Blog · Programmes
Impact reporting for funders: numbers you can defend
A defensible impact number is one where you can show, on request, which records it was counted from, what rule was used to count them and when each record was last checked. Reporting to funders goes wrong less often because results are poor than because nobody can reconstruct how a figure was reached.
Outputs, outcomes and impact
These three words are often mixed up, and funders notice.
- Outputs are what the programme did. Workshops held, teams admitted, hours of mentoring delivered. You control these and can count them exactly.
- Outcomes are what changed for participants afterwards. Companies formed, funding raised, people employed, products launched.
- Impact is the wider, longer-term difference that can fairly be linked to the programme. It is the hardest to show.
Many organisations set this out as a logic model, sometimes called a theory of change. It runs from inputs to activities to outputs to outcomes to impact. Writing one down, even on a single page, forces you to say how you think each step leads to the next. It also tells you what to measure at each stage.
A report that lists only outputs says you were busy. A report that claims impact without evidence invites doubt. Most of the useful work sits in the middle, with outcomes that are clearly defined and carefully evidenced.
Five things every number needs
- A definition. What exactly is being counted? "Jobs" might mean full-time employees at a given date, or everyone who was ever paid.
- A counting rule. Which records are included and which are left out? From which cohorts? Over what period?
- Evidence per record. For each item in the count, what shows it is true and where is that stored?
- A date. When was each item last confirmed? A total built from checks of very different ages should say so.
- An owner. Who can explain the figure if the funder rings up?
If a number has all five, a new member of staff can rebuild it from scratch and reach the same answer. That is the practical test. The habit behind point three is covered in data provenance: keeping receipts for every record.
Where reports usually go wrong
Double counting. A team that went through two of your programmes is counted twice. A funding round is counted once when announced and again when it closes. Two partner organisations both claim the same company. Stable identifiers for each team, and a rule for shared credit, prevent this.
Shifting definitions. Last year "active" meant trading. This year it means registered. The trend line now shows nothing real. If a definition must change, restate the earlier years or mark the break clearly.
Mixing evidence grades. Some figures come from official records and others from a founder's reply to a survey. Both are legitimate. Adding them together without saying so is not. Report them apart, or label the total.
A missing denominator. "Forty alumni companies are trading" means little without "out of how many?" Always start from the full list of participants. Tracking startup outcomes explains why the teams you do not hear from matter most.
Silent gaps. If a third of teams did not reply and could not be found, say so. An honest "unknown" column protects you. A total that quietly treats unknown as zero, or worse, as success, does not.
Stale data carried forward. A status confirmed long ago and never rechecked is not a current fact.
Attribution and contribution
The hardest question in impact reporting is "would this have happened anyway?" Evaluators call the answer the counterfactual. A team might have raised money without your programme. Strong teams are more likely to be selected, so good outcomes partly reflect who you chose.
Proving attribution, meaning that the programme caused the result, usually needs a comparison group and a careful study design. Most programme offices do not have these. The honest alternative is to talk about contribution. Describe what the programme provided, what happened next and what participants themselves say about the link, without claiming sole credit. "Alumni companies have raised" is a statement of fact. "We generated" is a claim about cause.
Weak and defensible wording
| Weak | Defensible |
|---|---|
| "Our alumni have created hundreds of jobs" | "Of all teams in these cohorts, this many confirmed headcount at the last check. They reported this many roles in total. The status of the rest is unknown." |
| "Most of our companies are still going" | "This share of teams were confirmed as operating at the last check, this share were confirmed closed and this share could not be confirmed." |
| "We helped teams raise significant funding" | "Alumni teams have raised this amount since taking part, from published announcements and founder reports, listed per team in the annex." |
| "Our programme drives regional growth" | "These alumni companies are registered in the region. We have not measured the wider economic effect." |
The defensible versions are longer and less exciting. They are also the ones that survive a follow-up question.
Build the evidence file as you go
The expensive way to report is to rebuild everything in the fortnight before the deadline. The cheaper way is to record evidence at the moment you find it, all year round.
- Keep one record per participant with a stable identifier.
- Log each status check with its date, source and the person who made it.
- Store a copy of, or a link to, the evidence.
- Have a person review anything uncertain before it enters a total. See human in the loop.
- Keep every earlier report and the data behind it, so that you can explain why a figure moved.
With this in place, a report becomes a query over records you already trust. If you use AI tools to help draft the text, hold them to the same standard. Every figure in the draft should trace back to a stored record, as why AI answers need sources explains.
Read the grant terms first
Many funders set their own definitions, reporting periods and templates. Those come first. Where your internal definition differs from the funder's, keep both and label them. This guide explains general good practice and is not legal or financial advice. For what your agreement requires, check the grant terms and ask the funder or your own adviser.
Where Prism fits
Prism Labs runs a live custom pipeline that tracks 51 real startup teams after their programme ended. How that pipeline works is private. In general terms, the engine Prism builds keeps a receipt for every record, has a person approve or reject each finding, and refuses AI answers that do not cite stored evidence. There is more on the innovation programmes page.