Tag: process improvement

  • Shift Notes to Shift Handoff: We Tested 3 AIs on the Same Messy Floor Notes

    Shift Notes to Shift Handoff: We Tested 3 AIs on the Same Messy Floor Notes

    A good handoff note has one job. Make sure the real risk gets seen before the small stuff does.

    Most don’t. A downtime event and a broken ice machine end up sitting in the same paragraph, same weight, same font. The next shift lead sorts out which one really matters standing up, usually with someone already asking them a question before they’ve finished reading.

    So we ran an experiment. Same messy notes. Same exact prompt. Three different AI tools. And one uninvited fourth guest we’ll get around to that later.

    The question wasn’t whether they could clean up the grammar. Any of them can do that in their sleep. The real question was whether they’d know what mattered.

    The Note We Fed Every Tool

    This is the raw version, typed the way a tired shift lead really types, half-punctuated and running downhill:

    Line 2 down 40 min around 10am, changeover took longer than usual, think it was the die swap on station 3, Marcus was working it. Forklift 4 still has the hydraulic leak, maintenance said Tuesday but never showed. Ran short on the 6mm fasteners again, used the backup bin, only maybe 200 left in it now. New temp on line 1 afternoon, did fine but slow, needs another day of shadowing before solo. Quality flagged 3 units off the line 2 run after the changeover, pulled and set aside by the press, not scrapped yet, waiting on Denise to look at them tomorrow. Everything else ran normal. Oh and the break room ice machine is out again.

    The prompt isn’t new either. It’s pulled directly from a piece we already published on decoding exactly this kind of inherited note. No reason to make you go dig it up, here it is in full, ready to paste into whatever you’re using tonight:

    You are an operations management assistant helping me process a shift handoff note. I am giving you the raw start-of-shift note from the previous shift lead, written by hand or dictated in a rush. Treat every line as something that might be incomplete.

    Structure your output using the exact sections below:

    1. WHAT HAPPENED

    – A plain-language summary of what actually occurred last shift, in the order it happened if the note makes that clear.

    2. STILL OPEN

    – Anything unresolved that needs my attention or a decision from me today.

    – For each item, note whether it’s urgent (needs action in the first hour) or can wait.

    3. TOP PRIORITY

    – The single most urgent item on this list, and one sentence on why it outranks everything else.

    4. WHO I NEED TO FOLLOW UP WITH

    – Based on what’s in this note, list anyone I likely need to contact today: maintenance, quality, the site leader, HR, a specific operator, anyone named or implied.

    – For each person, say what they need from me and how urgent that follow-up is.

    – Flag anything that looks like it needs to go to Quality or EHS specifically. Those two categories don’t wait.

    5. UNCLEAR OR MISSING

    – Anything in the note that’s ambiguous, contradicts standard work, or is missing information you’d expect to see (headcount, safety incidents, quality holds).

    – Do not guess at anything not stated in the note. Flag it as missing instead.

    Same prompt, same notes, word for word, into ChatGPT, Claude, and Gemini. Free tier on all three, memory cleared first, so nobody walked in with a head start.

    The Verdicts

    ChatGPT

    ●        Top Priority: the forklift’s hydraulic leak, reasoned as outranking the quality hold because the note never confirms whether the leak was tagged out or is still in use.

    ●        Gave the ice machine something none of the other tools bothered with the benefit of the doubt. Filed under “can wait,” but with a real condition attached, unless there’s a sanitation or heat-stress angle nobody mentioned.

    ●        Stayed disciplined on the die swap, calling it “not confirmed” twice instead of letting a guess quietly become a fact.

    ●        Didn’t ask a single clarifying question. Just went straight to work.

    Claude

    ●        Top Priority: also the forklift, same safety-first logic as ChatGPT.

    ●        Did something almost embarrassingly simple that the other two skipped entirely: it filled in every box the prompt asked for, site leader included. Nobody else even mentioned that category existed.

    ●        Caught a genuine distinction nobody else made. Denise’s review is scheduled for tomorrow, but confirming the units’ physical status and paperwork shouldn’t wait that long. Claude split those into two different clocks instead of treating them as one task.

    ●        The only tool willing to say “urgent leaning” out loud instead of forcing the fastener count into a clean yes or no.

    ●        Went further than repair status and questioned whether the leak had ever been formally logged as a safety hazard at all, or just quietly sat as a maintenance ticket nobody escalated.

    Gemini

    ●        Top Priority: the forklift too, and the most procedurally fluent of the three. It reached for real lockout and tag-out language, not a vague “someone should look at this.”

    ●        Most literal about the flagging instruction, tagging entries directly with [FLAG: EHS] and [FLAG: QUALITY] instead of burying urgency in a sentence.

    ●        Also the one that broke the rule. It reasoned that if the die swap caused the delay, current Line 2 production might also be non-conforming, connecting two dots the notes never touched. That’s the exact mistake the source prompt was written to prevent. Gemini’s the one that made it.

    ●        Caught something sharp nobody else did: “set aside by the press” isn’t the same thing as a documented hold. A press ledge is not a quarantine bin, no matter how urgent the tone around it sounds.

    Where They Agreed, Where They Split

    All three landed on the same Top Priority. Three companies, three completely different models, and every one of them decided an unresolved safety risk beats a contained quality hold, even one sitting closer to the loading dock. That’s not a small thing to agree on.

    Where they earned their keep was underneath that headline call. Claude was the only one that followed every instruction in the prompt to the letter. Gemini was the only one that invented a connection nobody asked it to make.

    Read only the top line of each response and you’d think all three did the same job. Read the whole thing. You find out which one you’d really trust with a decision that matters.

    We Also Ran It Through Grok, Out of Curiosity

    Full disclosure: not a tool I reach for often. My husband swears by it. Tells me not giving it a fair shot for months. Challenge accepted.

    Grok broke from the pack entirely. Its Top Priority was the quality hold, reasoned as needing immediate attention to protect product integrity and keep good material from mixing with bad. It also said flatly that nothing in the note pointed to an EHS incident, the opposite read from every other tool in the test.

    Not necessarily wrong. A hold sitting close to shipping is a real argument, not a lazy one. But it’s the one tool in the whole lineup that walked away from consensus on the single call that mattered most, which is worth knowing if it’s the one already living in your phone.

    How to Run This Yourself

    Pull a real handoff from your own floor, one with at least one safety-adjacent item and one genuine non-issue tangled up together.

    Clear memory and any saved custom instructions first, in every tool you’re testing. That’s the step people skip. It’s also the one that quietly rigs the result toward whichever tool already knows your plant.

    Paste the identical prompt above and your own notes into each one, back-to-back, same sitting. Read Top Priority first, before anything else. That single line tells you more about how a tool reasons than the rest of the output combined.

    The Actual Takeaway

    Run this on your own notes before you trust any one tool’s version of “what matters most” without a second opinion standing next to it. Free tier, five minutes, the exact prompt is sitting above, copy it and go.

    If two tools agree and a third doesn’t, that disagreement is the useful part. It’s pointing directly at the judgment call that’s genuinely hard. The kind of call that turns into a real decision nobody remembers making unless someone catches it first.

    The Floor Manager’s $20 AI Stack is worth a read if you haven’t settled on which of these is worth paying for yet. We ran this same test live on camera too, over on The Constraint Files, if you’d rather watch the answers land in real time than read about them after the fact.

    Comment below with which tool you’d trust to run your next real handoff, and which one just lost your trust after reading this.

  • Can AI Help You Get More Production from the Same 10-Hour Shift?

    Can AI Help You Get More Production from the Same 10-Hour Shift?

    You walk into your shift with a plan.

    You know the volume. You know the headcount. You know what yesterday missed and where today’s trouble spots are likely to be.

    For approximately seven minutes, everything looks great.

    Then operations happens.

    Someone calls out. A process starts falling behind. Equipment goes down. Quality starts moving in the wrong direction. Someone needs an answer about yesterday’s miss. Your labor plan needs to change. There’s a meeting in 20 minutes, and you’d really like to know what caused the problem before you’re expected to explain how you’re going to fix it.

    By lunch, you’ve pivoted so many times your hair feels like it’s on fire from the friction.

    Welcome to operations.

    Your Operation Has a Constraint. So Do You.

    You spend your shift thinking about capacity.

    How many people do you have? How many labor hours? How much volume? What’s your expected rate? Where’s the bottleneck? What can you move, change, eliminate, or improve to get more from the resources available?

    But there’s another finite resource affecting your operation every day.

    YOUR time.

    You get the same 10 hours whether everything runs beautifully or the building appears determined to test your commitment to operational leadership as a career choice.

    And some days, those hours disappear remarkably fast.

    A woman warehouse manager reviews shift notes and labor records while handling the demands on her time.
    AI can’t add hours to your shift. Used well, it can help you reclaim more of the hours you already have.

    You still need to:

    • Investigate yesterday’s misses.
    • Understand today’s numbers.
    • Find bottlenecks.
    • Analyze problems.
    • Explain findings.
    • Coach your people.
    • Follow up on corrective actions.
    • Work on process improvements.
    • Learn whatever new metric, system, process or methodology landed in your lap this week.

    Oh, and presumably run the operation.

    Wouldn’t want to forget that part.

    Enter AI. But Keep Your Cape.

    There is plenty of content online right now suggesting AI can do practically everything short of unloading the trailer for you.

    It can’t.

    There are operational problems AI simply cannot solve.

    AI cannot walk your floor.

    It can’t hear the strange noise coming from a machine that supposedly has nothing wrong with it.

    It can’t notice a queue forming where your dashboard says everything is fine.

    It doesn’t know that your highest-performing associate quietly developed a better method three months ago.

    It doesn’t know that everyone ignores Step 7 of the standard work because Step 7 makes absolutely no sense in the real world.

    It doesn’t know your operation.

    YOU do.

    REALITY CHECK

    AI is a tool. It isn’t Gemba. It isn’t judgment. It isn’t leadership. And a beautifully formatted answer based on bad information is still a bad answer.

    So why am I building an entire publication about using AI in operations?

    Because AI doesn’t have to replace your expertise to make YOUR expertise more valuable.

    The Opportunity Is in the Hours Between the Fires

    Think about how much time you spend on the work surrounding the work.

    You collect information.

    You compare numbers.

    You research something you don’t understand.

    You organize observations.

    You prepare for meetings.

    You turn findings into explanations.

    You rewrite the same information for different audiences.

    You try to build a coherent case for an improvement you’ve known for three weeks needs to happen.

    Individually, these tasks don’t create unreasonable time crunches.

    Collectively?

    There went Tuesday.

    This is where learning to work effectively with AI starts getting interesting.

    What Can AI Actually Help You Do?

    Used well, AI can help you:

    • Analyze problems faster. Organize what you know, explore potential causes, challenge assumptions and determine where to investigate next.
    • Get more from operational data. Ask better questions, compare information and look for patterns worth investigating.
    • Investigate bottlenecks. Work through process steps, queues, cycle times and constraints to narrow your focus.
    • Structure root-cause analysis. Turn a pile of observations into an organized investigation.
    • Develop improvement ideas. Explore alternatives, anticipate unintended consequences and identify what you need to measure before testing a change.
    • Communicate your findings. Turn your analysis into a clear update, proposal, presentation, action plan or training document.
    • Learn faster. Get help understanding unfamiliar metrics, calculations, Lean concepts, Six Sigma tools or process-improvement methods when you actually need them.

    There is something deliberately missing from that list.

    Let AI make the decision.

    That’s still your job.

    What If You Could Buy Back 75 Minutes?

    Imagine three tasks on a particularly busy day.

    • Analyze yesterday’s miss: 60 minutes → 25 minutes with AI
    • Organize findings for a meeting: 30 minutes → 10 minutes with AI
    • Draft an action plan: 30 minutes → 10 minutes with AI

    Without AI assistance: 120 minutes
    With AI assistance: 45 minutes

    That’s 75 Minutes Back in Your Shift.

    Those numbers are illustrative, not a promise of specific productivity gains. Your results will depend on the task, the information available and how effectively you use AI.

    A woman operations manager verifies source records against an organized analytical report.
    AI in operations management can help you spend less time on routine analysis and administrative work, leaving more of your shift for the work that requires you.

    AI hasn’t added a single minute to your shift.

    It hasn’t produced another unit.

    It hasn’t coached an associate.

    It hasn’t fixed the process.

    It HAS removed some friction from your day.

    And that raises a much more interesting question.

    What Could You Do With 75 More Minutes?

    You could walk the process you’ve been meaning to observe.

    Coach the associate who needs more than a hurried conversation in passing.

    Follow up on yesterday’s corrective action.

    Investigate the bottleneck everyone keeps working around.

    Talk to the people actually doing the work.

    Test an improvement.

    Look behind the metric instead of spending all your available time reporting the metric.

    That’s where AI can begin affecting operational results.

    Not because the chatbot produced another widget.

    Because the operations manager had more time to improve the system that did.

    A woman operations manager uses reclaimed time to coach an associate at a warehouse packing station.
    The value of AI in operations isn’t the time it spends doing the work. It’s what you can do with the time you get back.

    There’s a Catch

    Of course there is.

    If AI automatically gave everyone who opened an account another 75 productive minutes every day, this would be a very short article.

    The quality of what you get from AI depends heavily on how you work with it.

    You need to know what information matters.

    You need to provide context.

    You need to recognize when an answer doesn’t make sense.

    You need to protect sensitive information.

    You need to verify important conclusions.

    Most importantly, you need to know when AI is the wrong tool for the problem.

    Those skills improve with practice.

    And as you get better at working with AI, you can begin finding more practical places to use it throughout your shift.

    Not because you need to become an AI expert.

    Because you want to become better at operations.

    That’s Why The Daily Constraint Exists

    I’m not particularly interested in AI for the sake of AI.

    I’m interested in what happens when you put a powerful new tool into the hands of someone who understands operations and teach them how to use it well.

    That’s what we’re going to explore here.

    We’ll look at practical ways you can combine AI with:

    • Lean.
    • Continuous improvement.
    • Gemba.
    • Six Sigma.
    • Root-cause analysis.
    • Theory of Constraints.
    • Standard work.
    • Operational data.
    • And the experience and judgment you’ve already developed by actually doing the work.

    Sometimes AI will be useful.

    Sometimes it won’t.

    We’ll talk about both.

    Because the question isn’t:

    “What can AI do?”

    The question that matters here is:

    Can This Actually Help You Run a Better Operation?

    That’s the standard.

    You aren’t getting another two hours added to your shift.

    Your labor hours aren’t magically multiplying.

    And tomorrow will undoubtedly arrive with an entirely new collection of problems nobody bothered to put in today’s plan.

    But you can get better at using the resources available to you.

    Your own time is one of them.

    Subscribe to The Daily Constraint and we’ll keep exploring practical ways to use AI to understand your operation, solve problems, improve processes and get more accomplished with the hours you already have.

    After all, you’ve got enough fires to put out.

    Figuring out which AI tools are actually worth your time shouldn’t be another one.