A side-by-side read for procurement evaluators shortlisting EHS monitoring vendors — what each option actually does across the six questions enterprise buyers ask first, grounded in the surfaces the platform ships today.
Positioning
Vision-only AI tools, periodic audit spreadsheets, and cross-signal correlators were built for different problems. The matrix below names what each is actually doing — so the shortlist reads against claims a reviewer can verify.
Vision-only AI
Trained on PPE compliance and unsafe-act detection from camera frames. Strong on a single signal — was a hard hat worn in frame X? — but light on sensor and permit reasoning, so a confined-space breach with the right PPE still lands as an alert instead of a contextualised incident.
Spreadsheet audits
Sample a window, walk the floor, file a report — no always-on coverage, no cross-referenced evidence at the moment of breach. Reports are hand-written after the fact, and incidents outside the sampling window simply go unseen.
Cross-signal correlator
Always-on across sensor + permit + CCTV, reasons before it files, ships an audit-ready PDF at the moment of the breach, and routes the corrective work order to the right supervisor channel — the same loop enterprise buyers ask for on the FAQ.
Side-by-side
The matrix below mirrors the six procurement questions enterprise buyers ask first — what monitors, what reports defend in audit, how dispatch lands, what wires in, how long to first incident, and how the bill grows.
| Feature | Pactwell | Protex AI | Intenseye | Spreadsheet audits |
|---|---|---|---|---|
| Always-on monitoringWhether a violation surfaces the moment it happens — including off-shift — or only during periodic review. | 24/7 cross-signal correlator over sensor + permit + CCTV — reasoned before it files. | Real-time on camera, but only inside trained coverage windows. | Real-time on camera, model-confidence-gated on PPE & unsafe-act classes. | Periodic sampling windows — off-shift incidents simply go unseen. |
| Audit-ready reportsWhether the evidence pack — timestamped CCTV frame, sensor reading, permit state — is captured at the moment of the breach or compiled afterwards. | PDF written at the alert step with cross-referenced CCTV frame + sensor + permit state. | Video-clip export per event — narrative context written by hand afterwards. | Video-clip export per event — narrative context written by hand afterwards. | Hand-written report from whoever remembered to take notes that week. |
| Work-order dispatchWhere the alert actually lands — the supervisor channel that can close the loop — and how the corrective fix turns into a verified record. | Email · SMS · Slack · webhook routes to the right supervisor channel per bay + shift. | Notification to a wall display and a single dashboard — supervisor copy-paste to dispatch. | Notification to a wall display and a single dashboard — supervisor copy-paste to dispatch. | Ticketing copy/paste from the report — no automatic routing, no escalation path. |
| Sensor compatibilityWhat kinds of signals are reasoned across, and whether new signal categories require an engineering engagement or a settings-page config. | Gas · temperature · vibration · restricted-area contact · lockout-tagout state — added by config per the FAQ. | Camera-only — every new signal category is an engineering project. | Camera-only — every new signal category is an engineering project. | Manual entry — every reading typed in by the floor crew on the day. |
| Setup timeFrom kickoff to the first flagged incident in verified state — and whether the cameras, sensors, and permits you already run stay where they are. | ~2 weeks to first flagged incident — four-step onboarding (connect sensors, register permits, point at VMS, author rules). | Model-training dependent — first reliable alerts typically arrive after a learning window. | Model-training dependent — first reliable alerts typically arrive after a learning window. | Instant — and still not live, because nothing fires between sample windows. |
| Pricing modelHow the bill grows with your floor, and whether the crew sees the system without hitting a seat-friction wall. | Per bay under active monitoring — never per seat — sized against real signal volume after pilot. | Per-camera plus a model-training subscription line item. | Per-camera plus a model-training subscription line item. | Labour cost — nights the safety team spends compiling the evidence pack before an audit. |
Competitor capability claims are written against publicly described positioning; for Pactwell, every claim in the leftmost column maps to a shipped surface — see the enterprise FAQ, pricing, and security & architecture. For the dedicated spreadsheet status-quo read, see the spreadsheet comparison.
Run the comparison yourself
Start the same pilot the matrix is written against — connect sensors, register permits, point Pactwell at the VMS you already run, and watch a flagged incident arrive inside two weeks.