January 7, 2026
AI Autoclassification: How Golden Plains Shire Council Classified 800,000 Property Files
Golden Plains Shire Council held 800,000 property files with no way to sentence, retain, or dispose of any of them. AI Autoclassification found what could go, and let the council dispose of it for the first time in twenty years.
90%+
- 800,000 property files, one governed archive
- Files migrated years ago from Arnold EDRMS with no metadata now carry classification, retention, and disposal status.
- First digital disposal in twenty years
- Compliance is built into daily workflows, reducing manual effort and risk
- 11.5 years of manual work, run in daily batches
- Classifying this by hand would have taken one person 11.5 years. The engine now runs 65,000-file batches a day.
- Run by GPSC's own team
- GPSC's Information Management Lead operates the classifier herself. Her review dropped from hundreds of files per batch to a random 20–30 as trust in the model grew.
Challenges
Twenty years of property records nobody could sentence, retain, or dispose of.
Golden Plains Shire Council held 800,000 property files in SharePoint, migrated years earlier from an old EDRMS with no metadata attached. Staff could find a document by property. They could not sentence it, apply a retention rule to it, or dispose of it once its hold period had passed.
In more than twenty years of digital record-keeping, the council had never completed a single digital disposal. Classifying the holding by hand, at 90 seconds a file, would take one person 11.5 years, if that person did nothing else.
Solution
Information Leadership partnered with Golden Plains Shire Council to tune an AI Autoclassification engine against the council's own records judgement.
The engine is Azure-hosted. It reads each file, classifies it against the council's own rules, and returns the metadata. GPSC started with a proof of concept, sized at $9,000, easy to approve, because it cost about the same as the SharePoint storage the council would have paid for anyway.
The core of the work was the informing document - a structured expression of the council's own classification judgement, built iteratively with its Information Management Lead, Lena Wealands, at the centre. She opened individual pilot files, confirmed classifications, and refined the language, sometimes spending ten minutes on a single file to work out enough context to make a call.
"I'm not sure if I can do this... I just work with records, I can look at something and tell you what it is. How do I get this tool to do that as well?"- Lena Wealands, Information Management Lead, Golden Plains Shire Council
She got there. Delivery ran in three gated phases: setup and tuning against two batches of 1,000 files, a first production batch of 50,000, then rolling 65,000-file batches now run daily by GPSC's own team. Each phase ended in an explicit go or no-go, review the results, then choose to proceed. The largest risk came mid-engagement, when the engine was rebuilt from on-premises to Azure and GPSC became the first live client on the new platform. The trust built during tuning carried it through.
Results
GPSC has completed its first digital disposal action in more than twenty years.
The engine has classified around 600,000 files, roughly three-quarters of the holding, in daily batches, running at 90%+ accuracy in production. A quarter to a third of the archive has been flagged as low or no value and queued for defensible disposal, without the formal sign-off that permanent records require.
Staff now find property records by type and by council activity, not by guesswork. A rates officer can see at a glance that a building record sits outside their scope. FOI responses are faster because the metadata says what a record is before anyone opens it.
"It means I don't have to look at 800,000 individual records."- L ena Wealands, Information Management Lead, Golden Plains Shire Council
The model is reusable. GPSC's Manager ICT estimates at least 60% of the informing document would transfer directly to another Victorian council. Inside GPSC, the same approach could classify another million files across shared drives and other migrations.
"We've done the hard work for you. If you have unclassified files, this is an engine that is going to do it for you in a much more efficient and effective time than you're going to do in any other way. There's a light at the end of the tunnel that's not an oncoming train."
Not sure where to start?
Most organisations come to us at a moment of change. If something has shifted and you're not sure what good looks like - that's enough of a reason to talk.
