What is impact reporting?
Impact reporting is the recurring practice of telling a funder, board or community what changed for the people a program serves, measured against metrics agreed in advance, with the evidence behind each number and what is still unknown. An impact report is the document one cycle produces; impact reporting is the work that makes it trustworthy.
That work has two sides. A funder, such as a foundation, impact fund or CSR team, adds up reports from the organizations it funds. A funded partner, the grantee or investee, often writes a different report for every funder it has.
THE SHORT VERSION
- Agree the few metrics that matter, and how each one is counted, before the first report is due; a report is only as good as the agreement behind it.
- Collect once into one evidence base with one ID per person, then give each funder its own format without rebuilding the numbers.
- Let AI draft if you like, but give it the sources, definitions and taste, make it flag missing numbers, and open every figure before you send.
Why can so few impact reports be compared?
Because each report is built to a different template with different definitions, so the same word, such as "placed" or "served", counts different things from one report to the next. Usually nobody made a mistake; nobody agreed first.
Funders receive partner reports as PDFs, spreadsheets, survey exports and emails, then stitch a portfolio report from all of them. A funded partner answers a quarterly template, an annual narrative, an IRIS+ metric list and a board deck, rebuilding the same evidence each time.
The fix sits upstream. When funder and partner agree the few metrics that matter at the onboarding call, each with one written definition, the report becomes a read-out instead of a reconstruction.
What should an impact report include?
Seven parts cover most readers: a summary with the decision asked for, the scope, method and definitions, results against the agreed metrics, people's experience, what is missing or held back, and next steps with an owner. Put the finding first and keep the method one step away.
| Section | What the reader needs | Check before sending |
|---|---|---|
| Summary | The main finding, its limits, the decision requested | Does the headline match the evidence? |
| Scope | Who, which period, which programs or partners | Are exclusions stated? |
| Method and definitions | Sources, definitions, collection timing | Could a reviewer recount it? |
| Results | Each agreed metric with its denominator | Same definition and period as agreed? |
| People's experience | Words that explain a number, with permission | One voice, or a pattern? |
| Missing and held | Gaps, differing definitions, when each closes | Is every gap named? |
| Next steps | What changes, who owns it, when you review | Does each action follow a finding? |
Keep full definitions in an appendix, and never expose participant records in a public report. For a copyable outline filled in line by line, use the impact report template. The free guide How to Write an Impact Report covers the writing, and the report examples library shows finished reports and dashboards.
How does impact reporting work, from agreement to report?
It runs as one loop: agree the metrics, define each in a shared dictionary, check every report against the agreement, add up only what matches, write the report, and feed what you learned into next year's agreement. Funders and partners walk it from opposite ends.
Take a fictional regional workforce fund that funds four job-training partners, A to D. Its agreement, written from the onboarding call, names five metrics: participants enrolled (unique people, not visits), completed training, placed in a job within 90 days of exit, retained in the same job at 12 months, and starting wage by track.
| Step | Funder | Funded partner |
|---|---|---|
| Agree | Turn the onboarding call into a few metrics | Confirm each definition before signing |
| Define | One dictionary row per metric | Map each funder's ask to your rows |
| Check | Compare each report with the agreement | Run the same check before sending |
| Add up | Total only what shares a definition | Group the same records each funder's way |
| Report | Portfolio report for the board | One report per funder, in their format |
Partner C's report shows why the check matters. Enrolment matched the agreement: 80 unique people. But placements were counted within six months instead of 90 days, 12-month retention was missing, and starting wage arrived as one average instead of by track.
The fund sent three questions back instead of guessing, then added up only what matched. Partners A, B and D placed 42, 31 and 27 people within 90 days, a total of 100; Partner C's 55 used six months, so it is held beside the total until C confirms the 90-day count. The chapter Check each partner report against the agreement walks through it.
The walkthrough below takes one job-training and placement program through the loop end to end: its theory of change, the IRIS+ metrics that fit it, and the report built from them. Watch how each reported number keeps the definition it was collected under.
How do you report to several funders without rebuilding the numbers?
Collect once into one evidence base, with the same records, the same dictionary and one ID per person, then map each funder's request to a metric you already hold. Each funder gets its own format; none gets its own data collection.
Now look from Partner C's side, the picture at the top of this page. It has four readers: Funder A wants a quarterly template, Funder B an annual narrative, Funder C IRIS+ metrics, and its own board a two-page update. Teams in this position often run four small data systems, one per template; the picture shows the alternative, one evidence base with four layouts on top.
When a funder defines a metric differently, report their number, labelled, with one sentence on how it differs. Funder B asks about work within six months, so Partner C gives Funder B that count and keeps 90 days as its own measure; noted on the map, the six-month figure never reaches the workforce fund.
One ID per person makes outcomes countable: "retained at 12 months" needs the exit record and the follow-up to belong to one person. Girls Inc. of Metropolitan Dallas runs pre, mid and post surveys plus follow-ups at 6 months and 1 year on the same principle. The chapter Map every funder's ask to one evidence base builds the map row by row.
Which reporting mistakes change what a number means?
Three do most of the damage: adding numbers counted under different definitions, changing a definition without marking the break, and treating a gap as a zero. Each produces a clean-looking figure that means something other than it says.
Adding numbers that were counted differently
Add Partner C's 55 to the other three partners and the portfolio shows 155 placements. That derived total is wrong under either window, because it mixes six-month and 90-day counts. Keep the 55 beside the total with its reason, and put the 90-day rule into the reporting template for the next cycle.
Changing a definition quietly
A team tightens its completion rule or changes a denominator. The new rule may be better, but an unmarked year-on-year chart now compares two different things, so record the date and reason, restate earlier years where you can, and mark the break where you cannot.
Treating a gap as a zero
Keep missing, not applicable and zero apart. A financial file that never arrived is a gap to name, not evidence of zero spending.
Can AI write an impact report?
AI can write a good first draft in minutes when it has four things: the sources it may use, your data dictionary, the funder's agreement and the funder's taste file. Without them it fills gaps with plausible guesses, and a guessed number looks exactly like a found one.
The dictionary lets a draft catch its own gaps: if it lists five metrics and the draft finds two, the other three are flagged before anyone reads it. Frame the request fairly, too. "What changed for trainees this quarter, and what did not?" leaves room for an honest answer.
PROMPT · PASTE INTO CLAUDE, CHATGPT OR YOUR AI TOOL
Draft our [PERIOD] impact report for [FUNDER NAME]. Use only these inputs: - the data sources selected for this report: [LIST SURVEYS OR FILES] - our data dictionary: [ATTACH] - our agreement with this funder: [ATTACH] - this funder's taste file and our own: [ATTACH] Rules: 1. Report every metric in the agreement, using our dictionary's definition. If the funder defines one differently, give their number, labelled, with one sentence on the difference. 2. Put the source next to every number. 3. If a number is not in the selected sources, write "MISSING: [metric], not in our data". Do not estimate, round up or borrow from another period. 4. Keep each limitation beside the number it limits, and include shortfalls. 5. Do not invent quotations, causes or explanations. 6. End with a table: claim | number | source | status (found, missing, differs from agreement).
Taste is the part most teams skip. One program officer reads about 40 reports the week before the board and wants the table first. A short taste file per funder, plus one for your own voice, changes a report's words and order, never its numbers.
Drafting really can get short, as Marco Botha of Open Play Foundation says in the quote at the top of this page. The check does not shrink with it; the chapter Write each funder's report with AI, then check it gives the full pre-send checklist.
The video below explains why an accurate report can still go unread. Watch for the difference between a dictionary, which fixes what a number means, and the context around it: who reads the report and what they decide with it.
How do you choose impact reporting tools?
Choose by the work behind the report, not the look of the dashboard: one small, stable report needs only a spreadsheet and a document, while recurring reports across partners need evidence that stays connected. Test each tool on your own workflow.
| Capability to test | Ask to see it done |
|---|---|
| One record per person | Add a follow-up survey to the intake record without retyping names |
| Several kinds of evidence | Take metrics, a narrative and a document for one partner and period |
| Differences from the agreement | Show a partner report whose definition differs, and how you spot it |
| Answers you can trace | Open the record behind a headline figure |
| Scoped AI | Choose which surveys and fields AI may read; keep names out |
| Repeatable figures | Rebuild last quarter's headline figure from the same records |
Count the whole cost, including setup, cleaning imports, review and partner training. A low subscription can still cost a lot of staff time.
Sopact Sense fits recurring reporting across partners: one ID per person from the first form, an AI Assistant that sees only its team's folder and stays locked until you choose its surveys, and answers whose every line links to a record. See the portfolio solution or the separate guide to impact reporting software.
What can an impact report not prove?
A report shows what partners observed and what can honestly be added up; on its own it does not prove that the program caused the change. "100 people placed within 90 days" is a count of outcomes, not evidence that the funding produced those jobs.
Outputs are not outcomes: sessions delivered describe activity; placement and retention describe change. A before-and-after difference needs a comparison group before it supports a causal claim. Wages may be self-reported, and people who stop answering follow-ups may differ from those who reply.
AI makes confident mistakes faster too. People decide what a result means, and they check every AI line against the record before a funder reads it.
Start with one funder report this cycle
Pick the next report that is due and run the whole loop on it once, before changing anything else.
- Write the brief in three lines: who reads it, what they will decide, and the period it covers.
- List every metric you agreed with that funder, with your dictionary's definition beside each one.
- Name the sources the report may draw on, and remove names, emails and case notes from anything you share with AI.
- Draft with the prompt above, count the MISSING flags, and decide how to close or name each gap.
- Open the source of every number and confirm its definition, period and denominator.
- Send it, then add the funder's questions to that funder's taste file.
After the first cycle you have one report you can defend line by line, a list of gaps to fix, and the start of a taste file for your most demanding reader.
Frequently asked questions
What is the difference between impact measurement and impact reporting?
Impact measurement decides what change to track and collects the evidence: metrics, definitions and the records behind them. Impact reporting communicates the results, methods, gaps and next steps to a specific reader, such as a funder or board. Reporting depends on measurement; a polished report cannot replace evidence that was never collected. The two meet in the reporting agreement.
Which framework should an impact report use?
Use the one your reader asks for: a theory of change, logic model, logframe and results framework are four layouts of one change model, not proof. Impact Frontiers' Five Dimensions help you describe each result, and the GIIN's IRIS+ catalog supplies standard metric definitions where a funder wants codes. Check that a code's definition matches yours before attaching it.
Can numbers from different partners be added together?
Only when they share a definition, unit, period and population, and duplicates are handled. Otherwise report them side by side or recount from the source records. In the fictional workforce fund, partners A, B and D counted placements within 90 days, so 42, 31 and 27 add to 100; Partner C's six-month 55 is held beside the total.
How often should you produce an impact report?
Follow the agreement and your readers: often quarterly for progress and annually for outcomes such as 12-month retention. Look at the evidence more often where you can still act on it. Keep each period's approved report separate from any dashboard that keeps updating, and date both, so one figure never appears as two different answers.
Can AI write an impact report?
AI can write the whole first draft when it has the selected sources, your data dictionary, the funder's agreement and a taste file. It should not decide what the numbers mean, which shortfalls matter or whether a quote may be used. Ask it to flag missing numbers instead of estimating, then open the source of every figure before sending.
What should an impact report say when evidence is missing?
Name what is missing, why, and when it will be available. "12-month retention: not yet available; due after the annual follow-up" is a complete answer. Keep missing, not applicable and zero separate, and never fill a gap with an estimate. If two sources disagree, report the number you can trace and name the open question.
