Verifying an NPO's Annual Impact Report: Deduplication, Learning Gains and Pressure
Annapoorna Trust is a Section 8 company registered on the Social Stock Exchange and has raised zero coupon zero principal instruments. You are the empanelled social impact assessor reviewing its annual impact report. The trust reports 18,000 children reached through remedial education across 240 centres, a learning gain of 42%, and 96% attendance. Its management information system is a shared spreadsheet maintained by centre coordinators. You visit 30 centres selected at random. Attendance registers at those centres add up to 11,700 children after removing exact name-and-age duplicates, and several children appear at two centres in the same month. Two centres have no records for three consecutive months. The 42% learning gain is computed on 1,140 children who sat both a baseline and an endline test, and the tests were administered and marked by the same teachers whose performance pay depends on learning outcomes.
1. How do you treat the 18,000 beneficiaries figure in your assessment?
Read the worked answer
First, apply a stated materiality threshold so the treatment is principled rather than improvised - for example, restate any figure where the verified variance exceeds 5% of the reported number.
The verified position is 11,700 unique children across the 30 sampled centres, against enrolment records that would gross up to 18,000. That is a 35% gap, well past any reasonable threshold, and the cause is identifiable: enrolment-level records, no unique child identifier, and double counting where a child attends two centres.
Report it as two figures, not one: "18,000 enrolment records; 11,700 verified unique beneficiaries". State plainly that the 18,000 is a record count, not a beneficiary count, and explain the deduplication method so the reader can reproduce it.
Do not extrapolate the 11,700 from 30 centres to all 240 as if it were a precise population count - the sample supports an estimate with a range. Present the verified figure for sampled centres and a stated estimate with its confidence range for the whole programme, or restrict the headline to what was directly verified.
Finally, make a system recommendation: a unique child ID, monthly reconciliation, and centre-level sign-off. A recurring reporting weakness is a finding in its own right.
2. Can the 42% learning gain be reported as a programme outcome? Explain your reasoning and the conditions under which it could.
Read the worked answer
Not as currently framed. Three problems compound:
- Coverage. 1,140 children is 6% of the reported base and 10% of the verified base. Presenting a 42% gain without stating the denominator invites the reader to apply it to 18,000.
- Who assessed. Teachers whose performance pay depends on learning outcomes administered and marked the tests. That is a direct conflict of interest in measurement; the assessor cannot verify independence.
- Test design. If the endline mirrors the baseline, teaching to the test inflates the gain. Without a comparison group or an external tester, the 42% cannot be attributed to the programme.
It can be reported, with conditions: state the denominator and the selection rule (who sat both tests, and why others did not); disclose that testing was internal and incentivised; and describe the gain as "among children who completed both assessments", not as a programme-wide effect.
To make it a reportable outcome, recommend a random sample of centres tested by an independent assessor, with the endline version not shared with teachers in advance, and a comparison group of children on the waiting list. If the trust cannot do that this cycle, report the 42% as an internal monitoring indicator and exclude it from headline outcomes.
3. The chief executive officer warns that restating the beneficiary count could jeopardise the next tranche of funding, and a board member suggests you "align with management's view". What is your response?
Read the worked answer
Restate, and route the communication through governance rather than the executive.
The assessor's duty runs to the readers of the impact report - donors and instrument holders who cannot inspect the centres themselves - not to the issuer's fundraising calendar. Suppressing a 35% variance because a tranche depends on the number converts the report into a fundraising document, which is precisely what the Social Stock Exchange disclosure regime exists to prevent.
The sequence:
- Put it in writing. Send a short note to the chief executive and the audit committee or board setting out the verified figure, the method, and the proposed disclosure wording. Note the funding pressure explicitly.
- Offer a constructive framing. A trust that discloses a deduplication problem and a remediation plan retains credibility; a trust whose numbers are later found inflated loses it entirely.
- Escalate if obstructed. If the board directs publication of the unverified 18,000 without the reconciliation, record the direction and decline to sign. If the report is published over your objection, withdraw and state the basis for withdrawal in the engagement file.
Do not negotiate the number with a board member. Negotiate only the wording that accurately describes it.