
Operational excellence
Did the improvement hold?
Keep the evidence and expected result with the improvement decision. Check the outcome after the change and record what the team will do next.
Most improvement work I've seen, on production lines at Unilever, in launches at BAT and in operations teams since, is organised around the fix. Someone maps the process, someone finds the loss, a countermeasure is agreed, a task is closed. The dashboard turns green for a week.
The team also needs a follow-up date and a way to check whether the improvement lasted.
Five tools, no thread
A typical improvement loop lives in five places. The process map sits in one tool, the data in a spreadsheet, the analysis in a slide deck, the actions in a task tracker, and the result, if anyone checks it, in a dashboard that knows nothing about the rest.
Each tool is fine on its own. The trouble is the gaps between them. The evidence doesn't travel with the finding, the finding doesn't travel with the decision, and by the time someone asks whether the fix worked, nobody can reconstruct why it was chosen.
Write down what you expect
When a KPI moves, record the affected station, loss and supporting observations. Test the proposed cause and write down the result you expect from the fix. At follow-up, compare that prediction with what happened and check other changes that could explain the result.
In the fictional DMAIC filling-line example, the team compares the pilot with readings from the old setting, sampled across filler heads and shifts. After handover, the line lead checks average pack weight every hour. At the follow-up, compare the new readings with the old baseline and check the count of packs below the 495 g internal limit. Keep both results with the decision to lower the setting.
Keep the follow-up with the decision
I’m building Vrolen to keep the observation, evidence, decision and follow-up together. The team compares the follow-up readings with the expected result and decides whether to update the standard or investigate further.
If a proposed change needs testing on a model first, use simulation with recorded assumptions. Set a separate follow-up for the change in use; the model result cannot replace it.