From Outcomes to Indicators: How to Write SMART Indicators Your Data Can Answer
A good indicator states exactly what you’ll count, out of what, how often, from which record, and who is responsible for it. SMART (specific, measurable, achievable, relevant, time-bound) is a fair checklist, but it doesn’t tell you how to write one. This post gives you a recipe, ten rewrites, and the data questions to settle before you build a form. It is the deep dive on Section 8 of our free theory of change template and data model guide.
Why most indicators fail in practice
Indicators usually fail for one of three reasons, and none of them is about wording.
- No denominator. “Number of girls enrolled in Grade 9” can rise while the share of eligible girls falls.
- No source. The indicator is in the logframe, but no form captures it, so someone reconstructs it from WhatsApp messages at report time.
- No repeat measurement. Outcomes are about change, which means you need the same measure at least twice. A system that records only the first contact can report reach but never change.
A five-part recipe
For every outcome in your theory of change, write down:
- What counts: the event or state, defined precisely. “Enrolled in Grade 9” means on the school register by the end of the admission window, verified by a mentor.
- Out of what: the denominator. Girls in Grade 8 at the end of last year, not girls currently in the programme.
- When and how often: annually, after admissions close.
- From which record: the annual status check form, one record per girl per year.
- Who owns it and why: the MEL lead, for the programme lead’s decision on where to focus caregiver work.
Then add disaggregation (sex, age band, disability, location, as relevant) and a baseline if one exists. If you can’t fill in part 4, the indicator isn’t measurable yet. Mark it “no source yet” rather than pretending.
Ten before-and-after rewrites
| Before | After |
|---|---|
| Improved attendance | % of active girls with ≥75% school attendance in the month (days present ÷ school days), from the school register, monthly |
| Girls' learning improves | % of girls who moved up ≥1 level since the previous round ÷ girls assessed in both rounds, each assessment round |
| Increased confidence | % of girls scoring ≥4/5 on the 6-item self-efficacy scale at endline who scored <4 at baseline |
| Parents more supportive | % of caregivers answering "Yes, definitely" to "Will you support your daughter to enrol in Grade 9?", twice a year |
| Number of trainings conducted | Sessions held ÷ sessions scheduled, per centre, weekly (an output, labelled as one) |
| Reduced dropout | Grade 8 → 9 transition: girls enrolled in Grade 9 by end of admissions ÷ girls in Grade 8 at year end |
| Youth employed | % of graduates in paid work at 3 months after course end ÷ graduates reached for follow-up, with non-response reported separately |
| Better health awareness | % of mothers who correctly name 3 of 4 danger signs in pregnancy, at 6-month follow-up |
| Community engagement | Number of village meetings held with ≥10 caregivers present ÷ villages, termly |
| Women empowered | % of women participants who report making decisions on their own income alone or jointly, at baseline and endline, using a standard survey question |
Two patterns stand out. Vague outcomes like “confidence” or “empowerment” need a defined instrument, not an adjective. And some “indicators” are really outputs. That’s fine, as long as they’re labelled as outputs.
Disaggregation: decide now, not at report time
If you’ll need results by sex, age or disability, the fields must exist on the registration form from day one. Adding them later means missing data for everyone already enrolled. For disability, the UK government’s guide to disaggregating programme data by disability recommends the Washington Group short set of questions, which ask about difficulty with functioning rather than a diagnosis. UK Aid Match requires all beneficiary data to be disaggregated by gender and disability at minimum.
The flip side is restraint. Every extra field is personal data you must justify, store and protect. Collect caste, religion or health information only if a named decision depends on it, and record consent. If you work in India, our DPDP Act guide for NGOs explains what the law expects.
Quantitative or qualitative?
Some outcomes resist counting, and forcing a number on them produces a misleading one. “Girls feel safe travelling to school” might be better captured through a short set of structured interviews each term, coded for themes, alongside one countable proxy. The test is the same: what decision will this inform, and what record will hold it? Voice interviews can make this practical at scale; our guide to AI interviews for research covers when they work and when they don’t.
Baselines and targets
Don’t invent targets before you have a baseline. Collect one round of data, then set targets with the people who’ll be held to them. Where a target is about reducing something (the share of girls in the lowest learning level), express it as a maximum, and show “met / not met” plus the gap rather than a ratio, which misleads when the number is small.
Check your indicators against your data
Once each indicator has a source form, look at the records that form will create. Can you calculate the numerator and denominator from them without manual work? If not, change the form, not the indicator. Our worked theory of change example shows the full chain for ten indicators, from definition to the entity and form behind each one.
Your next step: find out whether your current spreadsheets and tools can produce these indicators every month. Take the free 10-minute technology assessment.