Every operation has them.

The WhatsApp message that triggers the real process. The printed checklist taped to the wall. The form people fill in by hand because there's only a non-editable PDF available, and nobody remembers where the Word doc is. The person everyone goes to because looking up information is too slow, or the SOP is too vague.

Managers often treat these workarounds as bad habits to be tidied away. Sometimes they are. More often, they're signals — quiet evidence that the formal system isn't doing the job people need it to do.

There's a useful new piece of context for this. On 1 May, Made Smarter — the UK government-backed programme for SME manufacturers — published an AI adoption toolkit built around the deliberately unglamorous instruction: start with a real operational problem. Two weeks later, Indeed's Hiring Lab put out the most uncomfortable workforce projection of the year — the sectors with the worst labour shortages are also the sectors AI is least able to help with. Manufacturing, construction, healthcare. Physical work, site-specific judgement, hands-on problem-solving.

So here's the reality in May 2026: fewer people available, more pressure on the people you've got, and an AI conversation that mostly isn't aimed at your industry.

In that environment, the first useful AI move is making the gap between what's written down and the real work visible. And the fastest way to find that gap is the workaround.

The workaround is the signal. Read it.

Made Smarter's new AI toolkit — built with Professor Chris Dungey, the Department for Business and Trade's AI champion for advanced manufacturing — is structured around a "Scan, Pilot, Scale" framework. Scan the operation for real friction, pilot in a contained way, scale only what proves its value in live conditions. Case studies cover Ritherdon & Co, Arden Dies, ELE Advanced Technologies and others — real Cheshire and North West manufacturers, not Silicon Valley reference customers.

My take: The most useful sentence in the entire piece is "practical adoption matters more than technical novelty." That's the one to print and pin up. Almost every failed rollout I've seen had the same fingerprints — chosen for what it could theoretically do, not for the operational pain it was supposed to remove. Scan, Pilot, Scale is just operational discipline applied to a new toolset. If your supplier or trade body isn't pointing you at something this concrete, ask why.

Indeed's Hiring Lab published "The Great Mismatch" on 14 May — an original piece of workforce modelling worth a coffee break. It's US data, but the structural picture maps onto the UK closely: ageing workforce, lower fertility, reduced migration. The numbers that matter: retirement rates in manufacturing are already running above 3% per year, and the sharpest near-term drop in the labour force lands between now and 2032. AI is most useful in sectors with no shortage — finance, professional services, information. It's least useful in the sectors where the shortage is real — construction, healthcare, manufacturing, frontline operations.

My take: This is the part the loud AI commentary keeps missing. In our world — manufacturing, warehousing, logistics, chemical distribution — AI isn't going to fill the gap left by retiring shift leaders or the operators who didn't come through the apprenticeship pipeline. What AI can do is take repetitive cognitive work off the plate of the people you do have — so they spend their time on the parts of the job that only humans can do. That's a smaller, more honest brief.

On 5 May, Apple's Manufacturing Academy held its inaugural Spring Forum in East Lansing, Michigan — hundreds of small- and medium-sized manufacturers, paired with Apple engineers and Michigan State experts, working on real machines. The case study they put forward was Block Imaging, a Michigan company that services and refurbishes CT scanners and MRI machines. Their training director described what they got as "practical tools and techniques we've been able to apply immediately on the floor."

My take: Look at the format, not the badge. Practitioner-to-practitioner. In person. On working equipment. That's the only training I've ever seen change behaviour. The reason workarounds exist in the first place is usually that nobody who properly understands the job was in the room when the system was designed. The fix is the right people, in the right room, on the right machine. When you're picking AI partners or training suppliers this year, the question to ask is "whose face shows up on my site," not "what's your platform."

A two-part exercise. About ninety minutes total. No software, no budget, no permission needed.

Step one — run a workaround walk.

Walk one process end-to-end. Pick something everyday — order picking, a maintenance routine, the start-of-shift sequence, a goods-in check. Look only for unofficial aids: spreadsheets, sticky notes, taped checklists, side messages, saved email templates, the person being asked the same question repeatedly. Don't fix anything. Don't even tidy. Just count them and write them down.

Step two — pick the most repeated workaround, and write the shop-floor reason.

Before you reach for any tool — AI or otherwise — write one paragraph using this structure:

If you can't write that paragraph, the use case isn't ready. If you can, you've now got a brief that any supplier, supervisor, or internal team can respond to without three meetings.

That's it. Diagnosis, then brief. The order matters.

I'm building SOPwise — it turns the SOP library into something that actually trains the people using it. If the workaround you just identified on your walk is one where the SOP exists but nobody's using it, that's the exact problem I'm working on. Reply and tell me about it.

Issue #5 in two weeks.