Investigations Get Closed Faster
The loop notifies the owner of a stalled investigation and the safety admins of an unassigned high-severity incident, with mean time-to-close on the console — off by default, notify-only.
HOW IT WORKS
How it closes an investigation
From reported to closed — using the same severity rules, investigator assignments, and site scoping you already enforce. It works the queue so nothing sits untouched.
1. Detect
An investigation sits untouched for 7 days or more, or a high or critical incident is still reported with no investigator. The hourly loop reads both signals before EHS has to look.
2. Decide
Each detection is a plain database fact and carries a fixed confidence of 95. The path follows the signal — nudge the incident's investigator (or reporter), or escalate to the safety admins for a high-severity unassigned case.
3. Act
Notifies the owner or the safety admins — and only notifies. It never edits, assigns, or closes an incident. Re-notifies every 24 hours until someone acts, and gives up after 90 days.
4. Log
Every nudge and escalation is recorded as an autonomous action naming the incident and its severity, listed in the console's "What the AI handled" feed and in Automation Hub.
AUTONOMY YOU CONTROL
Three levels of autonomy. You pick.
It ships off. Turn it on and it starts on probation, where every action waits for approval. Trust levels move as approvals accumulate; repeated rejections put every action back in the queue.
Off — the default
The hourly scheduler skips the tenant entirely. Nothing is detected, nothing is sent. Turn it on when EHS is ready.
Probation — approve everything
Each action is held in the console's approval queue until EHS approves or rejects it. Three rejections in 30 days put every action back in the queue.
Standard and trusted
Stalled-investigation nudges go out on their own within a daily limit — 50 actions a day by default, and admins can set a different limit per level in Automation Hub. A high-severity escalation is a critical action, so it still waits for approval at every level. An admin can pause the loop from AI Operations.
Every nudge and escalation is a logged autonomous action
Each row names the incident and its severity. A high-severity escalation also records how many safety admins were reached, and one that reached nobody is logged as failed. Each row lands in the console feed and in Automation Hub, so the retro reads the log instead of memory.
- Stalled investigation — investigating status, untouched for 7 to 90 days; the assigned investigator (or the reporter) is nudged.
- High-severity unassigned — reported, high or critical, no investigator; safety admins are escalated.
- 24-hour re-notify — the same detection is not repeated more than once a day.
- Notify-only — the loop never changes the incident record itself.
One console — EHS's home for investigation autopilot
The Safety Hub AI console is the landing for EHS leaders and safety managers. Mean time-to-close (closed minus occurred) sits front and center. The "What the AI handled" feed shows the most recent actions. The "AI actions to approve" list holds actions waiting for an admin. The AI mode is set on the same page; the pause switch is in AI Operations.
- Hero metric — mean time-to-close in days, with actions today and the 7-day success rate beside it.
- "What the AI handled" feed — the most recent nudges and escalations, each with a timestamp.
- "AI actions to approve" list — held actions approved or rejected inline.
- AI mode — set Off, Approve or Auto on the console; an admin can pause the loop from AI Operations.
Where Safety Workflows Break Down
Safety Hub AI attacks the specific failures that turn a healthy safety program into a binder full of late-reported incidents — without changing the investigation routing or corrective-action flow EHS already runs.
Near-Misses Don't Get Reported Because The Form Is Painful
A pallet jack tipped. Nobody was hurt. The hazard is real, but filing the form means opening a desktop app, picking 8 dropdowns, and writing a paragraph. The employee thinks "I'll do it later" and never does. The pattern that would have surfaced the dock-floor issue stays invisible.
Certification Expiries Surface At The Worst Possible Moment
Priya's forklift certification expired four days ago. She's been operating the forklift all week. Nobody caught it because the certification report lives in a compliance dashboard nobody opens. The miss becomes obvious during an audit, when it's already a violation.
Investigation Timelines Are Stitched Together From Memory
The lost-time injury happened May 7. The retro is May 14. Who responded first? When was the corrective action assigned? When was the medical clearance signed? Without a clean timeline, the retro becomes a guessing exercise — and the root-cause analysis suffers.
"Is That Hot-Work Permit Still Open?" Gets Asked After The Shift Change
The day crew issued a hot-work permit that ended at 16:00. The night supervisor does not know whether it was closed, whether the fire watch finished, or whether the contractor doing the welding was ever pre-qualified. The paper copy is in a truck. A one-line question in chat should answer all three.
"What Do I Do In A Chemical Spill?" Gets Asked During The Spill
Procedure binders sit in supervisor offices. Nobody re-reads them quarterly. When an actual emergency hits — chemical, medical, weather — the response is what someone half-remembers from new-hire training. A 3-second query in chat would surface the actual procedure.
Hazard Observations Get Logged And Then Nothing Visible Happens
An employee reports a slippery loading-bay floor on Tuesday. EHS triages it Friday. The repair ticket goes out Monday. From the employee's side, the system swallowed the report and nothing happened — so the next hazard doesn't get logged because "what's the point?"
Safety Hub AI At A Glance
Safety Hub AI
Hazard capture, permit and contractor status, muster roll-call, and an incident access log.
Inside Safety Hub AI — The Actual Capabilities
Every block below maps to a real tool the agent uses against your Safety Hub records. 31 tools: 24 are read-only; 7 writes (report_incident, create_safety_observation, compose_alert, send_alert_now, cancel_alert, withdraw_alert_approval, set_safety_kb_entry_owner) are confirmation-gated. Identifiers are masked in every tool result. Capture is fast; investigation routing stays in the Safety Hub app.
Incident Visibility — Live Summary, History, Timelines
Surface what's happened — incident counts by type, location, and severity over a configurable window. Pull a single incident's full record or its investigation timeline. The piece a site manager pulls Monday morning before the EHS huddle.
list_incidents— injuries, near-misses, property damage, and environmental incidents with type/status/location filters.get_incidentandget_incident_timeline— full record and the investigation timeline with corrective actions.get_incident_summary— counts by type, severity trends, location breakdown for any window.get_safety_metrics— incident rates, observation trends, and compliance scores;list_safety_locations— the sites those numbers are cut by.
Capture — Report Incidents And Safety Observations Fast
The two capture writes. An employee describes what happened in chat; the agent parses a draft (type, description, location, severity, when); the user must confirm before the record is filed. Once filed, the incident enters the standard investigation workflow.
report_incident— confirmation-gated write. Files an injury, near-miss, property-damage, or environmental incident.create_safety_observation— confirmation-gated write. Files a hazard report or positive-behavior recognition.- Mobile-friendly — chat capture is faster than the form, so near-misses actually get reported.
- Investigation routing unchanged — once filed, the record enters the same investigation flow EHS already runs.
Permits And Contractors — Who Is Cleared, What Is Open
The night supervisor's question. Which permits are active, awaiting approval, or overrun at this site; which contractors hold a current pre-qualification and which are expiring or lapsed. Both are read-only — issuing, suspending, and deciding stay in the Safety Hub app.
list_work_permits— by state (open, active, awaiting approval, overrun, closed), permit type, and site, with counts.list_contractor_prequalifications— current, awaiting decision, expiring, or lapsed, by vendor name.- Members see their own — a requester or holder sees the permits they are on; managers see the site board.
- Read-only by design — approval requires every precaution confirmed in the app, and a contractor permit can be blocked without a current pre-qualification.
Emergency Alerts — Compose, Send, Muster
Four confirmation-gated writes and seven reads over the emergency-alert engine Safety Hub owns. Compose a draft, send it, cancel it (which broadcasts the stand-down), or withdraw a pending approval — each shown as a card the user must confirm. Then read the roll-call.
compose_alert·send_alert_now·cancel_alert·withdraw_alert_approval— confirmation-gated;send_alert_nowadditionally requires the tenant's dispatch-alerts capability.muster_status— safe, needs help, and unresponded from the check-in roster;list_pending_alert_acknowledgments·alert_acknowledgment_summary— who has not acknowledged.list_recent_alerts·get_alert·diagnose_alert_delivery— what went out and which channel failed.summarize_alert_incident— the post-incident summary for the retro.
Observations And Toolbox Talks — Programmatic Safety
Pull the observation list (hazards plus positive behaviors), drill into a single observation, surface toolbox-talk attendance, and see the topic library. The mechanics of running a safety program — not just reacting to incidents.
list_safety_observations·get_observation_details— hazard and positive-behavior captures.list_toolbox_talks·get_toolbox_talk— past and scheduled safety meetings with attendance.list_toolbox_talk_topics— reusable topic library (PPE, Fire Safety, Ergonomics, Chemical).- Filter by attendance — "show me talks I attended" or "talks I should attend".
Certifications And Knowledge — Compliance + In-The-Moment Procedures
Your active certifications, what's expiring in 30 days, what's expiring across your team, and the safety knowledge base for procedure lookups during an actual incident. The Safety Hub AI earns its keep most when a supervisor asks "what's the chemical-spill protocol?" mid-spill.
get_my_certifications— your active, expiring, or expired safety certifications.list_expiring_certifications— manager view of certifications expiring within N days.search_safety_kb— search procedures, PPE guides, hazard info from the safety knowledge base.get_emergency_procedures— fire, medical, chemical spill, severe weather protocols by type.
Outcomes Teams Can Measure
The agent is built to compress incident-reporting time, surface certification expiry before it lapses, and make safety procedures queryable during the moment they're needed. Measure against your pre-agent baseline.
- Mean time-to-close — the console's hero metric: days from occurred to closed, trending against baseline.
- Near-miss reporting rate — share of operational near-misses captured as observations, trending against baseline.
- Time from incident to filed record — minutes from the event to
report_incidentconfirmation. - Certification-expiry lead time — days between an upcoming expiry being flagged and the renewal class scheduled.
- Permit overruns — permits the hourly sweep flagged past their window, trending down as close-out becomes routine.
7 Confirmation-Gated Writes, Investigation Routing Always Enforced
Safety Hub AI has 31 tools. 24 are read-only (incident lookups, summary, timeline, metrics, observations, toolbox talks, permits, contractor pre-qualifications, certifications, knowledge base, emergency procedures, alert status and muster). 7 writes — report_incident, create_safety_observation, compose_alert, send_alert_now, cancel_alert, withdraw_alert_approval, and set_safety_kb_entry_owner (a manager changing who owns a safety article) — show a confirmation card and wait for the user. Once filed, the standard investigation flow takes over.
- 7 confirmation-gated writes — two capture writes, four alert writes, and a manager's change of a safety article's owner;
send_alert_nowalso requires the tenant's dispatch-alerts capability. - Investigation routing unchanged — filed records enter the same workflow EHS already runs in the Safety Hub app.
- Permission-aware and PII-masked — site scoping and role gates mirror the web app, and every tool result replaces people with pseudonyms and withholds contact details.
- Every tool call logged — read or write, every tool call is logged with the requesting user and which tool ran; the AI request log keeps detail for 30 days; agent reads of an incident land in its access log.
WHAT TEAMS TRY INSTEAD
The four alternatives — and why none of them connect floor capture to permits, certification, knowledge, and the investigation workflow
EHS teams have piloted safety AI for years. The honest gap is that most options live in a separate EHS platform, surface incidents but not permits or certifications, or capture observations but never route them into a real investigation.
Pasting incidents into ChatGPT, Claude, or Copilot
General-purpose AI summarizing a copied incident report
- Files into the same incident workflow EHS already runs — not a draft pasted into an email
- Reads live permits, contractor clearance, certifications, hazard observations, and toolbox-talk records — no copy-paste
- Identifiers are masked in every tool result, and AI requests go to the platform's configured model providers
SafetyCulture AI / Intelex AI / Cority AI
Vendor-trapped safety AI behind a separate EHS platform
- Lives where employees already are — no separate EHS app sign-in for a quick incident report
- Reads certifications and trainings from the platform's HR record — not a synced shadow copy
- Available to frontline crews who never had a separate EHS seat
Custom safety dashboards and Sheets-based observation logs
An EHS team's spreadsheet that nobody updates after week three
- Live incident summary, open permits, certifications expiring this month, hazard observations — all in one prompt
- Investigation routing unchanged — filed records enter the same Safety Hub workflow EHS already runs
- Every tool call logged with the requesting user and the tool — read or write; agent reads of an incident land in its access log
The manual fallback — paper forms and weekly safety meetings
Capture lag that buries patterns under months of unfiled observations
- Observation capture happens between tasks — not at the end of the shift when memory is thin
- Certifications expiring next month surface this month — not the day someone's lapsed
- Muster status, toolbox-talk topics, and emergency procedures retrieved by ask, not by remembering which binder
PLATFORM ADVANTAGE
Safety Hub AI inherits everything Safety Hub already enforces
A standalone safety AI has to plumb identity, investigation routing, and certification tracking. Safety Hub AI gets all of it for free.
Confirmation-gated writes
7 writes (report_incident, create_safety_observation, compose_alert, send_alert_now, cancel_alert, withdraw_alert_approval, set_safety_kb_entry_owner) require explicit confirmation. The model proposes; the human commits.
Investigation routing unchanged
Filed records enter the same workflow EHS already runs. The agent files; EHS investigates. No parallel system to keep in sync.
Permission-aware and PII-masked
Incident visibility, permit and observation scope, and certification scope all respect the Safety Hub app's existing rules — and every tool result replaces people with pseudonyms and withholds contact details.
Certification visibility
Certifications expiring this month surface conversationally — no separate alert configuration, no spreadsheet to maintain.
Cross-app data plane
Reads certification holdings from Skills & Credentials and Training completions, vendor records behind each contractor pre-qualification, and incidents, permits, and alerts from Safety Hub.
Every tool call logged
Every read and every write is logged with the requesting user and which tool ran; the AI request log keeps detail for 30 days. Agent reads of an incident also land in the incident's access log, kept seven years.
INDUSTRY FIT
Industries where safety capture decides outcomes
Safety Hub AI earns its keep where the work is physical, the certification regime is real, and the patterns hide in observations that nobody files.
Manufacturing
Hazard observations captured between tasks — patterns surface weeks before a recordable, with certification scope visible inline.
Construction
Field crews file observations from the same mobile app they use for shifts — and ask which hot-work permits are still open and which subcontractors are cleared.
Healthcare
Clinical-safety observations captured on the floor — investigation workflow routes per the same rules EHS already configured, with identifiers masked in every tool result.
Energy & Utilities
Certifications expiring this month surface before crews are dispatched — and overrun electrical-isolation or confined-space permits are one question away.
Logistics & Warehousing
Yard and dock observations captured by frontline supervisors — patterns visible before a forklift incident becomes an OSHA case.
Public Sector
Incident access logged for seven years, identifiers masked in every tool result, and muster status readable during an evacuation.
WHY MANGOAPPS WINS
An embedded safety agent beats a horizontal AI, an EHS-vendor module, or a custom build on every axis
The argument EHS, operations, HR, and frontline supervisors all share — and the one SafetyCulture or Intelex structurally cannot answer.
Cheaper than the alternatives
No SafetyCulture AI tier, no Intelex add-on, no Cority subscription, no engineering team building a custom safety observation form.
More secure
Confirmation-gated writes, permission-aware reads, PII-masked results, every tool call logged. AI requests go to the platform's configured model providers, like every MangoApps agent.
Easier to deploy
Already deployed if Safety Hub is on. Agent picks up active investigation routing, the permit board, and certification records the same day.
Easier to use
Frontline crews file observations between tasks from the same mobile app they already opened — capture stops losing to friction.
Easier to manage
Investigation routing edits in the app are immediately visible to the agent — no re-training, no parallel rule store.
Easier to extend
Shares the agentic-tool framework with every other MangoApps agent. New safety tools (a new observation type, a new certification cut) ship as tools.
AI is actually better
A horizontal AI can summarize a safety report. Safety Hub AI files an incident into the live investigation workflow, tells the night supervisor which permits are still open and which contractors are cleared, surfaces certifications expiring this month, and reads the muster roll-call — all confirmation-gated, with every tool call logged.
At our head office within HR, marketing, and finance, we use ChatGPT quite a bit, and MangoApps AI Helper is a useful tool for us. As I go around to the stores, I’ve been showing them how the AI Intranet Helper can find things and answer plain language questions about our resources, and they’re quite excited about that.
Alison van Zijl
Head of Human Resources
Full House
Customer Success
MangoApps Customers
Organizations that trust MangoApps to run every workflow across their workforce.
Customer Case Studies
Video Case Studies
Webinars
Customer Case Studies
Customer Case Studies
Customer Case Studies
Frequently Asked Questions About Safety Hub AI
31 tools across the safety lifecycle — list and filter incidents, view a single incident, get the investigation timeline, get the summary and metrics for any window, file an incident or safety observation (both confirmation-gated), list and read safety observations, list work permits and contractor pre-qualifications, list and read toolbox talks plus the topic library, view your own certifications and your team's expiring certifications, search the safety knowledge base, look up emergency procedures by type, and compose, send, cancel, or withdraw approval on an emergency alert (all four confirmation-gated) and read its muster and acknowledgement status. A manager can also change who owns a safety knowledge base article, after confirming. 24 of the 31 are read-only.
No. All seven write tools (report_incident, create_safety_observation, compose_alert, send_alert_now, cancel_alert, withdraw_alert_approval, set_safety_kb_entry_owner) show a confirmation card and wait for explicit confirmation before acting. send_alert_now additionally requires the tenant's dispatch-alerts capability, which defaults to managers, and refuses while an approval is pending. The actual investigation routing then proceeds normally.
Yes, read-only. list_work_permits filters by state (open, active, awaiting approval, overrun, closed), permit type, and site, and returns counts. list_contractor_prequalifications lists current, awaiting-decision, expiring, and lapsed pre-qualifications by vendor name. Issuing a permit — which requires every precaution confirmed by the authoriser — and deciding a pre-qualification stay in the Safety Hub app.
No. It is off by default and notify-only. When enabled, an hourly job detects investigations untouched for 7 days or more and high-severity incidents with no investigator, then nudges the investigator (or reporter) or escalates to safety admins — it never edits, assigns, or closes a record. It starts on probation (each action approved in the console's queue; up to 50 actions a day by default), and an admin can pause it from AI Operations.
get_emergency_procedures takes a procedure type (fire, medical, chemical spill, severe weather, etc.) and returns the configured procedure for the business. The intent is to give a frontline supervisor a 3-second chat lookup instead of digging for a binder during an actual event.
get_my_certifications is self-scoped. list_expiring_certifications takes an optional user_id that managers can use to view a direct report's expiring certifications — subject to the same site-scoping rules the Safety Hub app applies. Employees cannot use that parameter, and identifiers are masked in every result.
Mean time-to-close (the console's hero metric), near-miss reporting rate, time from incident to filed record, certification-expiry lead time, and permit overruns. Compare against your pre-agent baseline.
Let's Talk
Every team. Every employee. Every workflow. One AI platform — built for the frontline. Since 2008, trusted by 2 million+ users with an NPS of 78.
Why Choose Us?
- Frontline AI: Governed AI for every employee and workflow.
- Top Security: HITRUST, ISO & SOC 2 certified.
- Exceptional UX: Delightful on mobile and desktop.
- Proven Results: 98% customer retention rate.
Rolled out to every employee at AutoZone (125,000), PetSmart (50,000+), A.S. Watson and Raley's (20,000) — and at larger retailers we are not permitted to name.
Prefer to explore first? Ask AI about Safety Hub AI →