The best M&E system I’ve worked with cost almost nothing.
A program coordinator in a rural water and sanitation project had a WhatsApp group with community health promoters. Every two weeks she sent a voice message asking three questions: what’s working, what’s not, and what do you need. She synthesized the answers herself in a one-page note. The program director read it the same day.
No dashboard. No indicator matrix. No quarterly report.
I’m not arguing against rigor. I’m pointing out something that tends to get buried under the weight of frameworks: the purpose of M&E is to produce decisions, not documents.
The gap between those two things is where most M&E investment disappears.
────
What AI actually changes in this equation is the cost of synthesis at scale.
The coordinator with twenty communities can’t produce a two-week synthesis note for each one. But an AI system that processes structured field inputs, groups emerging themes, flags outliers, and drafts a summary for human review — that changes the math considerably. What she was doing intuitively for one community can be done systematically across twenty.
Liao et al. (2024) found that 81% of researchers already use AI in their workflows. In applied M&E, adoption is slower, but the same logic holds: the tasks that eat evaluator time — coding qualitative data, synthesizing across sites, adapting findings by audience — are exactly the tasks where AI assistance is most reliable.
────
Two things don’t change.
The questions have to be right. The coordinator’s three questions worked because she’d spent two years figuring out what her program team needed to hear. No AI generates those questions. They come from knowing the context, the theory of change, and what decisions are actually on the table.
The relationship has to hold. The reason her system worked is that the health promoters trusted her. They sent honest messages because they knew she’d act on them. Data quality in M&E is a relationship problem as much as a technical one, and that’s still true whether you’re using WhatsApp or a machine learning model.
────
The most useful question I can offer to M&E teams thinking about AI integration: what decisions are you currently making slowly because synthesis takes too long? Start there.
#MEL #LearningAndAdaptation #ImpactEvaluation #DevelopmentPractice