From Clipboard to Chart: A Plant Moves Its Measurements Into a System

A three-line plant, ninety staff, making a moulded consumable product to customer specification. Every hour, on every line, an operator recorded seven readings on a printed sheet and signed it: weight, two dimensions, a temperature, a pressure, a cycle time and a visual check. At the end of a run, the quality technician added a sample inspection. The sheets went into lever-arch files, kept for five years, because customers audit.

The paper did one of its three jobs. It proved to an auditor that checks had been done. It did not catch drift while drift was cheap, and it never once improved the process, because nobody could read three thousand sheets.

Nothing in the shape of it is unusual. Most plants we walk into are running exactly this system.

What the sheets were really doing

Two days on the floor turned a vague sense of "our data is on paper" into a specific list.

• Readings were written after the fact. On a busy shift, three or four hours of hourly checks were filled in at once, from memory and from the last reading. Everybody knew. Nobody said it out loud until we asked in a way that made it safe to answer.
• Out-of-tolerance values were recorded, not acted on. The sheet had a spec range printed at the top. A reading outside it was written down in the same handwriting as everything else, and the supervisor found out at the end of the shift, if at all.
• Nothing could be compared. Shift A against shift B, machine 2 against machine 3, this material lot against the last one: all theoretically possible, all requiring somebody to type up a filing cabinet.
• Scrap lived somewhere else entirely. Reject quantities were counted at the end of the run and written on a different form, so the reading that would have explained the scrap was never sitting next to it.
• Recall queries took two days. A customer asking about a batch from March set off a hunt through files, followed by a nervous wait to see whether the sheet had been signed.

Paper records are excellent at proving you did something and useless at telling you what to do next. A plant that only has paper is compliant and blind at the same time.

What we changed, in order

The temptation with plant data is to start with sensors. We did not, and would not. Instrumenting machines is the second project. The first is the form people already fill in.

Five changes, in this sequence:

1. The form moved to a tablet at the station. Same fields, same order, same language as the paper sheet the operators already knew. Familiarity was worth more than elegance.
2. Tolerance checking happened at capture. Green, amber, red as the number is entered. Amber means trending; red stops the operator and asks two questions: what did you find, and what did you do.
3. Escalation became automatic. A red reading pings the supervisor's tablet. Two in a row pings the production manager. Who gets what was argued about for a week and was worth every minute of the argument.
4. Analytics arrived once there was something to analyse. Run charts per line, first-pass yield, scrap by cause, comparisons by shift, machine and material lot. Nothing exotic, just the questions the quality manager had always wanted answered.
5. Traceability came free. Because every reading already carried a batch, a machine, an operator and a lot number, a recall query became a search box.

What the charts showed in the first month

This is the part nobody promises, because it depends on what the plant has been hiding from itself. Here, within four weeks:

• The night shift ran 3.1% heavier on the flagship product. Consistently, on every line. Giveaway on that scale, across a year of volume, was a material sum in raw material handed to customers for nothing. The cause turned out to be a setting habit, not a machine.
• Machine 2 drifted forty minutes into every run, then stabilised. Visible in ten minutes on a run chart, invisible for years on paper because each sheet showed one shift at a time.
• One supplier's material lots correlated with a rise in scrap. The plant had suspected it. It had never been able to show it, and had therefore never raised it. The conversation with the supplier was short, because it came with a chart.
• The completion rate for hourly checks had been 71%, not the 100% the files implied. Missing hours had simply been filled in later, or not at all.
• One line's sheets had been signed for a machine that was down. Not malice, just a habit of copying the previous hour. It is the single strongest argument for capture at the moment of measurement.

None of those findings required clever analytics. They required the readings to exist in one place, honestly timestamped.

The training, and the fear that comes with it

The first reaction on the floor was not "this is easier". It was "so you will know exactly what I do all day". That reaction is reasonable, and if it is not handled it produces careful, useless data.

What worked:

• The purpose was stated plainly, by the plant manager, before we arrived with tablets. The system exists to catch the machine drifting, not the operator slipping. The commitment made in that meeting was that no one would be disciplined over a reading, and it was kept, including the week it was tested.
• Twenty minutes per operator, at the station, on their own line. Every operator entered a real reading on the real machine before we left them with it.
• The screen was built for gloves and noise. Big number pads, no free text where a choice would do, reason codes as buttons, and a screen readable at arm's length in bad light.
• Paper ran in parallel for three weeks. Duplicated effort, deliberately. It gave everybody a fallback and it gave us a comparison that proved the digital record was more complete than the paper one.
• Supervisors were trained on the charts, not the capture. Their job changed from collecting sheets to responding to amber. That is a genuinely different role and it needed its own session.
• The first month's findings were shared with the floor. Including the night-shift weight finding, framed as a setting to fix rather than a shift to blame. That single meeting did more for adoption than the software did.

Fourteen weeks, in order

What it cost

For scale: a single line with one form is a fraction of this, and is a sensible way to start.

What changed, measured

• Hourly check completion went from 71% to 99%, because a missed check is now visible within the hour instead of at the end of the month.
• First-pass yield improved from 92.4% to 96.1% over two quarters, most of it from catching drift early rather than at end-of-run inspection.
• Scrap value fell by 19%, with the biggest single contributor being the material-lot finding.
• Overfill giveaway on the flagship line dropped by 2.5% of raw material, which on its own covered a meaningful share of the project inside the first year.
• An audit or recall query went from two days to under five minutes, and now produces a document rather than a file box.
• Customer complaints related to dimension variance halved across the following six months.

What we would do differently

We would fix the specification master data before writing a line of the app, not alongside it. Three products carried tolerances on the printed sheet that did not match the customer drawing, and untangling that mid-build cost a fortnight and some credibility.

We would also involve the quality manager in the escalation rules earlier. Alerting is where a system either becomes indispensable or becomes noise, and the person who will live with the alerts has to design them. We got there, but a week later than we should have.

And we would resist even harder on the field count. Eleven fields at launch was right, and it was still a fight. A capture screen that takes ninety seconds instead of twenty is a screen that gets filled in later from memory, which is precisely the habit the whole project exists to end.

Where to start

Pick one line and one form. Digitise it exactly as it is, add tolerance checking at the point of capture, and leave the analytics until you have four weeks of honest data. What comes out of those four weeks will make the business case for everything after it, far more convincingly than a proposal can. The broader argument for this, beyond one plant, is in why the shop floor is not a spreadsheet, and the same discipline shows up in the warehouse that learned to pick.