Every operation I've run has had an SOP folder. Usually somewhere between a couple of dozen and a couple of hundred documents in it. Most opened twice a year, usually when an audit was on the calendar.

These documents are the closest thing your business has to a structured account of how the work actually gets done. They were written by people who understood the job. They've been revised, signed off, and version-controlled. They sit on the right side of compliance.

And almost nobody in the operation learns from them.

New starters don't read them. They watch someone who knows the job and pick it up that way. Supervisors don't open them unless they're pulling something out for a tick-box. The ones that do get used regularly are the ones printed and laminated next to the machine, because the digital copy is a chore to find.

That's the gap AI tools are quietly starting to close. Every capable model can now ingest a document and turn it into a conversation. That means the SOP library you've been treating as a compliance artefact is sitting there as a training asset. Feed it into a generative AI tool and you can ask things like: "What's the first question a new forklift operator would ask about this procedure?" or "Rewrite this as a five-minute briefing for a shift handover." The output won't be perfect. It'll often be 70% right, and the 30% that's wrong is the useful part, because it tells you where the written version drifted from what actually happens on the floor.

I've done this with my own. The most valuable thing wasn't the content the AI produced. It was the gaps it exposed. Steps that only made sense if you already knew the job. Contradictions between versions. Assumptions baked in and never spoken.

You don't need a platform for this. You need an hour, one procedure, and a willingness to look at what's been in the drawer for a decade with fresh eyes.

PwC and the Manufacturing Institute released the third in their series on frontline leadership and AI adoption on 7 April. Two numbers stand out: 54% of respondents say they have low or very low confidence that their frontline leaders are ready to guide AI-driven change, and 45% attribute failed AI initiatives to excluding those same leaders from the design of the rollout.

My take: The second number is the one to sit with. Almost half the failures trace back to a thing that costs nothing to fix — involve the people who run the shift in designing the tool that's meant to help them run the shift. If a pilot can't get that right at small scale, it won't correct itself when you multiply it by five sites.

Robotics and Automation News ran a piece earlier this month on how manufacturers are preparing their workforce for AI-powered operations. The headline finding: the tools are arriving faster than the people operating them are being prepared. Operators handed alerts they can't interpret. Maintenance teams given predictive dashboards with no guidance on what a failure signal actually looks like.

My take: This is the quiet version of an AI failure. The deployment went fine. The software works. Six weeks in it's being ignored because nobody has time to explain what it's telling them. If you're rolling something out this year, budget the training time honestly, then double it. That's closer to the real number.

SDS Manager expanded its AI capabilities this month for authoring and maintaining Safety Data Sheets aligned with GHS, OSHA and EU CLP. The pitch is straightforward: take the structured data you already hold, generate consistent documentation across regions and product lines, keep it audit-ready without the manual grind.

My take: SDS authoring is one of the few AI use cases I've seen that's actually built for what operations teams in chemical distribution and manufacturing spend real time on. High volume, heavily regulated, punishing when it's wrong. Worth a proper conversation with your compliance supplier before the next audit cycle. If they don't have an AI roadmap yet, they will by the end of the year — and you'll want to be on the version that saves you time rather than the version you're still paying to do by hand.

Pick one procedure. Ideally the one new starters ask the most questions about, or the one you keep finding yourself walking people through personally.

Open whatever AI tool you already have access to — ChatGPT, Gemini, Copilot, the assistant built into your safety platform if it has one. Paste the document in. Then ask two things:

Compare both outputs to what your new starters actually get stuck on, and to what your supervisors actually say at handover. The distance between the document and the real conversation is your training gap. You'll find it in under half an hour.

The Shift Report is now also going out by email. If LinkedIn isn't where you do your proper reading, or you'd rather have it waiting in your inbox every other Monday, you can subscribe here: https://vinces-newsletter-5a6441.beehiiv.com/

Same cadence. Same ratio of signal to noise. Just not dependent on the algorithm deciding to show it to you.

As always, reply if something landed, or if it didn't. I read everything.

Last thing: I'm building SOPwise — it turns the SOP library into something that actually trains the people using it. Have a look if that's a problem you're trying to solve, and reply if you want to talk about it.

Issue #3 in two weeks.