How to Use AI for Indicator Development
Good indicators are specific, collectible, and decision-linked. AI can generate dozens in seconds, but most will be generic unless you constrain the prompt with your actual project context and data collection capacity.
The best indicator is not the most technically rigorous one. It is the one your team can actually collect, that changes when something meaningful happens, and that someone will act on when they see the number.
The 4-Step Indicator Development Workflow
Move from a results framework to a complete indicator set with definitions, reference sheets, and quality checks. Each step narrows the list from "possible" to "useful."
Generate Candidates
Feed your results framework to AI and ask for 3-5 candidate indicators per result. Specify your donor (for standard indicators), sector (for technical norms), and disaggregation requirements. Cast a wide net at this stage.
Apply the Decision Test
For each candidate, ask AI: "What specific decision will this indicator inform? Who will make that decision? When do they need the data?" If AI cannot answer these questions convincingly, the indicator is not decision-linked. Cut it.
Write Reference Sheets
For each surviving indicator, ask AI to draft an Indicator Reference Sheet (IRS): precise definition, unit of measurement, data source, collection method, frequency, disaggregation, baseline value, and targets. This is where vague indicators get exposed.
Stress-Test for Collectibility
Ask AI to evaluate each indicator against your actual data collection capacity: "Given 2 M&E officers, 5 districts, and KoboToolbox as the primary tool, which of these indicators would require unrealistic data collection effort?" Trim the final set to what you can sustain.
Weak vs. Strong AI-Assisted Indicator Development
The difference between useful and useless indicators almost always comes down to specificity in the prompt.
Generating Candidates
Prompt: "Give me indicators for a food security project." You get a laundry list: Food Consumption Score, Household Dietary Diversity Score, Coping Strategy Index, prevalence of stunting, wasting, underweight, income levels, crop yields, market prices. Most require specialized surveys your team cannot run.
Generating Candidates
Prompt: "Propose 3 output indicators and 2 outcome indicators for a food security project that distributes drought-resistant seeds to 2,000 smallholder farmers in Malawi. Indicators must be collectible through quarterly field visits by 2 agricultural extension workers. Donor is USAID/BHA." You get a targeted, feasible set.
Indicator Definition
Indicator: "Number of beneficiaries trained." No definition of what counts as "trained" (attended vs. completed vs. passed assessment), no specification of training topic, no disaggregation.
Indicator Definition
Indicator: "Number of smallholder farmers (disaggregated by sex) who complete the full 3-day post-harvest storage training and score 70% or above on the post-training assessment." Precise, measurable, and verifiable.
Decision Linkage
The indicator exists because the donor template has a row for it. Nobody on the project team can explain what they would do differently if the number went up or down.
Decision Linkage
The indicator is tied to a specific decision: "If adoption rate of improved seeds falls below 40% at mid-term, the team will shift from group demonstrations to lead farmer follow-up visits." The indicator triggers action.
5 Rules for AI-Assisted Indicator Development
Start with donor standard indicators
Most donors have required or recommended indicators. USAID has standard foreign assistance indicators. WHO has core health indicators. Ask AI to list the required indicators for your donor and sector first, then add custom indicators only where gaps remain.
Fewer indicators, better data
AI will happily generate 30 indicators if you let it. A good MEL plan for a mid-size project has 15-20. Every indicator you add costs staff time to collect, clean, analyze, and report. Ask AI to rank its suggestions by decision value and cut the bottom third.
Match frequency to decision cycles
Do not collect data monthly if decisions are made quarterly. Do not collect annually if the donor wants quarterly reports. Ask AI to align each indicator frequency with the reporting schedule and decision timeline.
Always write the reference sheet
An indicator without a reference sheet is just a label. The IRS forces you to define exactly what is being measured, how, by whom, and how often. If AI cannot produce a coherent reference sheet, the indicator is not well-defined enough to collect.
Test with real data before committing
Before locking your indicator set, ask AI to generate mock data for each indicator and check: Does the format make sense? Can you visualize it? Would a decision-maker understand the trend? This 10-minute exercise catches problems that months of collection cannot fix.
Indicator Development Prompt
Use this prompt to generate a shortlist of decision-linked indicators from your results framework.
Help me develop indicators for my project's results framework. Results framework: [PASTE YOUR RESULTS FRAMEWORK TABLE HERE] Project context: - Donor: [USAID / FCDO / EU / World Bank / other] - Sector: [e.g., food security, WASH, education] - M&E staffing: [e.g., 1 M&E manager, 2 field monitors] - Primary data collection tool: [e.g., KoboToolbox, ODK, paper forms] - Reporting frequency: [e.g., quarterly to donor, monthly internal] For each result level, please: 1. Propose 2-3 candidate indicators 2. For each indicator, state: what decision it informs and who makes that decision 3. Rate each indicator's collectibility (easy / moderate / difficult) given my staffing 4. Flag any indicators that require a dedicated survey vs. routine monitoring data 5. Recommend which indicators to prioritize if I can only track 15 total Format as a table: Result Level | Indicator | Decision It Informs | Collectibility | Recommended Priority
Develop Stronger Indicators
Use the prompt library to generate indicator reference sheets, and explore the indicator library for sector-specific examples.
Related Quick Guides
How to Use AI to Design an M&E Framework
Build the results framework that your indicators attach to.
Read guideHow to Use AI for Baseline and Endline Analysis
Analyze the data your indicators generate.
Read guideHow to Write AI Prompts That Actually Work for M&E
Master the prompting techniques behind every guide.
Read guide