Maintenance Request Form: Triage Factory Requests by Urgency

Give every factory maintenance request the same form (machine, what is wrong, is anyone at risk, is production stopped, photo), score it against four written urgency levels, let any safety tick override the score straight to level one, and use an AI assistant to tag, route and flag mismatches. A maintenance lead reviews the queue weekly.

This differs from the repair forms tenants fill in for a landlord. A factory request competes for a small team's time against machines that earn the day's money, and some of those requests describe a hazard. So the form is built around two questions a tenant form never asks: is production stopped, and could anyone be hurt. Everything else, including the AI, serves those two.

Follow me on Instagram@sagnikteaches

What goes wrong with a notebook and a shout across the floor

In many small plants the "system" is a whiteboard, a clipboard and whoever the operator can find. It works while the maintenance team is two people who know every machine. It stops working the day a third shift starts, or one fitter is off, because three things go missing: the exact machine and fault, the urgency as the operator sees it, and a record that anyone can search afterwards.

Connect on LinkedInSagnik Bhattacharya

A form fixes the first two, and a sheet or system that keeps every submission fixes the third. The AI part is smaller than it sounds: it reads the free-text description, suggests tags and an urgency, and drafts a short note to the fitter. It does not decide what is safe.

Subscribe on YouTube@sagnikteaches

The fields on the form, and why each earns its place

Keep the form short enough to complete at a machine with gloves in a pocket. Phone access matters, so use a form tool that works in a mobile browser (Microsoft Forms and Google Forms both do; Google Forms can send answers to a linked Google Sheet, and Microsoft Forms can open results in an Excel workbook that keeps picking up new responses) and put a QR code on the machine itself that opens the form with the machine pre-filled where the tool allows it.

Two fields deserve a note. The "is anyone at risk" question must be the first yes/no on the form after the machine, because it is the question that matters most and people answer the top of a form more carefully. And the guard and lockout question is there because operators sometimes will not call a missing guard a "risk", but will agree it is a guard.

Four urgency levels defined by what happens if the job waits

Define each level by consequence, and print the definitions beside the urgency field. People pick a level by reading it, so the wording does the triage.

The targets are illustrative. Set yours with the people who have to meet them, and write them down once. A target nobody can meet teaches the floor to ignore the levels.

The safety-critical override, written as a rule rather than a feeling

A rule needs to be mechanical, so that a tired supervisor at 3 a.m. applies it the same way as the lead on a good day. A workable version:

1. If "anyone at risk" is Yes, the request is level 1, whatever the other answers say.
2. If the guard, interlock, e-stop or lockout question is Yes, the request is level 1.
3. If the free text contains any word on the escalation list, the request is raised to level 1 pending a human check. The list starts with: guard, interlock, e-stop, emergency stop, lockout, trapped, sparks, smoke, fire, gas, burn, electric shock, exposed wire, leak (on pressurised, hot or chemical systems), collapse, fell, injury, cut.
4. The system may raise a level on its own. Only a named person may lower a level 1, and the reason is recorded.

Rule 4 matters more than it looks. An AI step that can lower urgency will occasionally lower the wrong one, and the requests it lowers are the ones nobody looks at again. Make the direction one-way for anything safety-related. Also tell staff that level 1 is "make safe first": isolate, guard, cordon, then repair. A level 1 that sends a fitter straight to repair a live hazard is not the aim.

The form is not a substitute for your site's incident or near-miss process, and a serious hazard goes by phone or radio first. The form records it afterwards. For near misses that should feed the same review, see a near miss report template with AI logging and trends. For anything touching legal safety duties, your safety adviser sets the rules; this tutorial only builds the queue.

What the AI step does with each submission

Once a submission lands in the sheet, an assistant processes it. It has four jobs, in this order:

1. Tag the fault type from a fixed list (electrical, hydraulic, pneumatic, mechanical wear, controls and software, guarding and safety, utilities, building, cleanliness, other).
2. Propose an urgency level from the definitions, and say why in one sentence.
3. Raise any safety trigger it sees in the text, even where the requester ticked No.
4. Draft a two-line note for the fitter: machine, symptom, what to bring.

You can run this from a spreadsheet add-in, an automation tool, or by pasting a batch into an assistant at the start of each shift. Which route you use matters less than the prompt:

You triage maintenance requests for a small factory. For each request
below, return one line in this format:
ID | fault type | proposed level (1-4) | safety flag (YES/NO) | reason | fitter note

Fault types (use exactly one): electrical, hydraulic, pneumatic,
mechanical wear, controls and software, guarding and safety, utilities,
building, cleanliness, other.

Levels:
1 = anyone could be hurt, a safety device is missing or defeated, fire,
    smoke or uncontrolled leak.
2 = production stopped, or bottleneck machine below half speed.
3 = runs but slipping; no risk.
4 = comfort, appearance, improvement.

Rules: If the text or the form mentions any guard, interlock, e-stop,
lockout, sparks, smoke, burn, shock, exposed wire or injury, set
safety flag YES and level 1. You may raise a level but never lower the
requester's suggested level if the safety flag is YES. If you cannot tell,
say UNSURE and level 2. Do not invent causes. Fitter notes must be under
25 words.

REQUESTS:
[paste rows: ID, machine, description, at risk?, production status, requester level]

A worked day: twelve requests in a forty-person plant

Take an illustrative plant that makes sheet-metal enclosures, running two shifts with a three-person maintenance team. Here are twelve requests from one day, the way operators wrote them, and what the triage returned. Names of machines are placeholders.

Seven of the twelve moved, and all seven moved up, which is the pattern you want from a safety-weighted rule. R1 is the instructive one: the operator tagged it level 3 because the machine was running, and the tape on the sensor was the sign that told the whole story. R10 shows a middle path, an urgent interim action (contain the oil) separate from the repair. Put "make safe now" and "repair by" as separate fields in your sheet when you see this often.

Where the AI gets triage wrong and how the sheet shows it

Three patterns will turn up in the first month, so watch for them:

• Over-trust of operator wording. "Just a small leak" becomes level 4 because the text sounds minor. The fix is a rule keyed on the system, not the wording: any leak on a pressurised, hot or chemical system is at least level 2.
• Missing machine context. The assistant does not know that press 2 is the only machine that makes your best-selling product. Add a "criticality" column to the asset list (A for bottleneck, B for normal, C for spare capacity) and include it in the prompt. A level 3 request on an A machine may deserve level 2.
• Duplicates. Three operators report the same faulty door on one shift. Ask the assistant to group requests for the same machine within 24 hours and show the count. Several reports of one fault are themselves a signal.

The sheet should carry both the requester's level and the final one, plus a column recording who changed it. After a month, sort by the gap. A cluster of level-4 submissions that ended at level 2 tells you where operators underrate risk, and the answer is a short word at the next team meeting, not a rule change.

Routing: who gets which request

Routing by fault type is usually enough: electrical goes to the electrician, hydraulics and pneumatics to the fitter with that skill, building to facilities. Write the routing as a small table the assistant reads, so a change of staff means changing a row.

Level 1 and level 2 requests should alert a person rather than wait in a sheet. A message to a shared team chat channel or a text to the on-shift fitter is enough. Do not let the AI send these on its own without a human on the receiving end; the alert goes to a person, who decides.

The weekly review that keeps the form honest

Thirty minutes every Monday, with the maintenance lead and one production supervisor. The agenda is three lists the sheet can produce, and an assistant can summarise the week into them:

1. Safety flags. Every level 1, what was found, who made it safe and when it was closed. Nothing stays open here.
2. Repeat machines. Machines with three or more requests in the week or five in the month. These are candidates for a planned repair or a replacement part rather than another quick fix.
3. Level gaps. Requests where the requester's level and the final level differed by two or more, with the reason.

You can ask an assistant to draft the summary from the sheet, but read it against the sheet before the meeting, because a summary that smooths over a closed-late safety flag is worse than none. If you want to take the repeat-machine list further, link it to downtime data with reading downtime logs to find repeat stops, and to spare parts with AI-set minimums for maintenance spares.

The asset list behind the dropdown: a short filled-in extract

The dropdown is only as good as the list behind it, and the criticality column is what lets the assistant weigh a level 3 request on a bottleneck machine properly. An illustrative extract for the sheet-metal plant above:

The lockout column is a small extra with a large effect: when a level 1 request comes in, the fitter's note can say where to isolate before anything else, which is the "make safe first" step in practice. Build the list once, from the walk-round a maintenance lead already does in their head, and keep it to one row per asset a person would name out loud.

Handing night-shift requests to the morning team

The second place requests go missing is shift change. A request filed at 2 a.m. on a quiet shift may be seen by nobody until the day team has already started on its own list. Add a handover filter to the sheet: everything still open from the previous shift, sorted by final level, with any level 1 or 2 shown first. Have the assistant produce a six-line handover note at the start of each day:

From the open requests below, write a handover note for the day-shift
maintenance team. Order: any open level 1, then level 2, then level 3
older than five days. One line each: ID, machine, what is wrong, status,
who has it. End with the count of level 4 requests open. Do not
summarise or merge different machines. If a level 1 is still open, say so
in the first line.

The rule "say so in the first line" is deliberate. A handover that buries an open safety item in the middle of a tidy list is how it stays open for another day.

Measures that show the system is working

Pick three and track them from the first week. Avoid vanity numbers such as total requests, which rise when the form is easy to use.

• Time to acknowledge for levels 1 and 2. Average and worst case. If the worst case is hours, the alert route is broken, not the triage.
• Share of requests with a machine, a fault type and a final level. Aim for all of them. Gaps mean someone is working around the form.
• Repeat requests on the same machine within 14 days. A fall means fixes are holding; a rise means quick patches.

A quick sum shows why the effort pays. Suppose a plant logs 60 requests a week and the old method needed the lead fitter ten minutes each to interpret, call back and prioritise: ten hours a week. With a form and a tagged queue the interpretation drops to about three minutes a request, so three hours. Seven hours back is nearly a working day of maintenance time. The numbers are illustrative; time your own for a week before you change anything, so you have a real before to compare with.

When the sheet is not enough any more

A form and a sheet carry a small plant a long way. The signs that you have outgrown them are specific: you want planned maintenance tasks to generate themselves on a schedule, you need cost and parts used per machine, you want an audit trail for a customer or certifying body, or two sites share one team. At that point a maintenance management system (CMMS) is the right tool, and the request form simply becomes its front door. Whether the AI layer in a small-factory CMMS earns its cost covers that decision. Your written levels, safety rule and routing table transfer straight across, which is the main reason to write them first.

Finally, tell the floor what changed and why. A short briefing at a shift start, with the four levels on a card by the machine and the QR code, does more than a policy email. Make one promise and keep it: a request marked level 1 or 2 gets a person's response within the target, and the operator hears back.

Questions from maintenance leads setting this up

How many urgency levels should a factory use?

Four is enough for most small plants: stop-the-line or safety, this shift, this week, and planned. More levels invite arguments about which box to tick. Define each level by what happens if the job waits, not by how annoyed the requester is.

Can AI decide that a request is safety-critical?

It can flag likely ones, but the rule should be written so a keyword match or a requester's tick always escalates, and a human confirms. Treat the AI as a second pair of eyes that can only raise urgency, never lower it, on safety wording.

Do we need a CMMS to do this?

No. A form, a shared sheet and an assistant that tags rows will work for a plant logging a few dozen requests a week. A CMMS (maintenance management system) becomes worth it when you need asset history, scheduled tasks and parts tracking in one place.

Should operators be able to set the urgency themselves?

Let them suggest it, and let the rules or the maintenance lead confirm it. Operators know when a machine is dangerous or stopped; they are less able to judge a job's effect on the rest of the plant. Keep both fields so you can see where they differ.

How do we stop people filing everything as urgent?

Show the definitions on the form, record both the requested and the final level, and review the gap weekly. When people see that a wrongly urgent request is moved down with a reason, the habit fades without a lecture.

Further reads

• Factory Training Videos With AI: Safety Clips in Several Languages — Teach the four levels to a multilingual floor.
• AI Quality Control for Small Manufacturers: Where to Start — Link slipping quality to the machines behind it.
• 5S Audit Checklist: Build One for Your Shop Floor With AI — Tidier floors cut the small requests that clog the queue.
• Daily Production Report: Summarise Counts, Scrap and Stops With AI — Pair breakdown requests with the day's stops and scrap.
• Factory Energy Audit With AI: Find Peak Loads and Idle Waste — Spot failing equipment through its energy use.
• AI CCTV Cameras for a Factory: Detection, Cost and Staff Consent — What camera analytics adds, costs and requires from staff.
• Hotel Maintenance Software or AI? Log, Prioritise and Chase Requests — Decide between maintenance software and AI with a two-week test: one request form, urgency rules, a weekly chase list and a room-out-of-order tracker.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: the form, levels and figures are illustrative designs for a small plant, not vendor data; spreadsheet and form tools were described from their own help pages.