2. Data has no value unless analysed
Metering has zero value to business unless collected data is analysed. Drawing graphs and dashboards, aka visualization, does NOT constitute analysis. It’s just another way to show data.
Megabytes save no megawatts.
3. Analysis is a comparison of actual numbers with expectations
Absolute value of metered numbers has no meaning. Results of metering must be compared with expected numbers derived from models or from experience. If I tell you that plane landed at 8.45, is it good or not? – Depends on plane schedule.
No model is perfect – ignore small fluctuations, but …
4. Watch for sustained deviation from expectation
To see if small variation sustains – use CUSUM to notice them: CUSUM stands for cumulative sum of differences between actual and expected value. Due to its ‘cumulative’ nature, CUSUM averages out random fluctuations, but accumulates sustained ones. Small deviations do not matter, unless they are sustained over time. It’s like small steering wheel adjustments do not matter, until for every two left adjustment driver makes only one right.
Remember that any statistical model is an approximation, it calculates mean value, which is subject to built-in uncertainty. (Rare client understands this, so keep it for internal quality control 🙂 )
5. Check conclusions based on math against common sense.
Math is a tricky beast: it does not care about commons sense or input errors or typos. Garbage in – garbage out.
Watch for common sense before finalizing conclusions.