Case study · Hazard consultancy
Asbestos clearance reports are long PDFs full of certificate numbers, sample counts and lab references. Someone had to read each one and key the billing quantities into simPRO. Now a pipeline reads every page, extracts the quantities with a confidence score, matches the job and writes the note, in minutes.
Background
The consultancy issues clearance reports after asbestos removal: twenty or more pages of certificates, sample counts, lab references and background-air monitoring days. Billing depended on a person reading each report and keying the quantities into the matching simPRO job in a set note format.
In every one of the 22 reports we tested, the certificate numbers were not in the PDF text layer at all. They only existed as images, which is why a text-scraping prototype kept misreading them.
Our approach
The client's own text-scraping prototype was treated as guidance only. The agreed approach renders every page and reads it with vision, per the statement of work.
22 live reports run end to end before anything touched simPRO: 22 of 22 extracted, 20 of 22 matched to the right job, the two misses correctly excluded.
Reports are matched by the job number printed in the project details. Doubtful fields are flagged NEEDS REVIEW; a wrong job number means the report is held, not posted.
First live batch on 4 August 2026, then the client's and the office manager's rulings on note format and edge cases were built in the same week.
Challenges and solutions
Certificate numbers, sample counts and lab references sat in images across twenty-plus pages, with a second set of leak-test certificates in some reports.
A person skimming a long report misses things, in both directions.
Measurable results
Implementation timeline
What changed day to day
If your billing depends on someone reading long documents, this is the fix: read by vision, matched to the job, flagged when unsure.
Billing quantities land in simPRO the day the report is issued, without anyone reading the PDF to find them.
Items a person skimmed past are picked up and billed. Three errors caught in the first batch alone.
Billing no longer waits for the one person who knew how to read the reports.
Certificate numbers, sample counts and lab references captured in a consistent format with the source recorded, so the billing trail stands up to a client or auditor query.
Uncertain fields are flagged, wrong job numbers are held, billing queries ride along in the note.
Invoices go out sooner and match the report, so fewer queries come back.
“Skene is very efficient, a great communicator and came across as very knowledgeable and competent. We were very happy with the outcome of our first project and will use him again.”
Cameron, director, Google review, September 2026
More case studies
Court documents, client intake and submissions automated for a New Zealand law firm.
Inventory, supply chain and customer engagement automation for a NZ retail chain.
AI-driven IT operations and ticket triage for a New Zealand FinTech.
Claims processing, data entry and fraud detection for a NZ insurance broker.
Subcontractor agreements and supplier invoices handled automatically for a NZ property developer.
A custom CRM, scheduling and billing system for an NZ-wide HVAC company.