5S Audit Checklist: Build One for Your Shop Floor With AI

A small factory can build a 5S audit checklist with AI by listing each work area, asking AI to draft three to five checkable items per S from photos and short notes, and scoring every item 0 to 4 each month. A supervisor walks the area to verify, and AI turns the scores into a one-page trend.

The AI is a drafting and summarising tool here. It does not audit anything. A checklist item such as "tools are tidy" is useless; "every tool on the shadow board is in its outline" can be checked in ten seconds. Most of the value comes from turning vague standards into yes-or-no items, and AI is fast at that when you give it real photos and plain notes.

Follow me on Instagram@sagnikteaches

What a usable 5S checklist looks like

The five S words are sort, set in order, shine, standardise and sustain. A checklist that works has these properties:

Connect on LinkedInSagnik Bhattacharya

• Each item is observable. A person can see it from the aisle without opening a document.
• Each item has a pass description. "Floor markings visible and unbroken along all aisle edges" beats "floor marked well".
• Each item has an owner. The area's team lead, a maintenance person or the supervisor.
• The list is short. Three to five items per S, so 15 to 25 per area. Longer lists produce rushed scoring.
• It is specific to the area. A welding bay, a packing bench and a tool crib share the five S but not the items.

Scoring on a 0 to 4 scale per item gives more information than pass/fail without becoming fiddly. An example scale: 0 = not done, 1 = started, 2 = partly done with clear gaps, 3 = done with minor exceptions, 4 = fully done and has been for a month. Write the scale at the top of every sheet so the scores mean the same thing to different scorers.

Subscribe on YouTube@sagnikteaches

Main example: a small metal-fabrication shop with six areas

Illustrative. A 30-person fabrication shop has six audit areas: goods-in and raw stock, cutting, welding bay, finishing and paint prep, packing and despatch, and the tool crib. The owner wants a monthly score per area and a trend. No one has time to design a form from scratch, so the production supervisor takes five photos of the welding bay on a normal working day and writes down what she wants to be true.

Step 1: give the AI photos and plain notes

Use a business-tier assistant that accepts images. Take wide shots of each zone, plus close-ups of the places that tend to go wrong: the floor, the bench, the consumables shelf, the bins and the notice board. Keep people out of the shot, or crop them out, and avoid screens that show customer names. Then write the notes in the supervisor's own words and send both with this prompt:

I run a small metal fabrication shop. These photos show our welding bay on a
normal working day. My notes on what "good" looks like for this area:
- offcuts go in the labelled bins, not on the floor
- welding screens stay on their stands, cables coiled on hooks
- consumables (wire, tips, gloves) live on the red shelf, each in its marked
  space
- extinguisher and first-aid kit never blocked, 1 metre clear
- floor aisle markings visible
Draft a 5S audit checklist for this area: 3 to 5 items for each of Sort, Set
in order, Shine, Standardise, Sustain. Each item must be something a
supervisor can check by looking, with a one-line description of a score of 4
and of a score of 0. Mark anything that depends on information not visible in
the photos as NEEDS CHECK ON SITE. Do not assume the extinguisher is in date.

Illustrative output, shortened: "Sort 1. Only items needed for today's jobs are on the benches (4 = bench holds current job material only; 0 = unrelated items, packaging or tools from other areas present). Sort 2. No offcuts or scrap on the floor outside the bins. Set in order 1. Every consumable is in its marked space on the red shelf. Shine 1. Floor and bench are free of spatter, dust and spills at end of shift. Standardise 1. Aisle markings and shadow-board outlines are visible and not worn. Sustain 1. Last month's actions are closed or have a date. NEEDS CHECK ON SITE: extinguisher service date, glove stock level, whether the marked spaces match what was agreed."

Edit the output. Delete items the supervisor does not care about, add the three things the AI could not see, and rewrite any descriptor that two people could read two ways. Expect 20 minutes to get one area right, and 10 minutes per area after the first, because the wording of Standardise and Sustain is much the same everywhere.

Step 2: put the checklist into a scoring sheet

Ask the AI for the sheet layout, then build it in a spreadsheet. One tab per area with these columns, and a Summary tab:

The two score columns matter. A self-score that always exceeds the supervisor's by a point or more means the standard is not yet shared. Do not turn that into blame; ask the team lead to point at the item, and fix the wording.

Step 3: the monthly walk, with photo evidence

On the audit day, the supervisor walks each area with the sheet on a tablet or printout, scores each item, and takes one photo per item that scored 2 or below. Name photos with an area code and a number (W-07 for welding bay item 7). The next day, send the AI the scores and a short note per low item, and ask for the action list and the summary. The photos stay in a shared folder; they are the evidence if someone disputes a score next month.

The monthly trend summary AI writes for the owner

After scoring, the Summary tab shows each area's percentage: total score divided by the maximum possible (items times 4). Paste a small table of the last six months into the assistant:

Below are monthly 5S scores (percent of maximum) for six areas over six
months, with the supervisor's notes for the latest month. Write a one-page
summary for the owner: (1) a two-sentence overall picture, (2) the two areas
that improved most and the two that slipped most, with the numbers, (3) which
of the five S's scores lowest across all areas this month, (4) the three open
actions that are overdue. Use only the numbers and notes given. If a change is
under 3 percentage points, call it "no real change".

Goods-in: 71, 74, 76, 79, 80, 82 | Cutting: 78, 80, 77, 79, 81, 80 |
Welding: 64, 66, 70, 69, 66, 62 | Finishing: 58, 61, 65, 70, 74, 77 |
Packing: 80, 82, 83, 85, 84, 86 | Tool crib: 90, 88, 89, 86, 83, 79
Latest month by S, all areas: Sort 84, Set in order 79, Shine 81,
Standardise 66, Sustain 72
Notes: welding bay aisle markings worn; overdue actions are the welding
aisle repaint, the second offcut bin by the saw and outlines for new tools
on the tool crib shadow board.

Illustrative output: "Finishing improved most, from 58% to 77% over six months. Goods-in rose from 71% to 82%. Two areas slipped: welding fell from 70% in month three to 62%, and the tool crib dropped from 90% to 79%. Packing rose from 80% to 86%, and cutting shows no real change. Standardise scored lowest overall, at 66%. Overdue actions: the welding aisle repaint, the tool crib outlines, and the second offcut bin." Check the sums: finishing 77 - 58 is 19 points, goods-in 82 - 71 is 11, tool crib 90 - 79 is 11. Note that welding only slipped from its month-three peak; against month one it is down 2 points, which your own rule calls no real change, so ask which comparison the owner wants and keep it the same every month. The assistant is good at this kind of reading, but verify any number before the owner sees it.

The most useful sentence is the one about the lowest S across all areas. If Standardise is lowest everywhere, the shop has a habit problem (outlines, labels, markings that fade), not an area problem. If Sustain is lowest, the audit itself is not being followed up. For a related monthly pattern, the 30-minute monthly AI quality review uses the same idea for customer-facing work.

How the AI goes wrong on a shop floor

• It praises what the photo flatters. A wide shot of a clean floor hides the dust behind the bench. Ask it to list what it cannot see.
• It invents standards. Asked for a "5S standard", it may add safety or legal requirements you did not specify. Keep safety items, such as extinguisher dates and clear exits, as your own, checked against your own safety procedures.
• It confuses tidy with organised. A neat pile of unlabelled parts looks good in a photo and fails Set in order. Add items about labels and locations, not appearance.
• It over-reads small changes. A 2-point rise is not a trend; the "under 3 points means no real change" instruction prevents it.
• It writes generic actions. "Improve housekeeping" is not an action. Require an owner, a date and a physical change for every action.

From vague items to checkable ones, area by area

Before and after: three weak items

"Area is clean" becomes "No swarf, spatter or spills on the floor or machine bed at the end of the shift; bin emptied when over three-quarters full." "Tools are organised" becomes "Every hand tool is on the shadow board; any gap in an outline has a sign-out tag." "Standards are followed" becomes "A laminated one-page area standard with photos is on display and matches the current layout." Ask the AI to do this rewriting for every item your team has used before: give it the old checklist and the instruction "make each item checkable by looking, with a score 4 and score 0 description".

A packing bench with mixed stock

A packing bench holds cartons for three products, tape, labels and a returns pile. The audit finds the returns pile has been there for two weeks (Sort fail) and the labels are in a drawer (Set in order). The AI suggests a returns tray with a date label and a rule that nothing stays over 48 hours, which the supervisor accepts. It also drafts a two-line sign for the tray. The effort is small, and the next month's score on the item rises from 1 to 3.

A tool crib where the score slides

The tool crib dropped from 90% to 79% over five months, a pattern that looks like drift, not a one-off. The notes mention new tools without outlines, so the underlying cause is that the Standardise step has no owner for new purchases. The action is to add "outline made on arrival" to the goods-in procedure for tools. The AI helped by grouping five months of notes into one cause; the decision to change the procedure was the supervisor's.

The inspection room, scored with the same sheet

The same method covers the shop's own inspection room, which audits often skip. The areas are the gauge cabinet, the sample shelf and the drawing file. Sort is "no parts on the sample shelf without a job number and a date"; Set in order is "every gauge in its numbered foam slot, matching the calibration log"; Shine is "surface plate wiped and covered at the end of the shift". Scoring this room with the same 0 to 4 sheet tests your wording: if an item only makes sense to the person who wrote it, it is too vague to score anywhere.

Photo-based sample scoring as a calibration exercise

Once a quarter, have the three scorers independently score the same ten photos, then compare. Illustrative result: scores on "bench holds current job material only" range from 1 to 4 across the three. That tells you the wording is loose. Ask the AI to propose a tighter descriptor with a numeric rule, such as "no more than one item on the bench that is not part of a job currently running", and test again next quarter.

Turning item scores into an area percentage, with the sum shown

Keep the arithmetic visible so nobody argues about it. If the welding bay has 20 items scored 0 to 4, the maximum is 80. Suppose the supervisor's scores add up to 52. The area score is 52 / 80 = 65%. Next month the shop paints the aisle edges and adds the second offcut bin, and five items each rise by one point: 57 / 80 = 71.25%, shown as 71%. Where an item does not apply to an area (no chemicals stored, so no chemical-label item), remove it from both top and bottom: 18 applicable items give a maximum of 72, and 50 / 72 = 69%. Ask the AI to write the percentage formula for the sheet and to explain how it handles "N/A", then test it with the numbers above.

Pick a pass mark before the first audit, not after seeing the scores. One workable rule is 80% for the area and no single S below 70%, so an area cannot pass by being excellent at Shine and ignoring Sustain. Illustrative: finishing scores 82% overall but Standardise at 55%; it passes on the total but fails the minimum, and gets an action on its labels and markings.

A red-tag day to start the Sort step

For the first audit, hold a one-hour red-tag session in each area before scoring. Every item that is not needed for current work gets a red tag with the date and the reason, and goes to a holding spot. After 30 days, anything untouched is sold, scrapped or stored elsewhere. Ask the AI to turn the tags into a list: paste a photo of the holding area and your notes, and ask for a table of item, likely owner, suggested decision and what to check before disposing of it. Illustrative: a 30-person shop finds 41 items in the first session, 29 are moved back to proper stores, 9 are scrapped and 3 are returned to a supplier. Do not let the AI decide disposal; it does not know that a gauge is on loan or a jig is for a job in March.

The first audit after a red-tag day will score Sort highly, which flatters the shop. Say so in the summary, so the owner does not mistake a one-off clear-out for a habit.

Supervisor verification: what the human must do

The audit's credibility rests on a person who walks the area. These checks do not change from month to month:

1. Walk the area in the same order each month so changes stand out.
2. Open one drawer, cupboard or bin in each area; photos rarely show inside.
3. Check dates on anything that has one: extinguishers, first-aid stock, calibration labels, chemical containers.
4. Score before reading last month's score, so earlier results do not anchor the number.
5. Ask one operator in the area what the standard is for one item. If they cannot say, Standardise is not done, whatever the paperwork says.
6. Sign the sheet. A named person owns each audit.

Pair findings with other records where the floor connects: repeat stops in a downtime log analysis sometimes trace to a missing tool location or a cluttered changeover area, and gauge cabinets belong in the calibration log routine as well as the 5S sheet.

Setting it up in an afternoon, then keeping it up

For a first run, do this in order: choose the areas (six to eight is plenty); photograph each; let AI draft a checklist per area; spend 10 to 20 minutes per area editing; build the spreadsheet once, copying the tab for each area; run a trial audit with two people scoring independently; adjust the wording; then start the monthly cycle. Budget about half a day for the build and 20 minutes per area for each monthly audit.

To keep it alive, set one pass mark and one visible board. A simple board with each area's last three scores and the top action does more than a long report. Review overdue actions at the weekly production meeting, three minutes at most. When a score stays above 90% for three months, ask the AI to propose harder items for that area, such as time-to-find a tool, so the audit keeps asking something new.

Photos, people and what stays in the shop

Floor photos can include colleagues, customer paperwork and drawings. Crop or retake shots so that no faces, no names and no customer drawings appear, and tell the team the photos are for the audit. Use a business plan that does not train on your content by default. Keep safety-critical decisions with your own procedures and competent people; an AI-drafted checklist is a prompt for looking, and does not replace a risk assessment.

Questions before you start scoring 5S with AI

Can AI score a 5S audit from photos alone?

It can describe what it sees and compare a photo against your checklist wording, but it misses what a photo hides, such as a blocked fire exit behind a pillar or a gauge past its due date. Use photos as evidence and prompts, and let a supervisor walk the area and give the score.

How often should we audit?

Monthly works well for a small factory, with a quick weekly look at the two weakest areas. More often than weekly tends to become paperwork, and less often than monthly lets standards slip before anyone notices.

Who should score their own area?

Have the area's team lead do a self-score first, then a supervisor from another area score it independently. Differences between the two scores are the useful part, because they show where people read a standard differently.

Is a score of 100% realistic?

Not as a target. Set a pass mark such as 80% per area and a stretch of 90%, and treat a run of perfect scores as a sign the checklist has gone soft. Tighten the wording of the items that nobody ever fails.

Further reads

• AI Quality Control for Small Manufacturers: Where to Start — Where AI helps quality work in a small factory.
• AI Work Instructions From Photos and Short Videos of the Line — Turn photos of the line into standard work instructions.
• Small Factory CMMS With AI: Does the AI Layer Earn Its Cost? — Where maintenance software fits beside a 5S routine.
• 8D Report With AI: Answer a Customer Complaint Step by Step — A full 8D worked example for a customer complaint, with AI drafting D1 to D8 from your evidence, your team choosing the root cause, and every claim traceable.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: the standard five steps of 5S (sort, set in order, shine, standardise, sustain) as used in lean manufacturing; scoring scales are examples to adapt.