A Three-Lens Framework
Navigating AI & Digitalization in M&E: A Three-Lens Framework
Why a Framework Is Needed
Artificial intelligence and digitalization are reshaping how evidence is generated, analyzed, and used. Across the M&E profession, practitioners encounter a growing volume of AI-related tools, guidance, training, and organizational policies. Some of this is useful. But the sheer volume and diversity of the landscape — spanning technologies from satellite imagery analysis to large language models, and audiences from individual practitioners to those responsible for strengthening national M&E systems — has made it difficult for professionals and institutions to determine what is relevant, reliable, and appropriate for their specific circumstances.
The core proposition:
This paper offers a framework to help M&E and evaluation professionals and institutions navigate complexity. It proposes a way of thinking — a set of complementary perspectives that help practitioners and institutions ask better questions about if, where, how, and under what conditions AI and digital tools may have a role in their work.
Reframing the Discourse: Three Limitations of Current Approaches
The paper identifies three features of the current discourse that the framework is designed to complement and extend:
1. AI treated as a monolithic technology
Computer vision applied to satellite imagery, a predictive machine learning model for program targeting, and a large language model for qualitative coding differ in the data they require, the skills needed, the ways they can go wrong, and the evaluative questions they can address. Developing undifferentiated guidance on "the strengths and limitations of AI" is roughly as useful as guidance on research methods without distinguishing between randomized controlled trials and participatory action research.
2. Tool-first rather than need-first orientation
Much AI-related content adopts a tool-first orientation: beginning with an AI technique and illustrating potential applications. This is useful for practitioners who already know they want to use a particular technology. However, it is less well suited to more common situations: a practitioner facing a specific evidence need who wants to understand whether an AI-enabled approach might help address it. Good evaluation practice has always started from the question rather than the method.
3. Fragmented contributions
Existing contributions often focus on particular dimensions in isolation: applications mapped to evaluation phases, competency implications, ethical dimensions, or responsible practice guidelines. But in practice, these dimensions come together. A practitioner deciding whether to use NLP for qualitative analysis is simultaneously considering evidence needs, workflow, and capability. A framework that helps practitioners consider them jointly supports more integrated decision-making.
The Three Lenses: A Practitioner-Centered Framework
The framework is organized around three lenses — evidence needs, workflow, and capability — each capturing a distinct relationship between practitioners and technology. Notably, the framework does not include a lens organized around a taxonomy of AI technologies. This is intentional: a technology-taxonomy lens starts from the tool and works toward applications — precisely the tool-first orientation this framework is designed to move beyond. The three lenses are practitioner-centered: each starts from where practitioners are — what they need to know, what they need to do, what they need to become.
Lens 1: Evidence Needs
"What kinds of evidence do we need, and which types of AI-enabled approaches might help us generate each?"
This lens is epistemic. The practitioner has identified one or more evidence needs and wants to know whether specific AI-enabled approaches might expand what is possible — addressing questions that were previously infeasible, or enabling richer, more timely, or more granular evidence than conventional approaches allow. It draws on the GEI policy/program cycle framework.
Lens 2: Workflow
"What are the constraints in our evaluative work processes, and could AI-enabled approaches help address them?"
This lens is operational. The practitioner faces constraints — in data access, processing capacity, analytical bandwidth, or communication reach — and wants to know whether a specific type of AI-enabled approach can address them. It draws on the GEI task framework for evaluative activities.
Lens 3: Capability
"What do we need to know and have in place to work responsibly in an AI-influenced environment?"
This lens is developmental. The practitioner or institution wants to know what skills, knowledge, ethical frameworks, and institutional capabilities are needed to remain competent, credible, and responsible. It draws on the GEI Evaluation Competency Framework.
Lens 1 in Practice: Evidence Needs Across the Policy/Program Cycle
The evidence needs lens draws on the GEI policy/program cycle framework, which organizes evidence needs by the stages at which they inform policy and program decisions. The framework identifies five stages, each with central questions and established ways of answering. AI applications are situated within these ways of answering: not as replacements but as tools that may support, augment, or extend them.
A key insight from the evidence needs lens
The potential for AI to contribute to M&E is not evenly distributed. Ways of answering that involve large-scale pattern recognition, data integration, or systematic scanning of large evidence bases are well-suited to current AI capabilities. In contrast, ways of answering that are fundamentally normative or deliberative — convening stakeholder judgment on priorities, weighing competing values, interpreting findings in light of context — require human judgment that cannot be derived from data alone. Thematic analysis of qualitative data sits in the middle: LLMs can process large volumes of text, but validity depends on whether categories and interpretations are meaningful in context — requiring human evaluative judgment.
Lens 2 in Practice: Three Types of Workflow Enhancement
The workflow lens asks how specific AI and digital tools could enhance evaluative work processes. The paper distinguishes three types of workflow enhancement:
The paper draws on the GEI task framework for evaluative activities, which organizes evaluative work into task clusters: managing the overall process, framing, team engagement, design, data collection and analysis, and reporting and application of findings. Table 2 in the paper provides a detailed mapping of AI applications to each task cluster — from NLP-supported stakeholder mapping during process management, to LLMs as a sounding board for evaluation design, to AI-powered transcription and semi-automated coding for qualitative data processing.
Lens 3 in Practice: Individual Competencies and Institutional Capabilities
The capability lens asks what M&E professionals and institutions need to know, be able to do, and have in place to engage with AI and digitalization responsibly. The emergence of AI does not require wholesale rewriting of professional competency frameworks. Most of what makes a good M&E professional remains unchanged. Rather, AI requires additional, AI-specific competencies within the existing structure.
The paper also addresses institutional capabilities: data readiness, governance and policy frameworks, infrastructure and tools, role clarity and team composition, and institutional learning processes. A key insight is the sequencing perspective: before investing in advanced AI applications, are foundational capabilities in place — data infrastructure, governance mechanisms, basic digital literacy? Premature introduction of AI in settings where foundational M&E system strengthening is still underway risks wasting resources and eroding trust.
Ethics Across the Lenses: Differentiated, Not Generic
The framework addresses ethics not as a standalone topic but as a dimension that runs through each lens. Through the evidence needs lens, ethical questions center on the validity and fairness of AI-generated evidence. Through the workflow lens, ethical questions center on quality, consent, and the distribution of benefits. Through the capability lens, ethical questions center on professional responsibility and institutional accountability. This differentiated treatment enables practitioners to identify the specific ethical questions most relevant to their situation, rather than working from generic checklists.
Why Integration Matters: Two Illustrative Examples
Each lens individually provides a useful but partial view. The framework's central proposition is that the three lenses are interdependent in practice. Considering the lenses together surfaces trade-offs and interactions that a single-lens view would miss.
Example 1: Using an LLM for drafting evaluation reports
Through the workflow lens alone, this looks straightforward: it saves drafting time, tools are accessible, and capability requirements appear low. But adding the evidence needs lens introduces complexity: a well-prompted LLM could identify patterns a time-pressed human might miss, but LLMs also flatten nuance and fabricate content. Adding the capability lens surfaces the most critical risk: evaluators need the judgment to distinguish when AI-generated text accurately reflects evidence and when it subtly distorts it — a competency difficult to develop and easy to overestimate. A workflow-only view would adopt enthusiastically; the integrated view adopts cautiously, with substantial human oversight.
Example 2: Satellite-based nighttime light imagery for evaluating rural electrification
Through the evidence needs lens, this is transformative: satellite sensors capture data at geographic and temporal resolutions that household surveys cannot match. Through the workflow lens, it may replace expensive field-based data collection. But through the capability lens, critical constraints emerge: interpreting the data requires understanding of sensor limitations, geospatial processing skills, and crucially, ground-truthing against local data to validate what the imagery represents. Without ground-truthing, impressive satellite imagery can create false precision. The integrated view suggests strategic investment: high evidence and workflow value, but requires deliberate capability building.
Practical Applications of the Framework
Clarity and transparency in guidance and communication
Rather than generic recommendations about "AI in evaluation," guidance notes can specify which lens or lenses an AI application primarily addresses, helping practitioners assess relevance to their own situation. Conference panels can use the three-lens structure to specify which dimension is under discussion. Terms of reference and reporting can use the framework to specify AI-related expectations.
Prioritization and investment decisions
Organizations developing AI strategies can use integrated three-lens analysis to make differentiated investment decisions rather than adopting blanket directives. Appendix D of the paper provides illustrative assessments of four AI applications — nighttime light remote sensing, ML for heterogeneous treatment effect analysis, NLP for qualitative data analysis, and LLMs for drafting — demonstrating how integrated analysis supports nuanced decision-making.
Knowledge curation and resource navigation
The three-lens structure can serve as an organizing architecture for curating AI-related resources in a knowledge repository, tagging resources according to one or more lenses. This supports practitioners who approach the repository with a specific problem — an evidence gap, a workflow constraint, a capability need — and surfaces lens-specific considerations they might not have otherwise encountered.
Key Takeaways for Evaluators and M&E Professionals
Start with the question, not the tool
The framework explicitly rejects tool-first orientation. Good evaluation practice has always started from the question rather than the method — and AI does not change this.
Not all AI is the same
Computer vision, predictive ML, and LLMs differ fundamentally. Undifferentiated guidance on "AI" obscures critical differences in data requirements, skills, risks, and evaluative questions.
Human judgment remains central
Even where AI can process data at scale, validity depends on human evaluative judgment. The challenge is not to become technologists, but to remain evaluators.
Capability building is not optional
Critical judgment may matter more than technical skill. As AI tools become more accessible, the professional challenge shifts from "can I use this tool?" to "should I use it, and can I critically evaluate its outputs?"
Frequently Asked Questions
Why doesn't the framework include a lens organized around AI technologies?
A technology-taxonomy lens starts from the tool and works toward applications — precisely the tool-first orientation the framework is designed to move beyond. The three lenses are practitioner-centered: each starts from where practitioners are — what they need to know, what they need to do, what they need to become.
Does the framework apply only to formal evaluations?
No. The framework applies across the full range of evaluative activities — from needs assessments and design studies through implementation monitoring, impact evaluation, and evidence synthesis. The GEI policy/program cycle and task framework encompass this breadth.
How does the framework address ethics?
Ethics is not a standalone lens but a dimension that runs through all three, manifesting differently in each. Through evidence needs: validity and fairness of AI-generated evidence. Through workflow: quality, consent, and distribution of benefits. Through capability: professional responsibility and institutional accountability.
Can the framework be applied at the national M&E system level?
Yes. Each lens can be applied at multiple levels: individual/project, organizational, and national M&E system. The questions change across levels, but the three-lens structure remains relevant. This is particularly important for organizations supporting evaluation capacity development across individual, organizational, and system levels.
Staying Evaluators, Not Becoming Technologists
The paper concludes with a powerful observation: the M&E profession is well-equipped to navigate this moment. The same capacities that define good evaluation — rigorous thinking, ethical commitment, contextual sensitivity, methodological pluralism — are precisely those needed to engage with AI responsibly. The challenge is not to become technologists, but to remain evaluators.
The framework is offered as a structured starting point — to be tested, enriched, and refined through application. The author invites the global M&E and evaluation community to apply, test, challenge, and refine the framework.
Related Resources
