Pull inputs, outputs, and outcomes from a grantee report, mark every silent field MISSING, and grade the result green, amber, or red — so you report what's evidenced, not what you guessed.
In short: Don't paraphrase a grantee report into a number you can't defend. Extract inputs, outputs, and outcomes exactly as evidenced, write MISSING wherever the report is silent, and grade each element green, amber, or red — so the gaps are visible instead of quietly filled in.
Start in the dataset you already trust. For this walkthrough we work over DEMO-07 · SROI Investee — a single investee's report with clean records and persistent contact IDs. Load your Decision Brief first so the assistant knows the decision, audience, outcomes, indicators, and evidence standard before it reads a single line of the report.
You are the Sopact Sense Assistant working over the DEMO-07 · SROI Investee dataset (clean data + persistent contact IDs). Load my Decision Brief (decision, audience, outcomes, indicators, evidence standard) first, then wait for my task.
The extraction prompt does one job: pull what is there and flag what isn't.
From [ORG]'s investee report, extract inputs (incl in-kind), outputs, outcomes by stakeholder, indicators, proxies. Mark MISSING where silent. Grade green/amber/red.
Five elements make this reliable: the dataset you point it at, the extraction itself across inputs/outputs/outcomes, the instruction to mark MISSING wherever the report says nothing, the rule to don't guess a value that isn't evidenced, and the grade G/A/R that scores each element's evidence.
Run on the SROI Investee dataset (DEMO 07) already loaded in Sopact Sense.
GRADE: green | Outputs | evidenced and counted; amber | In-kind | named but not quantified; red | Income outcome | no indicator collected
The grade reads at a glance. Green means the element is evidenced — outputs are counted and sourced. Amber means it's named but incomplete — in-kind contributions appear in the narrative but were never costed. Red means it's MISSING — the income outcome has no indicator behind it at all, so reporting a number would be invention, not extraction.
Take the lowest-graded element and fix the one thing that moves it, using only what the program could realistically measure.
Take the lowest-graded element above and fix it using only what the program could realistically measure. Show the before → after grade and the single indicator/edit that moves it to green.
Turn the graded analysis into a branded "missing & incomplete" report, then publish a link anyone can open.
Create a 'missing & incomplete' report from this analysis in Sopact branding [or paste your website URL / brand guideline to apply your own]. List every element graded amber or red, what is missing, and the one input that fixes each. Lead with the decision this report informs.
Create a shareable link for this report and open it in a new tab.
MISSING is a finding, not a failure. A clearly marked gap is more useful to a funder than a confident guess. Name it and the report stays defensible.
Quantify in-kind or it disappears. Donated time and goods routinely go uncosted. Attach an estimated dollar value with a stated method so the input shows up in the value map.
Separate outcome from output. "500 people trained" is an output; "earnings rose" is an outcome. If the report only gives you outputs, the outcome is MISSING — say so.
Re-grade after every fix. Once you add an indicator, ask the assistant to re-score so you can see the element actually move.
Re-grade only the elements I changed since the last extraction and show the before → after grade for each.
Point an AI assistant at the report inside a clean dataset, then ask it to pull inputs, outputs, outcomes by stakeholder, indicators, and proxies — marking MISSING wherever the report is silent rather than guessing. Grading each element green, amber, or red turns the extraction into a readiness check you can act on.
Mark it MISSING and grade it red. Inventing a number breaks the chain of evidence and will not survive a reviewer. The fix is to add the missing indicator at intake and exit so the outcome can be evidenced next cycle.
Treat them as amber: real but incomplete. Attach an estimated dollar value using a stated, conservative method — for example a market rate for donated time — and note the assumption. That moves the input from a narrative mention to a usable figure.
ChatGPT, Claude, and Gemini are fine for a quick test — but not for an answer you'll put in front of a funder or board. When it has to hold up, run it in Sopact Sense.
Try it in Sopact →
