Grants That Actually Fit.
An eligibility-first pipeline replaced weeks of manual scanning with a 30-minute weekly digest.
What was broken before.
Australia's leading stillbirth-prevention charity was losing entire weeks per cycle to manual grant scanning. Hundreds of opportunities surfaced in each round. Most were ineligible the moment you read the fine print: wrong jurisdiction, wrong cause area, wrong applicant type. Staff were burning 5+ hours a week filtering noise before any real assessment work could start. The cost was not just time. It was applications written for grants that were never going to land, and real-fit grants missed because the team ran out of bandwidth before getting to them.
The system.
An eligibility-first grant pipeline. Source feeds and grant websites parse continuously into a normalised Supabase store, every record flattened into structured fields the assessment layer can reason over. A DeepSeek V3 assessment layer sits on top, scoring eligibility before fit. Anything that fails the hard gates never reaches the team. What survives goes through fit scoring, then lands in a verified weekly digest the COO can act on. No dashboards nobody opens. No spreadsheets. The moat is not the model, it is the operating sequence: ingestion, normalisation, eligibility, fit, delivery. Tuned weekly against real outcomes.
The outcome.
467 grants assessed through the production pipeline. 0% false positives after the eligibility-first rebuild. The weekly grant workload compressed from 5+ hours of manual screening to roughly 30 minutes of reviewing a verified shortlist. Still in production weekly use as of April 2026. Not a pilot, not a demo, not a future case study. The same charity that started with the manual process now runs the pipeline as a standing weekly habit.
“467 grants assessed, 0% false positives, weekly workload cut from 5+ hours to 30 minutes, production-live as of April 2026.”
Want one of these?
Book a free audit and I will tell you whether your grant workflow is a tooling problem or a sequencing problem.