Theme a whole grant round before you score a single application — cluster the proposed work by theme, population, and geography, with counts, quotes, and the priority gaps surfaced.
In short: To analyze a batch of grant applications, theme the whole round before you score any single proposal: cluster the proposed work by theme, population, and geography, count how often each appears, quote real applicant language, and surface the priority gaps. Hold single mentions aside rather than promoting them to themes. Sopact Sense grades the clusters green, amber, or red so you can see what the round is actually asking for — and fund accordingly.
Start with the round's open-text responses loaded as clean data with persistent contact IDs, so every theme ties back to real applicant language. Point the assistant at the dataset and have it read your Decision Brief first — the decision, audience, outcomes, indicators, and evidence standard.
You are the Sopact Sense Assistant working over the DEMO-04 · Open-Text Barriers dataset (clean data + persistent contact IDs). Load my Decision Brief (decision, audience, outcomes, indicators, evidence standard) first, then wait for my task.
The prompt names the dimensions and demands evidence per cluster. Paste this verbatim:
Across round [ROUND], cluster proposed work by theme, population, geography; count + quote; surface priority gaps. Hold single mentions. Grade green/amber/red.
The prompt works because of five elements: the dataset it clusters over, the instruction to cluster the round by theme/population/geography, the demand to surface priority gaps, the discipline to hold single mentions rather than over-theme, and the call to grade green/amber/red so weak clusters are visible.
Run it against the workforce barriers demo:
Run on the Open-Text Barriers dataset (DEMO-04) already loaded in Sopact Sense.
GRADE: green | transport/childcare | recurring; amber | other | ambiguous; red | language | single mention
The green cluster is recurring transport and childcare needs — named often enough to act on. The amber cluster is an ambiguous 'other' bucket mixing several meanings. The red item is a single-mention language barrier — real, but not yet a pattern, so it's held as a watch-item.
Take the weakest cluster and fix it with the one input that resolves it. Sense shows the before → after grade.
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 theming into a report, then a link that opens with no login.
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.
Theme before you score. Clustering the whole round first changes how you fund — you see which needs are widespread before any one polished proposal pulls your attention.
Hold single mentions. One applicant raising a barrier is a signal to watch, not a theme to act on. Keep single mentions in a separate list so they neither vanish nor distort the priorities.
Split the 'other' bucket. A catch-all category hides distinct needs. When 'other' grows ambiguous, ask the model to break it into named sub-themes or add a follow-up field.
Always keep the quote. Counts tell you how often; quotes tell you what people actually meant. Require a representative quote per cluster so reviewers can sanity-check the theme.
Split the 'other' cluster into named sub-themes with a count and a representative quote for each, and re-grade them.
Cluster the whole round at once — group proposed work by theme, population, and geography, count how often each appears, and pull a representative quote per cluster. Surface the priority gaps and hold single mentions aside so you understand what the round is collectively asking for before you score individual proposals.
Thematic analysis groups open-text responses into recurring patterns — themes — backed by counts and quotes. For a grant round it reveals which needs are widespread versus one-off, so funding decisions reflect the population's actual priorities rather than the most persuasive single application.
A theme is weak when it's a single mention dressed up as a pattern, or an ambiguous 'other' bucket that blends distinct meanings. The fix is to hold single mentions as watch-items and split catch-all buckets into named sub-themes, each with its own count and quote.
Open Sopact Sense, paste your program description, and put it to work.
Try in Sopact
