Most gemba walks produce a list. Then the list goes nowhere.

The walk itself usually isn't the problem. Leaders show up, ask questions, see real things. The trouble starts after: no plan for what to look at, no ground rules for how to ask, no time set aside to write anything up, no follow-through once everyone's back at their desk.

Here's how to run one that changes your floor.

What a Gemba Walk is (and isn't)

Gemba is Japanese for "the real place." A gemba walk means going to where the work happens instead of reading about it in a dashboard or a status email.

It isn't a tour. It isn't an audit. The moment operators feel scored, they stop showing you the real process and start showing you the version they think you want to see.

Genchi genbutsu is the root idea: go to the source and judge from what you see there, not from what a report says about it.

Step 1: Pick one process

Before you walk, name exactly what you're studying.

"Order entry through shipping for Product X" is a scope you can learn something from. "The whole fulfillment process" is not. You'll cover a lot of ground and understand none of it.

Anchor the scope in a real business question. What are you trying to do better for the customer? What are you trying to do better to survive? A vague purpose gets you a vague walk.

Step 2: Build a walking team

Six to ten people works well.

Leave out anyone who tends to shut people down. A senior leader who jumps in with "why didn't you just" kills the walk before it starts. Operators go quiet the second they sense judgment, and they don't come back for the rest of the session.

Give people real notice. A message like this works:

"On [date], we're walking through [process]. We'd like about 30 minutes of your time to talk through your work. We want to learn from you, not check compliance: what makes this hard, and what would you change."

Step 3: Set ground rules before anyone sets foot on the floor

Four failures wreck more walks than anything else: turning it into an audit, jumping straight to solutions, correcting people mid-interview, and letting the ground rules slide the moment things get uncomfortable.

A short list to run through with the team before you walk:

  • Question the process. Ask "what about this process makes that likely" instead of "why didn't you catch this."
  • Stay curious. A question like "why don't you just do it this way" pushes your own answer. Ask open ones: what happens when this goes wrong, what have you tried already.
  • Let people finish, even when they're wrong. The gap between what someone believes and what's true is often the finding. Correct it later, in training, not in front of their coworkers.
  • Stay present. Phones away, no side conversations. The floor gets full attention or it doesn't get anything useful.

No checklists or scorecards. No photos of "violations." Bring either and the walk becomes an inspection, and word travels fast.

Step 4: Ask what before you ask why

"What" questions get you facts. "Why" too early sounds like an accusation, even when you don't mean it that way.

Start with what: what's the purpose of this step, what does a good day look like, what problems do you run into, what would make tomorrow easier. Once you have the facts, why starts to land as curiosity instead of a challenge.

When you do get to root cause, keep asking why. Not a fixed five times, whatever it takes to reach something you can change. One team traced a server outage back through five whys: the server had overheated, the cooling fan had failed, the fan's bearing had never been lubricated, and the maintenance schedule that skipped it had been copied from a different piece of equipment. Three "whys" wouldn't have gotten there.

Then don't jump in with the fix. Ask "what do you think would help" instead. Handing someone a solution on the spot teaches them to stop bringing you their own ideas.

Step 5: Follow the work

Most waste hides at the handoffs between departments.

When a task moves from one group to the next, go with it. Watch what survives the handoff and what gets lost. That's usually where the real story is, and it's the part no single department can see on its own.

Time each step if you can. Compare lead time (the full elapsed wait) against process time (how long the work would take with zero interruption). The gap is almost always bigger than people expect, often 90% or more of the total.

Step 6: Write it down within 24 hours, act within 48

Memory fades fast. Capture what you saw, what surprised you, what people raised, and what you'd ask differently next time, the same day if you can.

Then go back to the people who walked with you, within 48 hours, and tell them two things: what you learned, and what happens next. Specific actions, not "we'll look into it." Vague follow-up teaches people the walk didn't matter.

Fix at least one of the top three issues inside a week. Speed here builds more trust than any amount of talking about how much you value feedback.

What breaks the routine (and how to protect it)

A single walk changes nothing on its own. The pattern that matters is the one that survives the calendar.

A few things quietly kill it: showing up only when something's wrong (your presence becomes a warning sign), skipping the details in favor of a quick lap around the floor, going quiet after the walk instead of closing the loop on what you found.

Block the time. Protect it the same way you'd protect a customer meeting. The first time you cancel a walk for something "more urgent," your team learns exactly how much the walk was worth to you.

Where the admin work usually kills it

This is usually where it falls apart, and it happens after the walk ends: writing the report.

An hour at your desk after every walk: copying notes into a template, pulling photos off your phone, chasing people for details they mentioned in passing, reconstructing the sequence from memory. It's tedious enough that it gets skipped.

Gemba Walk AI exists for that gap. Your team captures observations on their phone as they walk, tagged by waste type and location as they go. The report writes itself in one click. Every issue carries a status, and the still-open ones from the last walk in that area are the first thing you see when you open the next one.

The walk itself still happens the same way, on the floor, in person. This just removes the part that usually stops it from happening a second time.

Run one walk this week. If the follow-through is where yours keeps breaking, that's the part worth fixing first.