A European government authority took months to approve a permit, its clerks stitching answers together from four disconnected systems — and too afraid of a data leak to automate. We solved the speed and the privacy at once.
Citizens and contractors seeking permits, building approvals or rights waited long months — stuck behind a manual, cumbersome review of scanned PDFs and old archives.
To approve a single request, a clerk had to check four separate systems — municipal tax, engineering, legal and finance — that weren't connected to each other, which bred human error and delay. And management was afraid to bring in new automation at all, worried that sensitive citizen data (personal privacy) would leak out or be exposed to staff without authorisation. So nothing moved, and the queue grew.
A closed, privacy-safe environment where the systems can answer each other, and staff can just ask.
We built a closed data architecture — clean rooms — that lets the different departments share and query each other's information securely, with no copying of the underlying databases and no compromise of citizen privacy.
An engine connected to the public-enquiry system reads a complex application and its legal annexes — dozens of pages — in seconds, pulls out the relevant facts, and prepares a ready-to-check draft approval for the clerk.
Staff get a simple interface: they type in ordinary language — "does property X have open building violations or municipal debts?" — and get a verified answer drawn from across all the systems in seconds.
Instead of a clerk logging into four disconnected platforms and reconciling them by hand, the request arrives as a single, cross-checked picture — the errors and the delay of manual stitching gone.
The AI reads, cross-checks and drafts; the official reviews and approves. The judgement — and the accountability — stays with a person, which in the public sector is not optional.
Because data is queried in place rather than copied, and access is scoped, the whole thing meets privacy regulation head-on — the reason management could finally say yes to automation.
The speed came without the leak everyone had feared.
Average handling time for public requests and permit approvals fell from 45 working days to about 48 hours, and clerical, manual work dropped by 70% — letting the authority process three times as many requests with the same headcount.
There were zero incidents of data leakage or exposure of confidential records, full compliance with privacy regulation, and a reputational shift — from a slow, bureaucratic body to an efficient, modern one.
The speed and the privacy weren't a trade-off — the privacy-safe architecture is exactly what made the automation possible. Data queried in place, access scoped, a human on every decision: that is what let a cautious public body move at all. The same secure, human-owned approach runs our natural-language assistant work.
Reef TRH