A day with Maria Chen, EHS Manager at Apex Manufacturing.

Maria opens SE on her laptop. Three weekly leading-indicator KPIs greet her: hazard reporting rate (+18% on a 30-day baseline — good, she's been pushing the supervisors), near-miss volume (14 this week, above the floor of 12), and observation completion across the three plants. Observation completion is below target — 73% versus the 80% threshold she configured. Plant 3 is the laggard.

She clicks into the observation KPI. The drill-down shows Plant 3's safety lead has been on PTO for a week; the inspections queue piled up. Maria pings him on Teams to confirm catch-up is in flight, then moves on. The whole interaction took 90 seconds.

A near-miss landed at 11:47 PM. The Plant 2 third-shift supervisor used SE's mobile voice-capture flow on his phone: he spoke into the recorder, the system transcribed the audio and extracted structured fields (title, category, severity, location, people involved) using Azure OpenAI. Maria reviews the extracted record alongside the original audio transcript — they match.

She approves the extraction with one click. The system creates a corrective action ("inspect aisle B-4 racking before next shift uses the lane") and assigns it to the maintenance lead. Maria didn't have to chase this overnight — the supervisor reported it in 90 seconds from the floor while the memory was fresh, and the AI pre-filled the fields he would otherwise have skipped.

Yesterday afternoon a Plant 1 maintenance tech caught his thumb on a sharp edge while clearing a chip jam. On-site clinic cleaned and bandaged it; he returned to work. Today Maria walks the case through SE's Recordability Wizard. The Wizard asks the standard 1904.7 questions in sequence and cites the regulation behind each branch.

First-aid-only treatment terminates the cascade. The case is marked Not Recordable with the citation captured against the record. The annual Form 300 won't pick this up, but the case is preserved in the system with a full audit trail — if an OSHA inspector later questions the determination, Maria can point at the 1904.7(b)(5)(ii) reasoning + the override path that wasn't taken. Six months ago this would have been a paper-form decision Maria made alone and hoped wouldn't be second-guessed.

Maria opens the Form 300 for Plant 1's establishment to spot-check ongoing classifications. Five recordable cases year-to-date — two DART, three other-recordable. The log is current to the moment; case 004 from earlier this week is already on it.

Last year compiling Form 300 across all three establishments took 31 hours of Maria's team's time in December and the first week of January. This year she opened it in three clicks and confirmed the data while still on her first coffee. The annual posting on February 1 will generate Form 300A automatically from this same data. Two of the three plants fall under state plans (Ohio + Pennsylvania); the system surfaces the right variant per establishment.

The Compliance Calendar surfaces an SDS approaching expiry — Cool-Synth 4400, a water-soluble cutting fluid used on Plant 1's machining line. The 30-day window opened today. Maria sees the impact at a glance: four storage locations, three active containers, used by a defined production area.

She delegates the refresh to Plant 1's chemical safety officer with a single click. The system pre-fills the email with the manufacturer's contact details (held against the chemical product record), the current SDS PDF, and a one-line context note. When the updated SDS arrives, AI will parse all 16 GHS sections and the chemical product record updates automatically. The written HazCom programme regenerates from live data; Maria doesn't touch it.

Maria does a 20-minute walk through Plant 1's machining floor between meetings. She records three observations on her phone — two compliant (good housekeeping, correct PPE at the grinding station), one non-compliant (a LOTO procedure card without a verification signature on press 4's maintenance task).

Each observation ties back to a Work Observation Item template; the LOTO non-compliance specifically references LOTO Procedure Card v2.3 § 4.2, the version that was active when the observation occurred. The discipline engine picks up the non-compliant observation in the background. If the maintenance tech has crossed an escalation threshold for the LOTO behaviour, a disciplinary case opens in HR's queue automatically — preconfigured with the cited policy section. Maria doesn't have to remember to start that workflow.

At 1:15 PM Maria joins her weekly call with the HR Director and the legal counsel on retainer for disciplinary matters. The discipline engine has surfaced three new cases since last week. Each row shows the disciplinary tier (verbal / written / suspension / termination), the cited clause + version, the stage in the workflow, and the HR queue status.

Two cases get HR activation today — Maria adds a site-specific note to D-088 to explain why the second-written tier is appropriate. D-090 (the LOTO suspension) gets deferred for legal review; counsel wants to confirm the timing of the cited LOTO Procedure Card § 4.2 version before the letter generates. The system surfaces both the active version at observation time and the live version — counsel confirms they match. The letter will generate Monday after Maria adds the file note.

Mid-afternoon Maria gets a notification: Plant 2's anhydrous ammonia inventory will exceed the PSM threshold tomorrow when a scheduled delivery arrives. The 12-hour scheduled threshold-monitoring job caught it on this morning's run; the alert sat in her queue until she got to it.

She calls the Plant 2 plant manager. Two options: defer the delivery to next week (preferred — easier paperwork) or accept it and trigger the PSM program activation cascade (mechanical-integrity inspection cycle, MOC review, RAGAGEP confirmation, employee training records audit). They choose to defer; Maria emails the vendor while still on the call. The system records the threshold-proximity event against the chemical product so the audit trail shows the catch + the decision.

Maria spends 30 minutes on the Injury Analytics dashboard. The 24×7 timing heatmap for Plant 3 shows a clear uptick on third shift over the trailing 30 days — not catastrophic (no DART cases) but a visible departure from the prior 60-day baseline. She filters by injury category: the increase is concentrated in slip / trip and strain categories, not lacerations or burns. That points at fatigue or environmental factors (lighting, housekeeping) rather than machine-energy issues.

She creates a task assigning Plant 3's safety lead to walk the third-shift handoff, check lighting and 5S compliance in the high-incident areas, and report back within two weeks. The task is site-scoped and linked to the Injury Analytics filter that drove the assignment; the safety lead opens it and lands on the same view Maria was looking at.

At 5 PM Maria scans the Compliance Calendar. Two items are this week: the Pennsylvania state FROI for Case 005 (the Plant 2 strain) is drafted and awaiting the injured employee's signature; the SDS retraining session for Plant 1's chemical handlers is on the books for Thursday afternoon. Both are on track.

Two longer-window items show in the dashboard but don't need today's attention: the OSHA 300A posting cycle (Feb 1 – Apr 30 next year) and the ISO 45001 surveillance audit prep in Q3. Maria reviews them weekly; nothing surprises her because nothing has been hiding from her dashboard.

She logs off. Tomorrow she'll do the same loop — KPI scan, queue review, floor walk, calendar check. The platform isn't the work; the work is still the safety culture, the supervisor conversations, the policy decisions. The platform just makes sure none of it falls through the cracks.