What safety data your plant already has
The first step needs no new sensors and no year-long IT project. It is enough to list which records are created during daily work and who maintains them. In a typical industrial plant there are four streams: work permits (who works, on what, where and when), gatehouse entry control, the position of workers on site, and hazard and near-miss reports.
Each stream answers a different question. Permits show the intent and scope of work. Gatehouse records show who physically entered and when they left. BLE beacon location shows where that person actually is, without GPS. Reports show what people see on site and whether the team closes issues. Only the combination of these answers gives a picture that no single register provides.
- work permits: date, type of work, contractor, work location, authorised persons
- entry control: first scan as entry, second as exit, reason for refusal
- location: the worker's device periodically sends the beacons it hears, signal strength, battery level and status
- reports: location from the plant dictionary, type, description, photos, closure status
Step 1: a shared location dictionary
The most common reason safety analytics fails is the lack of a shared language for places. A permit names a building, the gatehouse records a gate, and a report describes a hall in the reporter's own words. Without a dictionary you cannot count how many jobs were done at a specific machine or compare reports with the place of work.
In an analysis of 106 permits from a continuous-process plant, as many as 44 gave only a general location, for example the building instead of the machine. The practical conclusion is to break the dictionary down to the level where someone can point to a filter, a section, a hopper or a chamber. The plant's online map in a 2.5D view can then serve as a shared register of places used by permits, reports and worker location.
Start with one hall or one line. List the places in the order people actually name them at work, not by the numbering in technical documentation, and only then agree the official names with maintenance.
What the analysis of 106 permits over 9 weeks showed
Step 2: connect the streams at the event level
Connecting data does not mean building a data warehouse. It is enough that every record has three shared keys: time, a place from the dictionary and an entity (a person or a company). Then a permit, a gatehouse entry and a worker's position can be put on a single event timeline.
Worker location comes from the signals of many beacons averaged on the map, not from the single strongest signal. This makes the position more stable, and comparing it with the permit can show whether a person authorised for the job is in the place stated in the application. This is not an assessment of the person but a quality check of the permit itself: does the description of the place match reality.
Minimum set of fields
For a permit: date and time of issue, type of work, company, place from the dictionary. For an entry: scan time and verification result. For a report: place, type, time of report and time of closure. That is enough to calculate the distribution over time, space and companies.
What not to connect at the start
Do not combine individual worker data with performance appraisals at the beginning. Safety analytics should detect patterns in the process, not judge people. This approach also makes the conversation with trade unions and the data protection officer easier.
Step 3: questions worth asking the data
Reports start with questions, not charts. Three questions give the most at the beginning: when external companies work, where particularly dangerous work is concentrated, and who generates the largest share of permits.
In the plant analysed over 9 weeks, 82% of permits were issued between 6:00 and 10:00, and 62 of 106 fell on Tuesday and Wednesday. The four largest contractors accounted for 63% of permits. No permit was issued at the weekend. Such numbers translate into decisions: when to schedule approver duty, where to strengthen the security presence, and for whom to prepare training and authorisation matrices first.
- When: distribution of permits across the day and week, peak load on approvers
- Where: map of hot work and work at height, places with a general location
- Who: share of companies in permits, training gaps, entry refusals
- How fast: time from hazard report to closure
Which reports to build and how to deliver them
A report nobody reads is not analytics. That is why the format is chosen for the recipient. The EHS manager needs a weekly summary with a list of open reports and training gaps. The plant director needs one page a month with a few indicators and a trend. The board needs a quarterly review assessing whether the risk is shifting.
Recurring reporting makes it possible to send such summaries automatically by e-mail: with tables in the body and a PDF attachment with the plant logo, optionally with an XLSX sheet. The location scope and recipients are set separately for each rule, so the hall manager receives their hall and the director receives the whole site. Such reports replace manually assembled e-mail summaries and cut preparation time to a few minutes of checking instead of several hours of compiling.
Weekly report for the EHS manager
Number of permits by type, open hazard reports with age in days, list of training gaps among external companies, entry refusals with the reason.
Monthly report for the director
Three to five indicators with a trend: share of permits with a general location, median report closure time, concentration of permits in a few companies, use of walkthroughs and inspections.
Audit evidence and the limits of analytics
Well-built analytics provides audit evidence: it is clear who approved a permit and when, what the data from the work site showed and how the team responded to a report. This supports meeting occupational safety management requirements, but it does not replace inspections by the EHS service or obligations arising from regulations.
The limits matter. Location data shows where the device signal was heard by beacons, not a person's intentions. Inspections and walkthroughs complete the picture but do not replace physical checks on site. Reports must be verified against reality from time to time: once a quarter it is worth comparing a sample of records with a walkthrough in the field.
Video
Frequently asked questions
Do we need new sensors for safety analytics?
Not at the start. Permits, gatehouse records and reports already exist. BLE beacon location adds a layer of place, but the first reports can be built on data the plant has today.
Which report should we start with?
With the weekly summary for the EHS manager: permits by type, open reports with their age and training gaps among external companies. It is quick to build and shows right away whether the data is consistent.
Will analytics replace EHS service inspections?
No. It provides audit evidence and points to where to look, but walkthroughs, inspections and checks required by regulations remain the plant's obligation.
How do we avoid assessing individual workers?
Report on processes, places and companies, not people. Individual data stays in the operational system, and only aggregated summaries go into analyses.
Let us talk about your plant's data
We will show which reports can be built from the permits, entries and hazard reports you already have. Book a conversation with the InnerWeb team.
Let's talk