Why data cleanliness decides trust in the report
A board trusts a report until the first figure it does not understand. If the logins in the report include technical accounts, and the number of alarms includes tests, someone will ask about the source and the whole document loses credibility. Data cleanliness is therefore a condition, not an extra.
In practice, plants generate data that does not describe the work of people: accounts used for integrations, virtual addresses that serve a kiosk, alarm templates triggered for tests. The report should skip them, and the recipient should know that it does.
Practice 1: exclude technical data at the source
Report data is cleaned: service accounts, virtual kiosk card addresses and technical alarm templates are excluded from queries. Excluding at the query stage is better than subtracting by hand afterwards, because it does not depend on the memory of the person preparing the report and works the same way in every period.
When the plant adds its own technical accounts, for example for integration with another system, it is worth agreeing with the administrator whether they will be excluded. Without such an agreement, system logins end up in the user activity section.
Average and median closing time of permits in the example
Practice 2: read times as a median or a percentile
Times in reports are presented as a median or a percentile, which are robust to extreme values. The average tells a different story: one permit left open for three weeks because someone forgot to close it raises the average time by a value that does not reflect the typical case.
The permit report shows the median time from opening to closing, and the report on reports shows the median reaction and resolution time. The median answers the question of how long a matter typically takes. Individual long cases are worth looking at separately, in the details of the report, rather than through distorted averages.
Numerical example
Take five permits closed after 2, 3, 3, 4 and 60 hours. The average is 14.4 hours and the median is 3 hours. The first figure suggests a problem with the whole process, while the second shows that a typical permit closes quickly and the problem is one forgotten permit. This is a sample calculation, not plant data.
Practice 3: do not interpret profiles from short periods
The daily delivery profile is calculated from periods of at least 21 days. This protects against conclusions drawn from too small a sample: the distribution of delivery hours over three days may be a coincidence. For shorter periods, the profile is not shown as a reliable pattern.
Treat the daily report in the same way. It shows the previous day, so it serves reaction, not trend assessment. Draw conclusions about trends from the weekly or monthly report.
Practice 4: count people, not permits, where people are concerned
The training report contains a section of people on site without a valid training, counted per person, not per permit. One person with several permits is counted once. This way the number matches the actual number of people who need training, not the number of documents.
Apply the same thinking when interpreting: ask what exactly the unit of counting is before you compare two indicators.
Practice 5: agree a shared dictionary of indicators
Before a report reaches management, agree with the safety department how indicators are named and what they mean. A common dictionary prevents the director and the inspector from reading the same figure differently.
Record in one document which data is excluded, how the median is calculated and what the cadence is. Such a document shortens audit conversations and makes it easier to onboard new report recipients.
Checklist
Video
Frequently asked questions
Why does the report show the median rather than the average?
The median is robust to extreme values. One forgotten permit does not inflate the typical time the way an average would.
Do service accounts appear in the activity report?
No. Service accounts, virtual kiosk card addresses and technical alarm templates are excluded from queries.
From what period length is the daily delivery profile calculated?
From periods of at least 21 days. Shorter periods do not give a reliable pattern.
Let us check whether your reports count what they should
We will show how data exclusions and the median work in reports, and go through the indicator dictionary for management. Book a short demo.
Let's talk