When a team chooses records manually, the check can become biased fast. Someone picks the newest orders, someone picks problem cases, someone picks the easy ones. The review then measures the selector's mood more than the process.
Short answer
- Define the full range before generating the sample.
- Choose the sample size before seeing the results.
- Use no duplicates if the same order should not appear twice.
- Record the date, range, and person who generated the sample.
- Review all selected items with the same checklist.
When this helps
Random sampling is useful when the full dataset is too large to review manually. Examples:
- an online store checks 20 orders from 800;
- a sales team reviews 10 calls;
- accounting checks random documents before closing a period;
- support audits operator tickets;
- a QA manager checks spreadsheet rows;
- a product team tests random export records.
The value is not that randomness is magical. The value is that nobody hand-picks convenient or inconvenient cases.
Prepare the range
The range must be clear and stable. If you use spreadsheet rows, exclude the header row. If you use order numbers, check that the format is consistent. If cancelled orders exist, decide in advance whether they participate.
Example:
Task: order packing quality check
Period: June 1-30
Order range: 15300-15980
Sample size: 25 orders
Duplicates: no
Exclusions: cancelled orders are skippedHow large should the sample be
For a quick weekly check, 10-20 items may be enough. For a more serious operational review, use 30-50. If you need formal statistical confidence, use a proper sampling method. For everyday quality control, a random list is still much better than “let's check whatever we notice first”.
Do not change the sample size after seeing the selected numbers. That turns a fair process back into manual picking with extra steps.
Example result
Sample date: July 5
Range: 15300-15980
Count: 10
Result: 15314, 15388, 15401, 15477, 15520, 15602, 15645, 15719, 15803, 15966
Owner: quality managerYou can paste this into an audit sheet, checklist, or team chat.
Common mistakes
| Manual selection | bias enters the process | use random selection |
|---|---|---|
| No recorded range | result cannot be verified | save period and number range |
| Duplicates allowed | review effort is wasted | use no-duplicate mode |
| Rules change after selection | team trust drops | set rules before generation |
| Different checklist per item | results are not comparable | one checklist for the sample |
What to do after selection
A random list is only the beginning. You still need one review checklist: delivery time, packing errors, customer comments, payment status, document accuracy, response quality, or whatever metric matters. Without criteria, the review becomes “I feel this is bad”, which is not quality control. It is weather with a spreadsheet.
FAQ
Can I use spreadsheet row numbers?
Yes. Enter the first and last row number, exclude the header row, and generate the required count without duplicates.
What if a selected order was cancelled?
Decide before sampling. For example: skip cancelled orders and generate replacement numbers for the same count.
Should the team see the sample result?
Yes, when the review affects performance or training. Transparency reduces arguments.
Is this enough for formal statistics?
Not always. Formal analysis may require a stricter method. For operational quality checks, this is a simple and fair way to choose cases.