Transactions Needed for Accurate Results

The answer to this question depends on what accuracy is acceptable to you for the data items that you capture. The common accuracy measures are 90%, 95% and 99% accuracy with 95% being the most widely used in business situations.

Click on appropriate link below to view a detailed discussion with examples.

Activity Sampling Accuracy            Activity Logging Accuracy

Activity Sampling

In most close elections 50% of the vote would be usually be a typical polling value. If the polling company (and the readers and viewers!) are willing to accept a plus or minus 4% error then 600 observations are required – hence the reason you usually see them using between 600 and 800 results in their polls. If pollsters had to be 50% plus or minus 1% confident of their numbers they would need to survey 9603 people to issue their results!

In real world business analysis using the common 95% confidence some examples are:

1. If you find one activity occurs 5% of the time and you want to be 95% certain that its true value was between 5% plus or minus 1% i.e. its true value is between 4% and 6% then you need 1824 observations.

2. If you find one activity occurs 2% of the time and you want to be 95% certain that its true value was between 2% plus or minus ½ % i.e. its true value is between 1.5% and 2.5% then you need 3011 observations.

3. If you find one activity occurs 2% of the time and you want to be 95% certain that its true value was between 2% plus or minus 1% i.e. its true value is between 1% and 3% then you only need 724 observations.

In general if you plan on between 1500 and 4000 transactions you will cover the vast majority of outcomes that would be of interest to you. Additional transactions will simply increase your confidence in the results.

The 95% confidence values are used in Profit Profiler reports as shown below:

Note: Our examples here assume 95% confidence limits and plus or minus the displayed percentage as an absolute value. Other typical confidence limits would be 90% (which requires less transactions to achieve) and 99% which requires more observations to achieve. The plus or minus values could also be expressed as a relative value which is rarely used.

.

Activity Logging

In Activity Logging, your users will be recording actual elapsed time to perform the various activities. The variance of the times for a particular activity is measured by what is known as the statistical standard deviation. Once again there are formula that express the various accuracy levels. 

Profit Profiler uses 95% accuracy so the values reported will show average value plus or minus the confidence limit.

Example - 4.15 Minutes +/- 0.37 

If however the combined average labor and overhead rate for the department in the study was $40/hour, you would be 95% certain that the cost for the transactions was between $2.52 and $3.01 - the average value would be $2.77. (This information would be particularly useful if you were currently pricing that transaction at a dollar or maybe ten dollars - as has happened with more than one client!)

In general the more observations you make the tighter the 95% confidence range will become. Also the greater the variability in the task times the larger the range will be. However if you can gather between 50 and 100 typical observations per task you can expect acceptable accuracy.

Remember, however, that it may be the non-typical occurrences of a task i.e. those flagged as “Other” by the users, that are causing problems in your process and inflating your costs. A high incidence of such transaction should be investigated.