In 2019 I walked into a shop in Bekasi that had a beautiful five-station line and a problem nobody had named. One man, Agus, could set the TDF former. When he took two weeks' leave for his brother's wedding, output dropped 31 percent and the rejection rate on flange height doubled. The line was fine. The knowledge was in one head.
That shop now runs a skill matrix on the wall of the tool crib, a laminated A3 with names down the side and stations across the top. It looks like a school chart. It is the most useful management document they own, and it took about nine hours to build.
A skill matrix is a grid: stations on one axis, people on the other, and a competence level in every cell. The value is not the grid. It is the argument you have to have while filling it in, because that argument forces you to define what "can run the machine" means in observable terms instead of by reputation.
I use five levels, and each one has a test attached. If there is no test, the level is a guess.
Level 0, shadow. Has watched the station. Cannot touch it without supervision. No test.
Level 1, assisted. Can load material and unload parts, can start and stop safely, can call for help before damage. Test: run 20 parts under supervision with zero safety interventions and no machine alarms caused by the operator.
Level 2, supervised setup. Can perform a changeover with someone checking each step. Test: complete three changeovers on different sizes with the supervisor signing each step, no scrap attributable to setup.
Level 3, independent. Runs the station alone across the full size range, diagnoses the five common faults, and holds tolerance. Test: 20 consecutive parts within tolerance, one full production shift with no help, and correct diagnosis of three faults injected deliberately by the supervisor.
Level 4, trainer. Can teach the station and write the changeover card. Test: has taken one other person from level 0 to level 2, and their card is the one pinned on the machine.
Level 4 is where most shops stop investing and where most of the value sits. A level 3 operator who cannot explain what he does is still a single point of failure, just a quieter one.
Below is the grid I build for a shop running a five-station auto line with TDF flanging and a separate lock former. Levels are what each role should hold, not what they currently hold. The gap between the two columns is your training plan.
| Station | Line operator | Setter | Shift lead | Hours to level 3 | Cost of one hour of downtime |
|---|---|---|---|---|---|
| Decoiler and feed | 3 | 3 | 3 | 40 | USD 55 |
| Plasma nesting table | 2 | 3 | 3 | 120 | USD 70 |
| Shear and notcher | 3 | 3 | 3 | 60 | USD 45 |
| TDF flange former | 2 | 3 | 3 | 90 | USD 90 |
| Pittsburgh lock former | 2 | 3 | 3 | 50 | USD 45 |
| Beader and rollformer | 2 | 3 | 3 | 45 | USD 40 |
| Gorelocker and elbow assembly | 1 | 3 | 3 | 90 | USD 50 |
| Folding and seaming bench | 2 | 2 | 3 | 30 | USD 25 |
| Leak test rig | 1 | 1 | 3 | 16 | USD 30 |
Two things in that table are worth staring at. The plasma table takes 120 hours to independence, because nesting and cut quality are a genuinely different skill from folding metal. And the leak test rig needs only 16 hours, which is why it is the first thing I cross-train when a shop has a quality problem, since the person who tests has to understand what a failure looks like.
Every station needs at least two people at level 3. That is the whole rule. Not one plus a trainee, not one plus the shift lead who is busy. Two independent operators, per station, per shift.
On a two-shift operation with nine stations, that means at least 18 independent operator-slot combinations. A nine-person shop cannot meet that with specialists, so cross-training is not a nice-to-have. It is arithmetic. I usually find a shop of nine can reach full coverage with each person holding level 3 on three stations and level 2 on two more.
The first pass does not need to be perfect. It needs to be honest and finished, because a half-built matrix becomes an argument about the grid rather than the work.
Job titles lie. "Machine operator" in one shop means a man who changes coils and dies; in another it means a man who presses the green button. Write down what physically exists along the line: decoiler, feed, plasma, shear, notcher, beader, TDF rolls, lock rolls, gorelocker, bench, test rig. If the same man runs two of them as one job, that is two stations, not one.
Stand at each station for a full shift and write down every decision the operator makes. On the TDF former, that includes: reading the flange height on the drawing, setting roll gap, checking the first part, judging whether the flange is cracking at the corner, deciding when to change a die insert. That list becomes your assessment criteria. If the list is longer than fifteen items, the station is two stations.
Do not ask people what they can do. Watch them. Give each operator three tasks and score them against the criteria from day two. A realistic first matrix in a shop that has never done this usually looks like: two people at level 3 across the whole line, four at level 2 on their own station only, three at level 1, and nobody at level 4. That is a business running on two heads.
Put it on the wall in the tool crib, with real names. Some managers resist naming people publicly. In my experience the operators themselves want it up, because it is the first time anyone has written down that they can do the difficult thing.
Random cross-training fails. The sequence matters more than the hours.
Nearest neighbour first. Train the shear operator on the notcher, not on the plasma table. The material handling is the same, the tooling language is the same, and the confidence transfers. Each step should share at least half its physical routine with the station the person already knows.
Setup after production, never before. Get the person to level 3 on running the station, then teach changeover. Teaching setup first produces an operator who can change a die but cannot hear when the machine is running badly.
Faults are a separate lesson. Put three deliberate faults into the machine during a quiet shift: a wrong roll gap, a dull insert, a mis-set backstop. The operator has to diagnose each without help. Passing this is what separates level 2 from level 3, and it takes about four hours per station.
Write the card together. The changeover card should be written by the person who does the changeover, with the trainer editing. A card written by an engineer and handed down gets ignored within a month. One that the operator wrote gets followed and gets improved.
The Bekasi shop took eleven months to go from two level-3 people to seven. Three numbers moved.
Changeover time on the TDF former fell from 38 minutes to 21, not because anyone got faster but because two people could now do the setup in parallel and the second one caught errors before the first part was run.
Rework on flange height dropped from 4.1 percent to 1.3 percent. The mechanism was unexpected: with more people able to run the station, the best operator was freed to spend his time on the first-off check across the whole line, and first-off checks catch the systematic errors that produce whole batches of scrap.
Overtime dropped 18 percent, mostly because absence stopped being a production emergency.
What did not change was throughput per hour on a fully staffed day. Cross-training is not a speed programme. It is an insurance programme that pays out in the weeks when someone is sick.
Three failure modes, all predictable.
Level 2 is comfortable. The person can do the job with someone nearby, so the supervisor stops pushing, and the matrix sits at level 2 for two years. Level 2 does not cover a shift. If a station's best operator is ill and the backup is level 2, you have a station that runs at 60 percent speed with a supervisor standing next to it. Set a date for each level 3 and put it on the matrix.
If a man who can run four stations earns the same as a man who can run one, the matrix becomes a poster. It does not need to be a large differential. In most of the shops I work with, level 3 on a third station is worth 4 to 7 percent, and level 4 trainer is worth another 5 percent plus a training allowance per person certified. The cost is small against the downtime it prevents, and the message is not subtle: skill is paid for here.
Day shift gets the training because the trainer is there in the day. Night shift inherits level 2 operators and no teacher. The fix is blunt: run the night shift's training in the first two hours of the night shift, with the trainer staying late once a week. It is inconvenient for one person and it is the difference between a two-shift plant that can absorb a resignation and one that cannot.
The matrix has a half-life of about six months. People leave, people forget, new machines arrive. I review it quarterly in a fifteen-minute meeting, and I add one thing every quarter: a new station, or a new fault to the diagnosis test, or a person moved up a level. A matrix that has not changed in a year is decoration.
One more habit worth building. When someone reaches level 3, the person who trained them signs the cell. It costs nothing and it makes the trainer's role visible, which is the only reward most of them get.
Three at level 3 and two at level 2 is a good target for a shop of nine to twelve people. Beyond four at level 3, you start to lose depth, because nobody runs a station often enough to stay sharp.
No, they do different jobs. The matrix tells you who is allowed to run a station. The instruction tells them how. A shop with a matrix and no instructions is one resignation away from reinventing its own process.
About nine working hours to produce the first honest version, spread over a week. Getting a shop of nine to full coverage takes nine to fourteen months, mostly because of the 90 to 120 hours each person needs on the harder stations.
Include anything that stops the line when the person is absent. In most shops that means the person who books coil, the person who raises the delivery notes, and the one who can operate the forklift. Those three cause more unplanned stoppages than a dull die ever does.
One last note for buyers. When you are specifying a line, ask what the changeover procedure looks like on paper, then ask who wrote it. If the answer is the engineering department and the operators have never seen it, you are buying a machine that will run on one man's memory. The machine pages here describe the operator-facing controls station by station, and the guides section has the throughput comparisons for shops still choosing a size.