Short answer

AI auto-classify reads an asset's photo and existing data, then writes back: a clean asset-type name, an asset-type description, and a first-pass condition assessment. It will overwrite the asset's description field, so before bulk-classifying any asset that has a manually-typed description, copy that text into Notes first (Notes are not touched by AI).

Two just-uploaded assets on the mobile list showing Untitled asset with a Classifying badge
Just uploaded: 'Classifying…' — the AI is working

Detail

Run it one of two ways:

  • Single asset. Open the asset's side drawer, scroll down, click Auto AI Classify. Updates name, description, and condition assessment.
  • Bulk. From the asset list, tick the assets you want to process, then click the robot icon in the bar at the bottom. Runs the same classification across the whole selection.

What it touches:

Field AI behaviour
name Replaced with a clean asset-type name (e.g. "Green Litter Bin").
description Replaced with the AI-generated description. Manually-typed descriptions are lost unless saved elsewhere first.
condition / condition_description Filled with a first-pass assessment from the photo.
note Not touched. Safe to use for any text you want to preserve.
purchase_value, current_value, etc. Not touched.
next_inspection_date, last_inspection_date Not touched.

The single biggest CS lever for any account that built up assets pre-AI is to bulk-classify after copying descriptions into Notes. It's also the largest single contributor to a low Data Quality score on the health leaderboard — the processing_status = NOT_CLASSIFIED ratio is the dominant signal.

The same list moments later: Outdoor cross trainer and Outdoor recumbent bike, named and typed with condition Good
Moments later: named, typed, condition set

Watch-outs

  • Description gets overwritten. Always copy manual descriptions to Notes before bulk-classifying. The classic flow is: open asset → scroll to Description → cut → paste into Notes → save → then run classify.
  • Local nicknames get lost. AI doesn't know that the council calls something "the BruBowl" or "the Mugger". Do a quick pass after classify to humanise names where it matters for the field worker.
  • AI doesn't always succeed. Some assets come back with processing_status = FAILED. Worth checking why on a sample (typically poor photo, missing photo, or a weird asset type the model doesn't recognise).

  • It can classify identical items inconsistently. A council with uniform stock (e.g. 10 identical hedge trimmers) may find the same model classified differently from one photo to the next. The fix is photo quality: take a clean, isolated shot of the item with nothing else in frame — clutter or other kit nearby confuses it. Always sense-check the result, and if something is genuinely wrong and you can't see why, send Civic.ly a link to the asset so we can look into improving it.

  • A just-uploaded asset shows "N/A" (web) or a "Classifying…" badge (mobile) while the AI is still working. Processing usually takes well under a minute — refresh the list and the name, type and condition appear. Only investigate if it never resolves (see the failed-classification note above).

An asset's detail screen with the AI-written name, type and description for a historical building
The AI fills the name, type, description and condition — you just correct what it got wrong