Purpose
Turn a program description and a target population into a usable MEL framework — anchored to a theory of change, with indicators at every level of the results chain, defined data collection methods, an evaluation design, and structured learning loops. Eliminates the common failure mode of tracking easy outputs while impact goes unmeasured.
When to Use
Invoke this skill when the user asks to:
- Design a MEL, M&E, or measurement framework for a program
- Define metrics or indicators for a nonprofit intervention
- Build a theory of change into a set of measurable outcomes
- Prepare a measurement plan for a funder reporting or evaluation requirement
- Choose data collection methods and an evaluation approach for a program
- Fix a program that tracks activity but can't demonstrate impact
Inputs Required
Before starting, collect:
- Program description: What the program does — activities, delivery model, duration
- Target population: Who is served, size, and key characteristics
- Intended impact: The ultimate change the program exists to create
- Program stage: New / piloting / scaling / mature
- Resources for measurement: Staff time, budget, existing data systems or tools
- Reporting obligations: Funder requirements, board expectations, or regulatory needs
- Timeframe: When results are needed and the length of the funding/program cycle
If program stage or measurement resources are missing, assume an early-stage program with limited dedicated M&E capacity and note the assumption — it changes how ambitious the evaluation design should be.
Instructions
Step 1: Construct the Theory of Change
Build the causal chain that everything else measures against. Express it as a linked sequence:
- Inputs → resources invested (staff, funding, materials)
- Activities → what the program does with those inputs
- Outputs → direct, countable products of activities (people served, sessions delivered)
- Outcomes → short- and medium-term changes in the population (knowledge, behavior, status)
- Impact → the long-term change the program exists to create
State the 2–3 core assumptions the chain depends on (e.g., "participants will apply skills after training") and 1–2 external factors outside program control that could interrupt the chain. A theory of change with no stated assumptions is not testable.
Step 2: Design the Tiered Indicator Set
For each level of the chain, define indicators. Aim for 12–18 total — enough to tell the story, few enough to actually collect.
Output indicators (4–6) — activity completion and reach
- Format each as: indicator name, definition, unit of measure
- These are the "did we do it" metrics — necessary but never sufficient
Outcome indicators (5–8) — the change in the population, split into short-term and medium-term
- For each: indicator name, what change it captures, direction of expected change, and how it connects to a specific outcome in the theory of change
- Prefer indicators that measure change in the person, not satisfaction with the program
Impact indicators (2–3) — long-term, population-level change
- Acknowledge attribution limits; specify whether the program claims contribution or attribution
- Where possible, reference an existing external benchmark or population baseline
For every indicator, mark baseline required (yes/no) and a target (or note that targets should be set after baseline).
Step 3: Define Data Collection Methods
For each indicator or logical group of indicators, specify:
- Method: survey, administrative records, interviews, focus groups, observation, or existing secondary data
- Source: who or what provides the data
- Cadence: baseline / midline / endline, or continuous / monthly / quarterly / annual
- Instrument owner: who is responsible for collecting it
- Disaggregation: which demographic cuts matter for equity analysis (e.g., gender, age, geography)
Balance rigor against capacity: recommend the lightest method that credibly answers the question. Flag any indicator whose collection method exceeds the stated measurement resources.
Step 4: Select the Evaluation Design
Match the design to program stage and resources:
- Pre/post single group — feasible for most early-stage programs; measures change but not attribution
- Comparison group (quasi-experimental) — matched non-participant group; stronger causal claim, moderate cost
- Randomized (RCT) — strongest attribution; only recommend when stakes, scale, and budget justify it
- Developmental/formative — for innovating programs where the model is still changing; prioritizes learning over proof
State the chosen design, why it fits, what it can and cannot claim, and the one biggest threat to validity for this specific program.
Step 5: Build the Learning Loops
Turn data into decisions. Define:
- Review cadence: how often the team reviews data (e.g., monthly ops review, quarterly outcome review, annual strategy review)
- Decision rights: who acts on what — which findings the program manager owns vs. the leadership team
- Adaptation triggers: specific thresholds that prompt a program change (e.g., "if outcome indicator falls below X, review delivery model")
- Feedback to participants: how those served hear results and shape the response
A learning loop is only real if a data point maps to a named decision-maker and a possible action.
Step 6: Set the Reporting Rhythm
Specify what gets reported, to whom, and when:
- Internal team: operational dashboard, frequency
- Leadership/board: outcome summary, frequency
- Funders: aligned to their reporting schedule and required indicators
- Note which indicators from Step 2 map to each audience so nothing is collected without a use.
Output Format
- Structure: Six labeled sections following the steps; theory of change as a linked chain; indicator set as a table with tier, definition, unit, baseline flag, and target; data collection and reporting as tables
- Length: 1,100–1,500 words
- Format: Structured document with tables for the indicator set and data collection plan; prose for theory of change assumptions and evaluation rationale
- Sections: Theory of Change → Tiered Indicator Set → Data Collection Plan → Evaluation Design → Learning Loops → Reporting Rhythm
- Tone: MEL director — rigorous but pragmatic, honest about attribution limits, biased toward what the team can actually sustain
Quality Checklist
Before delivering output, verify:
- [ ] The theory of change includes stated assumptions and external factors — not just a linear chain
- [ ] The indicator set has 12–18 indicators spanning all three tiers (output, outcome, impact) with correct counts
- [ ] Every outcome indicator connects to a specific element of the theory of change
- [ ] Each indicator has a defined collection method, source, and cadence — no indicator without a plan to collect it
- [ ] The evaluation design states what it can and cannot claim, plus its biggest validity threat
- [ ] Each learning loop maps a data point to a named decision-maker and a possible action
Notes
- The most common mistake is over-indexing on output indicators because they're easy to count — push at least half the indicator set into outcomes and impact
- Match evaluation ambition to resources: recommending an RCT to a two-person program with no M&E budget is a design failure, not rigor
- If baseline data doesn't exist yet, make establishing it the first recommended action — targets set without a baseline are guesses
- Disaggregation by equity-relevant characteristics should be built in from the start; it's costly to retrofit
How to use this skill
This file is a Claude Code skill in Anthropic's standard format — markdown with YAML frontmatter at the top.
Installing in Claude Code:
~/.claude/skills/<folder-name>/SKILL.md— auto-loads in every session anywhere on your machine (recommended for skills you'll use across multiple projects)<your-project>/.claude/skills/<folder-name>/SKILL.md— auto-loads only when working in that project (recommended for project-specific skills)
Restart Claude Code or open a new session to register.
Using in Claude.ai, ChatGPT, Gemini, or other AI tools:
- Drag this file into a new chat, or copy and paste its contents
- Send your task or request as your next message — the AI will apply the skill where it fits
For Claude Code skill details: https://docs.anthropic.com/en/docs/claude-code/skills