Learning agendas, adaptive management and after-action reviews

From “doing more learning” to learning that changes anything

Every programme claims to learn. The test is behavioural: can the team point to a decision that went differently because of evidence, and to the evidence that turned it? Most cannot — not for lack of data, but because learning was never attached to decisions in the first place. The three instruments on this page attach it. A learning agenda decides which questions are worth answering and who will act on the answers. Adaptive management re-plumbs the programme cycle so evidence arrives while decisions are still open. The after-action review extracts lessons from completed work in a form a team can act on within weeks. All three are cheap relative to the M&E machinery most programmes already run; what they cost is management attention.

The learning agenda: questions with owners

A learning agenda is a short, prioritised list of questions whose answers the organisation genuinely intends to use, each carrying three attachments [1]:

• An owner — the named person or unit whose decisions the answer will inform. A question no decision-maker claims is a research interest, not a learning-agenda item.
• An evidence plan — how the question will be answered: existing monitoring data, a study, an evaluation component, a structured review of feedback trends (see feedback and complaints mechanisms). Many good learning questions need synthesis of what already exists rather than new collection.
• A decision point — the moment the answer must arrive to matter: the annual workplan revision, the mid-term review, the scale-up decision.

What a learning agenda is not: a calendar of workshops, a synonym for the M&E plan, or a graveyard folder called “Lessons Learned”. The failure pattern is well documented in evidence-use research — organisations collect and even discuss evidence, but no route connects discussion to authority to act [4]. The agenda’s whole value is that route.

CLA: the framework and its enabling conditions

USAID’s Collaborating, Learning and Adapting framework is the most widely used codification of what “a learning organisation” means for a development programme. Its structure makes one deliberately unglamorous point: CLA pairs the visible practices — collaborating (internal and external coordination, engaging the right stakeholders), learning (learning agendas, scenario thinking, using evidence in the programme cycle) and adapting (pause-and-reflect moments, adaptive management of activities) — with the enabling conditions that determine whether any of it survives: organisational culture (openness to challenge, tolerance for reporting failure), business processes (planning and budgeting that permit change), and resources (staff time and money earmarked for learning) [1].

The enabling conditions are the load-bearing half. A pause-and-reflect session in a programme whose budget cannot be revised, or whose culture punishes reported failure, produces performance, not adaptation [1][4].

Adaptive management: what it changes about M&E

Adaptive management is decision-making under uncertainty done honestly: where the path to results is not fully knowable in advance, the programme is run so that course corrections are expected, evidenced and legitimate. ODI’s roadmap frames the aspiration precisely — adaptation should be evidence-informed, and the evidence trail should be documented, while acknowledging the real trade-off between rigorous documentation and the time it consumes, and the reality that decisions sometimes must proceed on incomplete evidence [2].

For the M&E system, adaptive management is a requirements change [1][2]:

• Shorter feedback cycles. Quarterly reporting designed for accountability arrives too late for adaptation; the monitoring plan needs a fast layer — light, frequent, decision-focused — alongside the formal one.
• Decision-linked monitoring. Each major uncertainty in the programme’s theory gets a named signal that would trigger reconsideration; monitoring watches signals, not just indicators.
• Scheduled pause-and-reflect points with the authority present to change things — a reflection session that cannot revise the workplan is theatre.
• A decision log. Every significant adaptation recorded with its date, the evidence that prompted it, the options considered, and the expected effect — the audit trail that distinguishes adaptive management from drift [2].

Where the theory of change itself needs revisiting rather than the tactics beneath it, that is double-loop territory (below); the practice of maintaining a living theory of change is covered in Monival’s theory of change explainer. Where the evaluation function must itself operate inside this rhythm, the adjacent discipline is developmental evaluation.

The after-action review: a ritual worth protecting

The after-action review is the smallest complete learning instrument: a structured, facilitated discussion held soon after a defined piece of work — a distribution round, a campaign, an emergency response, a failed procurement — that turns the team’s own experience into recorded, assigned change. Its canonical spine is four questions: What was expected to happen? What actually happened? Why were they different? What will we do differently — and who will make sure?

WHO’s guidance, developed for reviewing public-health emergency response, is the most operationally specific open methodology and offers four tested formats scaled to the situation: the debrief (short, single-team, immediately after action), the working group format (multi-functional, over one or more days), the key informant format (interview-based, where participants cannot convene), and a mixed-method combination [3]. Its facilitation counsel generalises well beyond health: prepare with a factual timeline so the review argues about why, not what; keep the review close in time to the action; and separate the review from performance appraisal absolutely — an AAR that can hurt a participant’s rating will never again hear the truth [3].

The output discipline matters as much as the conversation: each recommendation gets an owner and a date, enters the decision log, and is checked at the next review. An AAR whose recommendations are not tracked is a debriefing, not a learning instrument [3].

From learning to change

Evidence-use research across African policy and programme settings converges on an uncomfortable finding: the supply of evidence is rarely the binding constraint — the machinery of use is. What distinguishes settings where evidence changed decisions: relationships and trust between evidence producers and decision-makers built before the moment of use; evidence timed and framed to live decisions; and institutionalised routines (standing review meetings, mandated management responses) rather than one-off products [4]. For a MEAL system the translation is direct — build the routines, not just the reports: a decision log, standing evidence slots in management meetings, and a tracked management-response process for evaluations (covered under use and management response).

Single-loop and double-loop, deliberately. Most adaptation corrects action within existing assumptions — the outreach schedule moves, the targeting criterion is tightened. Periodically, the assumptions themselves must face the evidence: is the causal logic still believed? Scheduling explicit double-loop moments — typically at mid-term or after an AAR pattern repeats — prevents a programme from efficiently optimising its way down a wrong road [2][4].

Anti-patterns

• Learning theatre. Reflection sessions with no authority present and no budget flexibility; outputs are flipcharts, not decisions [1].
• The lessons-learned graveyard. Lessons filed per project, never synthesised, never consulted at the design of the next project — the folder exists so that no one has to read it.
• Adaptation without evidence. “Adaptive” invoked to dignify improvisation; the decision log is empty because the decisions would not survive it [2].
• Learning as a terminal activity. The learning products all arrive at closeout, when the only actor who can still use them is the evaluator of the successor project.

Checklist for an operational learning system

• A learning agenda exists: prioritised questions, each with an owner, an evidence plan and a decision point.
• Monitoring has a fast, decision-linked layer, and pause-and-reflect points sit in the calendar with decision authority present.
• A decision log records each adaptation with the evidence that prompted it.
• AARs follow defined triggers (after each major activity cycle or shock), use the four questions, and their recommendations are tracked to closure.
• At least one scheduled moment per cycle questions the programme’s assumptions, not just its tactics.
• Learning time and money appear as budget lines, not as goodwill.