SiteVoice ‹ back to the doors

What this system does with twenty seconds of honesty

SiteVoice turns short, anonymous entries from the people doing the work into the risk picture their leaders are required to see — in the language regulators already use.

How an entry becomes intelligence

1

Someone says something. Green for a suggestion, red for an issue. Typed or spoken — speech is transcribed on this system's own server and the audio is deleted immediately. No login, no name, no device details.

2

The analysis engine reads it. A trained reading model extracts what the entry is about (the aspects: equipment, workload, management, pay, environment and so on) and how it was felt — the emotional reading, including intensity, understatement (mild wording over serious content is detected and flagged) and whether the content is escalation-grade.

3

The reading becomes recognised risk signals. Feelings and aspects compose into signals that match the structures workplace regulators already publish: the Australian model Code of Practice for psychosocial hazards, the UK HSE Management Standards, and the ISO 45003 risk areas. Free text goes in; "Low control over work, rising, worst severity 3" comes out — with the framework alignment printed in every report.

4

People see what they need, when they need it. Danger-to-life content alerts the supervisor immediately. Everything else lands in the daily brief, the board, and the weekly and monthly listening reports — retellings only, never the writer's own wording.

The anonymity model

What happens to original wording

This is a per-company policy choice, set by the account owner:

This deployment: …

In this demo, an audit view additionally shows originals beside the analysis so the reading quality can be assessed. Production deployments follow the policy above.

What leaders get

Signal categories are aligned with recognised frameworks (AU model Code of Practice, UK HSE Management Standards, ISO 45003); alignment does not imply approval or certification by any regulator. Demo data consists of real worker and incident texts from public corpora, analysed by the real engine; no fabricated readings appear anywhere in this product.