplication, categorization, record linkage, normalizing inconsistent entries.
Where a human is mandatory
Silent error propagation. Every transformation needs validation before you trust it.
Related resources
2 guides and playbooks
Quantitative Analysis
Patterns, summaries, analysis code
AI suitabilityMedium
What AI does well
Pattern detection, descriptive summaries, drafting analysis code.
Where a human is mandatory
Causal inference and interpreting significance stay with you.
Related resources
1 guides and playbooks
Qualitative Coding
Thematic analysis, open-ended responses
AI suitabilityHigh, with a human in the loop
What AI does well
Coding open-ends against a codebook at scale, surfacing candidate themes.
Where a human is mandatory
Codebook validity and intercoder reliability. This is both the most-cited win and the most-cited disappointment.
Related resources
Evidence Synthesis
Literature review, text-mining across reports
AI suitabilityHigh
What AI does well
Text-mining across hundreds of reports, surfacing and highlighting relevant passages.
Where a human is mandatory
Bias toward written, English-language evidence. What is not in the documents stays invisible.
Reporting
Drafting, gap-checks, visualization
AI suitabilityHigh
What AI does well
Drafting from structured evidence, gap-checks, first-pass visualization.
Where a human is mandatory
Strategic recommendations, framing, and accountability are yours. Always disclose and cite.
Related resources
Learning
Adaptive management
AI suitabilityMedium
What AI does well
Near-real-time insight in place of periodic reporting.
Where a human is mandatory
Insight without sense-making is not learning. A faster dashboard is not a decision.
Design Medium to high
Theory of change, indicators, evaluation questions, instruments
What AI does well: Draft items, menus, and variants; check coverage gaps against a framework you supply.
Where a human is mandatory: Context and political sensitivity, and the soundness of the underlying theory.
- How to Build a Theory of Change with AI (guide)
- How to Use AI to Design an M&E Framework (guide)
- How to Use AI for Indicator Development (guide)
- How to Write a Logframe with AI (guide)
- How to Use AI to Write a MEL Plan (guide)
- How to Build Better Surveys with AI (guide)
- Build a Theory of Change with AI (playbook)
- Build a Results Framework with AI (playbook)
- Develop Indicators with AI (playbook)
- Build a MEL Plan with AI (playbook)
- Build an Evaluation Plan with AI (playbook)
Data Collection Low to medium
Fieldwork, instruments, consent
What AI does well: Translation, transcription, formatting instruments.
Where a human is mandatory: Consent, sampling integrity, field ethics. AI does not make these decisions.
Data Cleaning High
Cleaning and harmonization
What AI does well: De-duplication, categorization, record linkage, normalizing inconsistent entries.
Where a human is mandatory: Silent error propagation. Every transformation needs validation before you trust it.
Quantitative Analysis Medium
Patterns, summaries, analysis code
What AI does well: Pattern detection, descriptive summaries, drafting analysis code.
Where a human is mandatory: Causal inference and interpreting significance stay with you.
Qualitative Coding High, with a human in the loop
Thematic analysis, open-ended responses
What AI does well: Coding open-ends against a codebook at scale, surfacing candidate themes.
Where a human is mandatory: Codebook validity and intercoder reliability. This is both the most-cited win and the most-cited disappointment.
Evidence Synthesis High
Literature review, text-mining across reports
What AI does well: Text-mining across hundreds of reports, surfacing and highlighting relevant passages.
Where a human is mandatory: Bias toward written, English-language evidence. What is not in the documents stays invisible.
No how-to guide yet.
Reporting High
Drafting, gap-checks, visualization
What AI does well: Drafting from structured evidence, gap-checks, first-pass visualization.
Where a human is mandatory: Strategic recommendations, framing, and accountability are yours. Always disclose and cite.
Learning Medium
Adaptive management
What AI does well: Near-real-time insight in place of periodic reporting.
Where a human is mandatory: Insight without sense-making is not learning. A faster dashboard is not a decision.
No how-to guide yet.
Beyond M&E: AI Across the Wider Program
Service delivery and beneficiary-facing uses call for even more caution because real people are directly affected.
Service Delivery and Beneficiary-Facing Tools
Chatbots, triage, translation in the field
AI suitability: Use with care
Does well: Translation, drafting information materials, routing and triage support behind a human.
Human mandatory: Anything that gives advice, makes an eligibility call, or touches a vulnerable person directly. Consent, data protection, and a clear human fallback are not optional.
In-Project Analysis and Operational Reporting
Operational data, internal updates
AI suitability: Medium
Does well: Summarizing operational data, drafting internal updates, flagging anomalies for review.
Human mandatory: The same silent-error and oversight cautions as the M&E stages. Keep a person between the output and any decision.
Guides and Playbooks for the Work
Choose a short guide or a longer playbook. Each result says which format you are opening.
Foundations
The skills and safeguards that apply to every AI-for-M&E task.