Purpose
Turn scattered data collection and shelf-bound evaluation reports into working feedback loops — mapping where insight enters the organization, where it stalls, and who acts on it — so that findings reliably change programs and strategy. Solves the "we measure but never adapt" problem that quietly undermines most nonprofit MEL efforts.
When to Use
Invoke this skill when the user asks to:
- Fix the gap between collecting data and actually using it
- Build a continuous learning or adaptive management practice into operations
- Diagnose why evaluation reports and dashboards aren't driving decisions
- Design feedback loops between program teams, leadership, and the board
- Establish learning routines, review cadences, or reflection cycles
- Prepare a learning system to satisfy a funder's adaptive-management expectations
Inputs Required
Before starting, collect:
- Program area(s): The program(s) or strategy this learning system covers
- What data is collected now: Types of data, tools used, frequency (surveys, service records, evaluations, dashboards, etc.)
- Existing review moments: Any current meetings, reports, or check-ins where data is discussed (staff meetings, board reports, funder reports)
- Decisions that need informing: The specific program or strategic decisions the org wants data to shape
- Roles involved: Who touches data — program staff, MEL lead, ED, board, funders
- Known friction points: Where the user senses data gets stuck or ignored
- Decision authority: Who can actually change a program based on findings
If review moments or decision authority are unclear, map the loops against typical program/leadership/board tiers and flag the assumption.
Instructions
Step 1: Inventory the Current Data Flow
Trace each data source from collection to (attempted) use. For every source, capture:
- What's collected and how often
- Who sees it and in what form (raw, dashboard, report, verbal)
- Where it goes next — or where it stops
- Label each source's current fate: Acted On, Reviewed But Not Acted On, Collected But Never Reviewed, or Collected For Compliance Only
This inventory becomes the evidence base for every recommendation. Do not skip sources that seem minor — the "collected for compliance only" pile is usually where the biggest waste sits.
Step 2: Map the Three Learning Loops
Every functioning learning organization runs feedback at three altitudes. Diagnose each:
Loop 1 — Operational Loop (weeks) Program staff adjusting delivery based on real-time signals (service data, participant feedback, frontline observation).
- Cadence: weekly to monthly
- Question it answers: "Is the program running as intended, and what needs adjusting now?"
- Owner: program/team leads
Loop 2 — Program Learning Loop (quarters) Program leadership reviewing outcome patterns and evaluation findings to refine program design.
- Cadence: quarterly
- Question it answers: "Is the program working, and how should we change the model?"
- Owner: MEL lead + program director
Loop 3 — Strategic Loop (annual/multi-year) Leadership and board using aggregated evidence to test the theory of change and reallocate resources.
- Cadence: annual, with a mid-year checkpoint
- Question it answers: "Is our strategy sound, and where should we invest, cut, or pivot?"
- Owner: ED + board
For each loop, state whether it currently Exists and Works, Exists but Stalls, or Is Missing — with the specific evidence from Step 1.
Step 3: Diagnose the Breakdowns
For each stalled or missing loop, identify which of the common failure modes applies:
- No forum: Data exists but there's no scheduled moment to discuss it
- Wrong altitude: Board reviews operational detail; frontline staff never see outcome data
- No decision rights: The people who see the data can't change the program
- Format friction: Findings arrive as 40-page reports no one reads instead of one-page decision briefs
- No follow-through: Decisions get made in the room but nothing tracks whether they happened
- Compliance orientation: Data collected to report up to funders, never to inform inward to the program
Name the dominant failure mode for each broken loop — most orgs have two or three, not all six.
Step 4: Redesign Each Loop
For every loop, specify the working design:
- Forum: The named recurring meeting or moment where data is reviewed
- Inputs: Exactly which data/findings feed this forum, in what format
- Cadence: How often
- Owner: Who convenes and prepares the inputs
- Decision rights: Who in the forum can authorize a change
- Output artifact: What the forum produces (a decision log, an adaptation memo, a reallocation)
Convert heavy reporting into decision-ready formats: one-page briefs framed around "What we learned → What it means → What we're deciding."
Step 5: Close the Follow-Through Gap
Design the mechanism that tracks whether decisions turn into action:
- A decision log capturing: date, what was decided, who owns it, by when
- A standing agenda item at the next forum to review the prior period's decisions
- A rule for surfacing findings that contradict current strategy so they aren't buried
Step 6: Sequence the Rollout (First 90 Days)
Not every loop can start at once. Provide a phased plan:
- Days 1–30: Establish the single highest-leverage loop (usually the Program Learning Loop) with its forum and one-page brief format
- Days 31–60: Stand up the decision log and the operational loop's lightweight review
- Days 61–90: Add the strategic loop checkpoint and the board-facing learning summary
Each phase names what to launch, who owns it, and the first artifact it produces.
Output Format
- Structure: Data flow inventory (table) → three-loop diagnosis → breakdown findings → redesigned loop specs → follow-through mechanism → 90-day rollout
- Length: 1,100–1,500 words
- Format: Mixed — inventory and loop specs as tables; diagnosis and rollout as structured bullets; brief connective prose between sections
- Sections: Current Data Flow Inventory → The Three Learning Loops (diagnosis) → Breakdown Diagnosis → Redesigned Loops → Follow-Through Mechanism → 90-Day Rollout
- Tone: Foundation strategy lead — practical, systems-minded, allergic to MEL jargon and dashboards-for-their-own-sake
Quality Checklist
Before delivering output, verify:
- [ ] Every data source in the inventory is labeled with one of the four fates (Acted On / Reviewed Not Acted / Never Reviewed / Compliance Only)
- [ ] All three loops are diagnosed with a status and specific supporting evidence — not generic descriptions
- [ ] Each broken loop names a specific dominant failure mode from the six
- [ ] Every redesigned loop specifies forum, inputs, cadence, owner, decision rights, and output artifact — all six fields
- [ ] The follow-through mechanism includes a concrete decision-log structure, not just "track decisions"
- [ ] The 90-day rollout sequences loops rather than launching everything simultaneously
Notes
- The most common finding is a missing Program Learning Loop: operational data flows and board reports flow, but nothing connects quarterly outcomes to program-design decisions. Start there.
- "Decision rights" is the field organizations most often skip and the one that most often explains why nothing changes — the people in the room can't authorize the adaptation.
- Resist adding new data collection. Most orgs already collect more than they use; the fix is almost always in the loops, not the instruments.
- Funder compliance reporting can double as strategic-loop input if reframed — the data is usually already there, pointed the wrong direction.
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