Turning a fragmented funding landscape into one intelligent platform.
How Optivus designed, built, and now operates an AI-powered grant-discovery platform for a UK funding intermediary — converting hundreds of scattered, ever-changing funder websites into a single, continuously-updated, human-verified pipeline that matches organisations to the money they qualify for.
The client is a specialist grant-funding intermediary serving SMEs, charities, and social enterprises. Their mission is to connect under-resourced organisations with the funding they qualify for — opportunities otherwise scattered across hundreds of funder websites, published in inconsistent formats and updated without notice.
To deliver this reliably and at scale, they engaged Optivus as their end-to-end technology partner: one team for strategy, build, and ongoing operation. The opportunity was real, but the operating model could not scale — finding, verifying, and maintaining grant data by hand was slow, error-prone, and impossible to grow with a lean team.
Grant opportunities were scattered across hundreds of funder websites with no common format, deadlines and statuses changed silently, and organisations were shown irrelevant grants — eroding trust, with no subscription infrastructure to monetise the service at all.
A single platform combining an AI data-acquisition pipeline, an intelligent matching engine, subscription monetisation, and dual web portals — built with human-in-the-loop approval, staged rollouts, and vendor-agnostic AI at every step, so automation never becomes an opaque black box.
All 41 eligible users were matched to relevant grants within a two-week window, with 90% receiving an ideal 3–5 recommendations. The platform now runs as an ongoing partnership, extending into B2B2C seat-licensing, CMS expansion, continuous model optimisation, and production hardening.
Engagement process
Each milestone delivered as a 2–4 week workflow with written commitments and a weekly working demo — never an open-ended build.
AI quality measured against real, client-reported cases first — model selection is a data-backed decision, not a guess.
Every grant discovered and every change detected is surfaced to a person for approval — fully auditable and reversible.
Staged rollouts, allow-lists, kill-switches, and log-only modes guard every production surface before a single live send.
Models swap via a single setting across every call site, enabling continuous quality and cost optimisation with zero code change.
Documented, tested, and operable by the client's own team — systems built to run, not thrown over the wall.
Continuous discovery, always current.
Autonomous crawlers visit funder sites and use LLMs to extract structured grant data, then re-check every live grant daily — flagging only status and deadline changes for human review.
One scoring engine, every channel.
A shared scoring engine ranks every grant against each organisation across five weighted signals, powering both the on-site recommendation feed and weekly email alerts.
One codebase, two purpose-built portals.
An internal admin dashboard for review and operations, and a public portal for registration, search, and bookmarking — with lifecycle email integrated across the funnel.
Delivered as a single monorepo, auto-deployed across managed cloud infrastructure with environment-level safety controls.
Have a project like this in mind?
Whether you're evaluating AI opportunities, modernizing existing systems, or building your next digital product, we'll help you identify where technology can create the greatest impact.