The Shift Report · Issue 16 · 8 November 2026
How to review your safety action list with AI before the site safety meeting
A row marked Closed on the action tracker often just means somebody stopped chasing it. AI can go through the whole sheet before the meeting and flag what’s overdue, what’s closed without proof, and what keeps coming back — deciding what to raise first is still yours.
It will be similar on many sites. Dave, who runs the floor, sits down to prepare for the site safety meeting later in the day. He goes down the actions list in a shared spreadsheet where everything ends up: follow up on near-misses, internal audit findings, the customer’s audit, the insurer’s visit, last week’s meeting.
A few rows are marked as Closed with no or very limited comments next to them. Sometimes the same thing turns up again as a new near-miss, with a new reference. On a quiet week Dave goes through it row by row. On a busy week, and that is mostly the case, he walks into the meeting with no prep at all.
A month ago, Dave’s driver reported that no trained banksman was present as required at a customer site. The action was closed by speaking with the ops manager at the customer site, only to have the same problem back on the table a month later.
The rows that belong together often aren’t next to each other. Two near-misses at the same loading door can end up far apart on the list, with different owners. Taken one row at a time, each looks like it’s being dealt with.
AI can go through the whole table in one go. Dave can upload it and ask what’s overdue, what’s marked Closed without anything saying it was fixed, which rows are about the same place or the same problem, and what has no owner. He can tell it not to take Closed at face value, then ask what to raise first.
How to do it before the next safety meeting
- Open the tool. Most people with a Microsoft work account have Copilot Chat. I find ChatGPT Plus or Claude much better at reading a table like this. An ordinary new chat is enough; there is nothing to set up.
- Give it the instructions. You can start with my example below. Paste it into the message box, and add today’s date.
“This is our site action tracker. Go through it with me before I go into the site safety meeting.
Tell me:
- which actions are overdue as of today, [day and date]
- which ones are marked Closed but the update doesn’t actually say the problem was fixed
- which rows are about the same problem or the same place, put them together and tell me if it looks like it’s getting better or worse
- anything with no owner or no due date
Don’t assume something is done because it says Closed. Don’t make up what was done. If you can’t tell from the sheet, say so.
Then give me the three things I should raise first in the meeting, most serious first, in a few lines each.”
- Add the actions list. Click + by the message box (the paperclip in Copilot) and choose the Excel file. If it lives on a shared drive, download a copy first. If the tracker has several tabs, say in the message which one to use. If the upload doesn’t go through, select the rows in Excel, including the header row, copy them and paste them into the chat.
- Tell it what only you know. It can only see the sheet. Click the microphone, or type, and add what you know that isn’t on it: who has left or moved jobs, what you saw on your walk round, what you think arrived but may never have been fitted. Don’t tidy it up. Your words appear in the message box; send them when you’ve finished.
- Ask where it got each point. “Which of these did you get from the sheet, and which from what I just told you?” Then “List every row reference you read.” Ask about anything that doesn’t match what you know.
- Ask for it in the form you need. Three lines to open the meeting, or a table of the overdue rows to put on the screen.
- Check it against the sheet. Look up the rows behind the top three, and anything else you’ll read out: reference, owner and date.
- After the meeting, write the note for the floor. Go back to the same chat (it’s in the chat list on the left) and tell it what the meeting changed: which rows were reopened and who took what. Then send:
“Now write a short note for the noticeboard and the nights WhatsApp group, so people see what’s happened to the things they’ve reported. Two parts: what’s been fixed, and what’s still open and who’s on it. Only put something under fixed if the sheet or I have actually said it’s fixed. Plain words, no names of who reported anything. Don’t add any instructions or rules of your own, and use the sheet’s own words for what was done.”
About 15 minutes before the meeting, most of it getting the file in, talking it through and checking the top three, and a few minutes after it for the note.
Two things worth knowing
1. AI on the cameras a site already has, logging the near-misses nobody writes down
Near-miss reporting is one of the most useful habits on any site, and one of the patchiest: it depends on someone stopping, remembering and filling in the form. AI is already being used to fill that gap. One of the companies doing it is Protex AI. It connects to a site’s existing CCTV. A small box on site processes the video, picks out people, forklifts and pallet trucks, and records each near-miss as an event with a short clip. Managers see the events by area and by shift, instead of watching footage.
On a podcast with The New Warehouse, Protex’s founders gave two examples. At one customer’s warehouses in the US Midwest, near-misses went up when the weather turned cold, and repeated talks about paying attention changed nothing. The clips showed people walking the floor with their hoods up. The company banned hoodies, handed out branded beanies, and near-misses fell by 73%. Another customer had two sites, and the one with fewer near-miss reports looked safer on paper. The cameras showed that people there were simply reporting less.
The figures are Protex’s own, with no independent check, and for a mid-size site this is a real purchase even on cameras it already has. What I take from it is the second customer: a short near-miss list can mean a safe site, or a site where people have stopped reporting. (The New Warehouse, 2 November)
2. A sensor on the pump warned before it failed, and the repair waited for a planned stop
At Britvic’s soft drinks plant in Rugby, a small sensor on a critical pump picked up abnormal running and sent an alert before the pump failed. The maintenance team fixed it in 45 minutes during a planned production stop. Britvic and the supplier, Alfa Laval, say that avoided six hours of unplanned downtime, 288,000 bottles of lost product and an unplanned clean of the line.
The sensor is attached to the pump and measures vibration and temperature around the clock. It sends the readings wirelessly to a gateway on site, which passes them to the supplier’s software. The software looks for vibration patterns that point to wear, a bad installation or a process problem, and sends a notification with step-by-step instructions on what to check. Britvic bought it rather than built it: the sensors and gateway, plus a yearly analytics subscription. The site’s engineering specialist says that one alert paid for the whole system and the first year’s subscription.
Britvic is a big company, but the set-up works pump by pump. A mid-size site doesn’t need it everywhere, only on the few pumps, gearboxes or fans whose failure would stop the whole line. The figures are Alfa Laval’s and Britvic’s own. (ProFood World, 23 October)
Final thought
A row marked Closed often only means somebody stopped chasing it. What was actually done is in the update column, when it’s filled in, or out on the floor. For the people who reported it in the first place, the note on the noticeboard is sometimes the only answer they get.
I spend a lot of my time working out where operations lose time, especially in training, and how the new tools can help. Direct message me if you want to find out more or set up a call.
SOPwise turns a Standard Operating Procedure into a training package: task-based steps, critical mistakes, and scenario questions. £35 per SOP, no subscription. Built for UK manufacturing and chemical distribution.