By Unmesh Sheth, Founder and CEO, Sopact. All fund, company and numerical examples are fictional.

A report should use a defined data cutoff and approved evidence, then adapt the explanation and level of detail to its audience. The donor's priorities change which question you answer. They do not change the facts or remove inconvenient gaps.

Cedar's community donor wants a quarterly update for the fictional Northside service area. Its LP wants an annual account of capital deployment, portfolio progress, risks and evidence quality. The board needs actions and company support priorities. Nina should be able to prepare all three from approved evidence without asking each company to complete three overlapping returns.

Write the report brief before the narrative

Name the audience, decision, period, included companies, required measures, allowed detail, cutoff and approver. A donor request is an input to that brief. Distinguish a required disclosure from a preferred style learned from an earlier report.

Historical reports can help identify structure and recurring questions. They are not current evidence, and uploading them does not itself establish that a model has been trained or that their old totals should be reused.

Save each report as its own record with its request, sources, calculation explanations, draft, approval and released version. If MarketLink submits later, the current dashboard may change after approval. The earlier board report should remain reproducible.

Save each donor or LP's reporting requirements once

Create a reporting profile for each audience, separate from the company record. Capture the donor or LP ID, agreement, purpose, included investments, eligible places, required measures, framework edition where relevant, fiscal calendar, cutoff, currency basis, format, permitted detail, approver and delivery schedule. A donor's mailing address does not define the area in which results qualify.

Use an earlier accepted report to propose headings and presentation preferences. Use the current agreement to confirm requirements. Retain a version of the profile so a changed template or donor request does not silently change earlier reports.

These are illustrative starting points. The actual agreement determines the required measures and disclosures. The course does not substitute for an LP's accounting or regulatory reporting requirements.

When the reporting profile requires external metric codes, use the validated mappings from the IRIS+ standards reference. Keep the required edition and any limitations with the profile.

Generate the report from an approved snapshot

At the reporting cutoff, select the saved audience profile and its eligible companies, investments, places and observation periods. Assemble the approved values and calculation explanations, then show which requirements are satisfied, missing or awaiting review. Generate the narrative and tables from that snapshot with evidence references attached.

Personalize the selected evidence, geographic scope, questions, structure and level of detail. Changing the recipient's name on a generic report does not do that work. Keep the numerical claim unchanged when two audiences use the same definition and scope, and explain why it differs when their scopes differ.

A donor may fund only part of a company's work. Its report can describe relevant supported outcomes using an agreed contribution convention. Do not allocate people reached or outcomes in proportion to funding unless a documented method and evidence justify that allocation. Financial allocation and outcome attribution are separate questions.

If the requested local breakdown is unavailable, the draft should state the gap and route it to the relevant owner. Reusing a valid company total is helpful for a company-wide report; it cannot satisfy an unsupported place-based claim.

Build the annual view without adding the wrong quarterly values

An annual report reuses approved evidence, but its calculations depend on the measure. Sum period flows only when their definitions and periods are compatible. Use the appropriate dated balance for a stock measure. Recalculate a rate from its eligible numerator and denominator rather than averaging quarterly percentages by default. Check currency and fiscal-period rules for financial values.

Quarterly reach also needs care. In a fictional example, a company reaches 40 people in one quarter and 50 in the next. Ten appear in both periods. If lawful, reliable matching establishes that overlap, the two-quarter unique count is 80. Without it, the defensible statement is a sum of 90 quarterly counts with cross-period overlap unknown. Four quarterly reports cannot by themselves establish an annual unique-person total.

Annual-only measures should appear when due, with their own source and period. A six-month outcome follow-up should use the appropriate cohort, not be forced into each calendar quarter. Record late submissions and corrections according to the report's cutoff and restatement rules.

Choose the views the decision needs

If a CRM geography field contains several locations, normalize it into separate location relationships without duplicating the company's entire result into each place. Allocate an outcome to locations only when the source supports that allocation. A map can otherwise show presence without claiming local outcome totals.

Calendar year, a donor's fiscal year and all-time views must use explicit date ranges. Maintain deployment dates, source periods and approval dates separately so the selected calendar does not quietly change what an observation means.

Write claims with their boundaries attached

Cedar's first board brief can say:

Two of three expected companies supplied approved Q2 reach observations. Their organization-level counts sum to 200 people reached under the agreed service definition. Cross-company overlap is unknown, so this is not a unique-person portfolio total. MarketLink's return remains outstanding. Its relationship owner will follow up on the agreed date; the impact lead will review any subsequent submission before updating the portfolio view.

The claim register names the RouteWorks and Harbor Homes observation IDs, mapping versions, calculation, missing record and reviewer. If the board needs company-specific outcomes, present them separately rather than converting the reach total into an impact-depth claim.

For donor reporting, apply the donor's actual geographic and funding scope. A multi-donor company does not justify giving each donor the full outcome as exclusively attributable impact. Record the reporting convention and distinguish contribution from attribution.

Use AI to draft within the approved scope

Draft a brief for [audience and decision] from this approved snapshot and claim register. Include the reporting boundary, result, coverage, unresolved exceptions and next action. Cite the observation and calculation references for each material claim. Preserve the limits on uniqueness and attribution. List unmet requirements separately. Do not add missing results, reuse historical totals or disclose restricted source material.

Review the numbers, quotations, explanation and linked files. A company story can illustrate how a service works; it cannot establish prevalence across the portfolio. Use only appropriately permitted stories and quotations.

An approved assistant connection can make governed evidence available to tools such as ChatGPT or Claude. Test the actual connector's scope and permissions before relying on it. A successful query does not establish that every source link or export respects the intended audience boundary.

Test impact reporting software against the released report

The impact lead approves impact claims and rollup explanations; the finance reviewer approves financial interpretations; the authorized report owner releases the final packet. On a small team, roles may overlap, but the responsibilities should remain explicit.

Keep the report snapshot, dataset version and explanation together. When a correction changes a material claim, issue a new version with the reason and the appropriate notification under the fund's process. Do not silently replace the evidence behind an earlier decision.

Build this part of your plan

Prepare a quarterly place-based donor brief and an annual LP report plan from Cedar's approved evidence. Save the two audience profiles. Identify which Q2 observations are reusable, which annual measures are not yet due, and which geographic breakdowns remain missing. Add MarketLink's absent return where relevant. Then simulate a correction to RouteWorks and explain which report versions change.

Check your reasoning: the donor's address is not used as impact geography, quarterly counts are not mislabeled as annual unique reach, and the old report remains reconstructable. A new report uses the corrected approved evidence with its history. Neither audience receives a stronger claim than the evidence supports.

Test a tool: ask for the same numerical claim from a fixed approved snapshot twice, then open its included records and calculation. The number should reproduce. Different narrative wording is acceptable; different scope or arithmetic needs investigation.

Next: Act on exceptions and test the whole cycle.