Impact reporting is the practice; an impact report is the product. Five-stage workflow, five questions every report answers, six design principles, and tooling.
.svg)
Impact reporting is the process of explaining what changed, for whom, why it matters, how the organization knows, and what it will do next. A credible report combines quantitative results, qualitative evidence, documents, and change over time. It distinguishes what the evidence supports from what remains uncertain, and it lets the reader inspect where important figures and conclusions came from.
The issue is not whether AI can write fluent prose. It is whether the team can show the definition, calculation, source record, and supporting voice behind a claim.
Key takeaways
The report is not where the work begins. Start with the decision the report must support, translate that decision into evidence requirements, collect the evidence during delivery, and review the claims before they become polished prose. The report is the final view of that work. This is the practical meaning of embedded impact measurement: measurement stays connected to delivery instead of becoming a separate exercise at the reporting deadline.
A report becomes useful when it answers a decision—not when it contains the largest possible set of indicators.
A foundation board may need to decide whether a portfolio strategy is working across places. A donor may need to verify whether an agreement was delivered. A training-program lead may need to know whether participants reached employment and what prevented others from doing so. Those are different questions, even when they draw from some of the same evidence.
Write one sentence before designing the report:
By the end of this review, [audience] must be able to decide [decision] using evidence about [people, outcome, period, and context].
Then document what is known about the reader’s context without pretending to know their private preferences. Keep direct requirements separate from patterns you have observed and assumptions that still need verification. The Academy lesson on building a sourced funder context profile shows how to retain the exact source, date, owner, and level of confidence behind each observation.
Alignment does not mean replacing the organization’s learning model with every funder’s metric list.
First state what the organization believes it is trying to change, how that change may happen, what it will accept as evidence, and how its core measures are defined. A Theory of Change, Logic Model, Logframe, Results Framework, or another framework can provide the change logic. The important point is to connect that framework to actual evidence rules and everyday decisions. The Organization Evidence Model lesson explains how convictions, frameworks, evidence standards, definitions, and external alignment fit together.
Classify each requested measure as a direct match, a translated view of the same concept, a valid extension, or a real conflict. For example, a funder may ask for a placement count while the organization cares about job quality and 90-day retention. Report the valid count, but retain the measures needed to learn whether the placement lasted and was worthwhile. Use the detailed metric-alignment walkthrough when the organization and funder use different language.
External standards such as SDG, IRIS+, GRI, IMP, or SROI help organize or communicate evidence. They do not decide which outcome matters for participants, make weak data strong, or remove the need for an agreed definition.
For every question the report must answer, specify the evidence, source, owner, timing, and missing-data response before collection begins.
This is the plain-language purpose of a data dictionary: it makes sure the same term means the same thing wherever it is collected, analyzed, or reported. The evidence map adds operational context—where the evidence will appear, who is responsible, and what should happen if it is absent. The Academy provides both a plain-language guide to building a data dictionary and a worked method for turning reporting requirements into collectable evidence.
| Report question | Evidence needed | Source and timing | Owner and missing rule |
|---|---|---|---|
| Who did the program reach? | Eligibility, enrollment, attendance, disaggregation | Enrollment and session records joined by participant ID | Program operations; flag unmatched or duplicate records |
| What changed? | Baseline, follow-up, definition, calculation | Participant record at agreed observation points | MEL lead; show missing follow-up separately |
| Why did the result occur? | Open response, interview, case note, program event | Collected close to the relevant moment | Program lead; do not infer a reason when voice is absent |
| Can the claim be trusted? | Source record, definition version, calculation, review | Stored with the result and reporting period | Evidence owner; downgrade or withhold unsupported claims |
Do this for a small number of decision-critical questions first. A 60-page reporting template can still be mapped, but the team should know which items can change a decision and which merely satisfy an administrative requirement.
A label such as “served,” “placed,” or “retained” is not a definition. Reasonable people can count it differently.
For each important measure, define who is eligible, what event or change qualifies, when it is observed, how it is calculated, which evidence is acceptable, how exceptions are handled, and what decision it should inform. The Academy’s one-definition-for-every-number lesson provides the complete Metric Definition Sheet.
In working conversations with Sopact, Open Play Foundation CEO Marco Botha described a practical problem. Young people could enroll in a program but not appear for the first session. Program data was collected diligently, yet analysis could arrive weeks later. By then, the team had missed the moment to understand the absence and respond.
The useful change was not a larger monitoring template. It was a small set of decision metrics, quicker review, and one neutral question that let young people explain the barrier in their own words.
| Decision | Who needs respectful follow-up, and do repeated barriers require a program change? |
| Who | Eligible young people with confirmed enrollment before the first scheduled session. |
| What | Program-approved presence recorded during the first session. |
| How | Eligible confirmed enrollees attending ÷ all eligible confirmed enrollees × 100. |
| Proof | Dated enrollment and attendance records joined through a persistent participant ID. |
| Exception | “Attendance not recorded” remains separate from “did not attend.” |
Ask:
“What, if anything, made it difficult for you to attend the first session? Please describe what happened in your own words.”
Do not begin with “Were fees the problem?” That inserts an explanation before the participant has spoken. Optional follow-up can explore transport, safety, family responsibilities, financial hardship, communication, timing, or program design after the participant raises relevant context.
Case boundary: this example is based on working conversations and the organization’s public mission. It should not be used to infer the prevalence of a specific barrier or the effect of an intervention without the underlying records.
Clean evidence is not just a valid form. It is evidence captured close enough to the work that the team can respond. The collect-clean-evidence walkthrough shows how to assign every requirement to the moment where it can be observed accurately and used.
For Open Play, a WhatsApp follow-up after a missed first session can ask the neutral question above. If a young person describes financial hardship, staff can follow up to understand the situation and determine an appropriate response. The record should not claim that assistance was offered unless that action is documented.
The same workflow can expose an operational issue that a traditional report would miss. Marco also described field data revealing a water leak. The value was not the survey itself; it was seeing a problem while the team could still act, then recording what happened after the intervention.
For each evidence item, record the natural moment, minimum signal, neutral context question where needed, owner, timestamp, participant or program identity rule, validation rule, action trigger, and follow-up event.
A number shows the size or location of a pattern. Qualitative evidence can explain experience and mechanism. Documents provide obligations and context. Longitudinal records show whether change lasted.
General-purpose AI can summarize a group of files, but it may interpret the same question differently when the prompt, file set, or conversation changes. For reporting, use approved definitions, source records, calculation rules, a qualitative codebook, and version history. AI can help find and organize evidence; fixed rules should calculate governed results; people should approve consequential interpretations. The Academy lesson on stable results from governed data explains this reliability architecture without requiring the reader to understand database queries.
Do not force the evidence to support the strongest possible story. State the strongest claim the current record can actually defend.
Every headline figure should be able to travel backward through a simple trail: report claim → calculation or qualitative theme → approved definition → source records → reviewer decision. The Academy’s trace-every-result lesson provides worked trails for a number, a qualitative finding, attribution, and SROI.
For SROI, this means more than showing a final ratio. Keep the benchmark source, financial proxy and justification, deadweight, attribution, displacement, drop-off, calculation steps, and sensitivity scenarios. A reviewer should be able to move from the headline ratio to the source and assumption behind it.
One governed evidence base can produce different reports without changing the underlying definitions or rewriting the evidence.
The report should make three layers easy to distinguish: the observed result, the interpretation of that result, and the action the organization will take. It should also name limitations in the relevant section rather than burying them at the end.
At Open Play, the operational value continued beyond a single report. Marco described viewing results across roughly ten programs on a shared dashboard. The important point was not the dashboard itself. The team could see progress and emerging problems regularly, respond, and carry the next observation into the following cycle.
The loop closes only when the evidence changes a decision. Record the action, responsible owner, date, expected response, and the follow-up signal that will show whether the action helped.
The method above does not depend on one platform. A team can implement it with documents, spreadsheets, a form tool, a qualitative coding process, and disciplined review. The difference is how much reconciliation, traceability, and repeated setup the team must manage as programs, partners, funders, evidence types, and reporting periods multiply.
| Approach | Works well for | Watch for |
|---|---|---|
| Manual tools | A small program, limited evidence types, and a stable team with strong controls. | Definitions split across files; report assembly and checking become slow. |
| General AI | Exploring language, summarizing supplied text, and drafting structure. | Answers may change with prompts or files; calculations and versions need separate control. |
| Sopact Sense | Connecting structured and unstructured evidence across programs and audiences. | People still approve definitions, access, interpretations, claims, and decisions. |
At minimum, have the decision and audience, approved measure definitions, reporting period, source records, calculation rules, qualitative evidence where explanation is needed, missing-data states, limitations, and an owner for each important claim. If a required item is missing, show the gap rather than inventing a complete narrative.
Use the smallest set that can answer the report’s decisions and obligations. Separate decision-critical measures from administrative fields. A few well-defined measures with context and follow-up are usually more useful than a large indicator list that nobody reviews during delivery.
Impact measurement defines and observes change. Impact reporting communicates the relevant evidence, interpretation, limitations, and action to an audience. Reporting is an output of the measurement and evidence workflow, not a substitute for it.
Yes. Open-ended responses, interviews, case notes, and partner updates can explain experience, context, mechanism, and unintended effects. Keep themes connected to quotations and source records, state how the evidence was selected or coded, and do not use one story to imply prevalence.
Keep missing, not recorded, not applicable, and negative results distinct. When sources conflict, preserve both, describe the disagreement, name the owner of the resolution, and narrow the claim if necessary. A transparent partial answer is stronger than a polished unsupported one.
They can help structure or draft a report from supplied material. They should not be treated as the source of a governed result. Important figures and conclusions still need approved definitions, fixed calculations, evidence boundaries, citations, version control, and human review.
Yes, when the underlying definitions and evidence remain stable. Create audience-specific views that map the governed evidence to each funder’s valid requirements. Do not silently change the meaning of a measure to fit a template.
The formal report may be quarterly or annual, but decision-critical evidence should be reviewed at the pace at which the team can still respond. Attendance may need same-week review; employment retention may require a 90-day observation; a board may review a portfolio quarterly.
