The problem
Before a developer commits to a site, someone has to read the paperwork. Planning certificates, zoning schedules, contracts of sale, environmental overlays — each one dense, inconsistently formatted, and full of clauses that determine whether a project is viable at all.
The client's team was doing this by hand. An analyst would open a PDF, find the relevant sections, cross-reference them against local environmental plans and zoning data, and write up an assessment. It took days per site, it didn't scale with deal flow, and — the part that worried them most — a missed clause in a document nobody had time to read properly could sink a project after money had been committed.
The bottleneck wasn’t analysis. It was getting the document into a form that could be analysed at all.
What we built
We engineered a document intelligence service that takes planning documents from ingestion through to a finished assessment report, with a web portal wrapped around it for project and stakeholder management.
- Ingestion and OCR. Documents are received in whatever state they arrive in — scanned, photographed, digitally generated — and normalised into machine-readable text with layout preserved.
- Structured extraction. The Document Agent locates and extracts the fields that matter: zoning classifications, permitted uses, height and density constraints, overlay designations, contract terms.
- Validation against reference data. Extracted values are checked against local environmental plans, zoning schedules and GIS data, so the system doesn't just report what the document says — it reports whether what it says holds up.
- Parallel processing. Document sets are processed concurrently rather than sequentially, which is what turns a multi-day review into a wait measured in minutes.
- Automated assessment reports. Findings are compiled into a report the analyst reviews rather than writes — with every extracted value linked back to the page it came from.
How the agents were used
The extraction, validation and reporting layer is the AIOBC Document Agent, configured to this client's reference datasets and rules. The bespoke engineering went into the parallel processing architecture, the GIS validation logic and the portal — the parts genuinely specific to their business.
The client paid for the work that was unique to them, and inherited a document pipeline that had already been proven elsewhere.
The outcome
Site assessment moved from a multi-day analyst task to a review of a generated report. More importantly, the process became consistent: the same document produces the same findings every time, and every finding is traceable to its source. The team's analysts stopped transcribing and started making judgements — which is the work they were hired for.
Where a human stays in the loop
The system does not make the investment decision, and it doesn't sign anything off. It surfaces what the documents contain, flags where they conflict with reference data, and shows its working. A qualified person makes the call. In regulated, high-value processes, that boundary is a design requirement, not a limitation.