How to measure the impact of a nonprofit programme, step by step

What are the steps to measure a programme's impact?

Impact measurement sounds like a job for economists. For most nonprofits it is a disciplined version of something you already do: asking whether the programme is working, for whom, and what to change. Here are the eight steps in order. Each one is small; together they give you evidence you can defend.

• 1. Write down why you are measuring and for whom: learning, participants, funders, the wider field.
• 2. Build a theory of change with staff and participants: activities, outputs, outcomes, long-term change and the assumptions in between.
• 3. Choose six to twelve indicators, each with one written definition, a baseline and a target.
• 4. Plan collection: methods, who you ask, how many, and the exact questions, translated and tested.
• 5. Get informed consent, follow your safeguarding policy and protect personal data.
• 6. Check data quality, calculate change and disaggregate by group.
• 7. Weigh other explanations and what would have happened anyway: contribution, not attribution.
• 8. Hold a learning review, change what needs changing, then report honestly with limits stated.

What is the difference between outputs, outcomes and impact?

Outputs are what you deliver: sessions run, kits handed out, people trained. You control them, and they are easy to count. Outcomes are what changes for people because of what you delivered: skills, confidence, behaviour, earnings, access to services. You influence them, but family, the economy and other programmes matter too. Impact is the long-term change in people's lives or conditions, such as steadier household income or more girls finishing school, which many organisations and forces contribute to.

The most common mistake in nonprofit reporting is presenting outputs as impact: '500 young people trained' is an output. It might be the start of an impact story, but on its own it says nothing about whether anyone's life changed.

A theory of change links the three. The evaluator Carol Weiss popularised the term in 1995 in work for the Aspen Institute's Roundtable on Community Change, arguing that programmes are hard to evaluate when the assumptions linking early steps to long-term goals are never written down. Build yours with frontline staff and participants, working backwards from the long-term change, and write the assumptions under every arrow.

How do you choose good indicators?

An indicator is a specific, observable measure that shows whether a change is happening. Test each candidate with plain-words SMART questions: is it specific about who and what counts, measurable with the tools you have, achievable to collect without burdening people, relevant to the outcome, and time-bound? Then ask the question that matters most: if this number moved, what would we do differently? If the answer is nothing, drop it.

Standard indicator sets can save work and let you compare with others. IRIS+, maintained by the Global Impact Investing Network as a free public resource, is one widely used catalogue of metrics and core metric sets; many funders and governments publish their own. Use a standard indicator only when its definition genuinely fits your participants, and say how you adapted it.

Every indicator needs a baseline, the value before you start, or a result means little. Set targets from evidence such as past cohorts, and pair outcome targets with reach targets so the easiest participants are not quietly favoured.

How many people do you need to survey?

Fewer than most people think, as long as they are chosen fairly and the same questions are asked before and after. The standard formula for estimating a percentage, often credited to the statistician William Cochran, needs your confidence level, the margin of error you can accept and the proportion you expect. With a small-population correction, a group of 600 needs about 235 responses for a margin of plus or minus 5 points at 95% confidence. Divide by your expected response rate to see how many to invite.

For small groups, the same formula often says survey everyone. A cohort of 60 needs about 53 responses at those settings, so a census makes more sense than a sample. For interviews and focus groups, choose people on purpose, for example two from each site and some who dropped out: those samples explain numbers but cannot produce percentages for everyone.

Survey design is a skill worth practising. If you want to check yours, AssessAll's Survey and Questionnaire Design Assessment is one way to see where you stand.

How do you know your programme caused the change?

Usually you do not, and that is fine if you say so. Attribution means showing your programme caused a change, normally by comparing participants with a similar group who did not take part. Contribution means showing your programme plausibly played a part, by evidencing each link in your theory of change and weighing the other explanations. Most small organisations can make strong contribution claims and should rarely claim attribution.

A useful middle step is to estimate deadweight: the share of people who would have improved anyway. You can take it from a comparison group, a waiting list, past cohorts or published studies, and you must say where it came from. Subtract it before you claim credit or calculate value for money.

Many funders frame evaluation questions around the OECD Development Assistance Committee's evaluation criteria, revised in 2019 to add coherence alongside relevance, effectiveness, efficiency, impact and sustainability. Use them as prompts for good questions, not as a form to fill in every time.

Worked example: from survey results to cost per additional outcome

Here is a small, realistic calculation for an illustrative after-school reading programme. It enrolled 120 children, 96 completed, and 88 completers were assessed at the end. Of those, 55 moved up at least one reading level. Last year's waiting-list group suggests about 20% of similar children move up a level over the same period without the programme. The programme cost 36,000 (in any currency), including a fair share of shared costs.

• The honest claim: 'Of the 88 completers we assessed, 55 moved up a reading level. Allowing for children who would have progressed anyway, we estimate about 41 additional children improved, at roughly 882 per additional outcome.'
• The assumptions to state: completers who were not assessed are like those who were; last year's waiting list is a fair comparison.

What are the common mistakes that inflate impact?

Inflation is rarely dishonest. It creeps in through counting habits and hopeful wording. Watch for these, and run every draft report against them.

• Counting attendances or rows instead of unique people, or adding the same people across years and partners.
• Calling outputs outcomes.
• Reporting only completers as if they were everyone who started.
• Percentages of tiny groups, such as '100% of participants with a disability' when there were two.
• Before-and-after change presented as caused by the programme, with no word about what else changed.
• Choosing only the best quotes and stories.
• Changing survey questions between baseline and endline.
• Ignoring who did not respond.
• Quoting a social return on investment ratio without its proxies and adjustments. Social Value International's principles for social value include one that says simply: do not overclaim.

How should you use stories and participant voice?

Stories explain numbers and reveal changes you did not think to measure. Collect them with consent, code them to a short agreed list of themes, and read the quotes behind any count before you conclude anything. Include stories that complicate the main pattern; they make findings more credible.

Participatory methods put participants in charge of naming change. In the most significant change method, developed by Rick Davies in the 1990s and later promoted widely by Jess Dart, people tell stories of the most important change they have seen, and panels choose the most significant and explain why. Outcome harvesting, developed by Ricardo Wilson-Grau and colleagues, collects evidence of what changed among partners or systems and works back to whether and how the programme contributed.

Use stories and photos only with recorded consent, written with the person, and in ways that show people as capable rather than helpless.

The same care applies to every piece of data. Tell people, in their language, what you are collecting and why, that taking part is voluntary and will not affect the support they receive, and that they can skip questions or stop. Give each participant an ID, keep names in a separate locked file, combine small groups before reporting, and never paste personal details into public AI tools. Data protection laws differ by country, so check yours, and involve your safeguarding lead whenever children or adults at risk take part.

What goes into an honest impact report?

Open with the year in two minutes: reach, completion, the main outcome, one thing you learned and changed, and cost per participant. Then who you reached, broken down by group with small groups combined; what changed, with each result's baseline and limit in the same paragraph; one consented case study; what did not work and what you changed; how you know and what you do not know; and money and value, with assumptions in words.

Before you publish, hold a learning review with frontline staff and, where possible, participants. Look at the numbers first, then the voices, then the differences between groups, and decide what to continue, adapt or stop. A report that says what you changed because of the evidence is far more convincing than one that only lists successes.

A clear structure saves hours. The Monitoring, Evaluation and Impact Report Kit from Bodhih Training includes a brand-ready impact report template, a MEL workbook that feeds it, and a checklist of mistakes that inflate impact. Writing a concise summary is a skill too: if you want to build it deliberately, a Jobulary development plan can turn it into small weekly practice.