One hypothesis-driven improvement per week — state prediction, test, measure, decide
Run one complete PDCA improvement iteration per week on a single operational system: state a hypothesis predicting how a specific change will improve a specific measurable outcome by an estimated amount, implement for one cycle, measure the result, then adopt, adjust, or abandon based on whether the prediction held.
Why This Is a Rule
Deming's Plan-Do-Check-Act cycle is the smallest unit of systematic improvement. Applied weekly to a single system, it compounds: 52 improvement iterations per year, each building on the previous one's learning. But the power of PDCA requires discipline in each step, particularly the hypothesis — the prediction that makes the cycle scientific rather than random.
The hypothesis step — "Changing X will improve metric Y by approximately Z%" — is what separates systematic improvement from tinkering. Without a prediction, you can't tell whether the change worked (because you didn't define what "worked" means), and you can't learn from the result (because you didn't specify what you expected to happen). The prediction makes learning possible regardless of outcome: if the prediction held, the change worked. If not, your model was wrong — which is also learning.
One system per week prevents the common failure of making multiple changes simultaneously, which makes it impossible to attribute improvements to specific changes.
When This Fires
- During weekly operational reviews when choosing what to improve
- When a process feels suboptimal but you're not sure what change would help
- As a standing practice in any continuous improvement program
- Whenever you want to improve a system without relying on intuition alone
Common Failure Mode
Making the change without stating the hypothesis: "Let's try batch-processing emails twice a day instead of continuously." Without a prediction ("This will reduce email time by 30% and increase deep work blocks by 1 hour"), you can't evaluate whether the change worked. You might feel better, but feeling is not measurement.
The Protocol
Weekly: (1) Plan: Choose one operational system. State hypothesis: "Changing [specific thing] will improve [specific metric] by [estimated amount]." (2) Do: Implement the change for one week. (3) Check: Measure the result. Did the metric change as predicted? (4) Act: If prediction held → adopt the change permanently. If partially → adjust and run another cycle. If wrong → abandon and choose a different hypothesis next week. One system, one change, one week, one measurement.