Every M&E framework I’ve seen in the last ten years has a learning section.

Almost none of them describe what learning actually looks like in practice: who reads the findings, in what format, by when, and what decision it’s supposed to inform.

I spent a week once reviewing quarterly monitoring reports for a social protection program. The reports were detailed, well-structured, and sent on time to the funder. The program team got a copy too. When I asked the field coordinator which indicator had triggered the most internal discussion, she said she hadn’t read the last two reports. Too long, too much formatting, not enough signal.

The data existed. The learning didn’t happen.

AI changes the first part of this problem in ways that are genuinely useful.

A system that processes field data continuously, flags deviations from expected trajectories, and summarizes findings in a format that field coordinators will actually open — that’s not a hypothetical. It’s available now. Korinek (NBER, 2025) documents how LLM agents run full research workflows, including synthesis and reporting. Applied to M&E, the same capacity compresses the gap between data collection and actionable insight from weeks to hours.

That matters. Programs make bad decisions partly because the feedback loop is slow. By the time a quarterly report reveals a problem, the problem has had three months to compound.

────

But the second part of the problem is harder.

Learning requires that someone with authority to change the program reads the finding, believes it, and acts on it before the next reporting cycle. That chain breaks at every link for reasons that have nothing to do with data quality or reporting speed.

It breaks because the program manager’s incentives are tied to delivery, not adaptation. It breaks because the funder approved a logframe and isn’t asking questions about what’s working. It breaks because admitting a design flaw mid-implementation feels riskier than continuing.

Faster synthesis doesn’t fix any of that.

────

The organizations doing M&E well right now are not necessarily the ones with the most sophisticated data systems. They’re the ones where leadership asks, genuinely, “what are we learning, and what are we changing because of it?”

If you have that, AI makes the cycle significantly faster. If you don’t, it makes the reporting look better.

There’s a difference.

#MEL #MonitoringAndEvaluation #ImpactEvaluation #InternationalDevelopment