A land-survey firm where the report wrote itself slower than the survey, and the reference library couldn't be trusted
Built two tools the firm asked for. The first turns a structured field-data sheet into the repetitive sections of a survey report, generated and ready to verify instead of retyped. The second is a private reference vault: the firm's own regulations, standards, and precedent documents, queryable in plain language, grounded only in what was loaded so it never invents a citation.
Every file the firm produces has the same skeleton: the parcel parameters, the regulations that apply, a description of the boundary and ownership chronology, an as-is situation with a diagram. The numbers live in a structured data sheet, the report is written in a separate document, and someone keeps the two in sync by hand, linking figures across, copying sections, and adjusting for the small twist that every file carries. A single file runs around fifty hours, and a large share of that is text handling: copy, paste, relink, revise. The drafting, not the fieldwork, was the long pole.
The second problem was research. The firm relies on a deep, accurate reference base, regulations, standards, annual review papers, and bulletins, but no single search covered all of it, so a question meant opening source after source by keyword. Public AI tools were worse than no tool: asked for a precedent, they would return citations that did not exist. In a field where everything has to be verified anyway, an answer you cannot trust is just more work.
The first tool takes the structured field-data sheet and generates the recurring report sections from it, in the firm's own chronology and layout. The figures and the prose stay linked, so when the data sheet changes the report follows, and the work left for a person is to read and confirm, not to retype. Extraction is AI-driven with a deterministic fallback behind it, so even a messy file produces a usable draft rather than failing.
The second tool is a confidential reference vault. The firm drops in its own PDFs, thousands of pages, growing every week, and asks questions in plain language with full context, not just keywords. Answers are grounded strictly in the loaded documents and cite where they came from, so the firm gets a real reference instead of an invented one. Nothing leaves the firm's own environment, which was a hard requirement given how sensitive the files are.
Drafting a file used to mean hours of relinking and retyping. The firm now recovers four to five hours on every file, on work that runs about fifty hours each, and the same team clears more files without adding a person, in a market where the hard part is hiring. The hours that remain go to judgment, checking and adjusting the draft, which is where the firm's value actually sits.
Research stopped being a tab-by-tab keyword hunt. A question now returns a grounded, cited answer drawn from the firm's own library, with no invented precedents to chase down, and the library keeps growing without anything leaving the building.
“The bottleneck was never the survey. It was turning the data into a written report by hand, line by line, and a search tool we could actually trust. The public ones kept inventing references that didn't exist.”
Have a mess like this?
Tell us where time and money are leaking. We’ll map the fastest high-ROI win, usually a Pilot that pays for itself.