Difference in throuput times of the Throughput Time Search Component and the OLAP-chart


#1

I am trying to reconstruct the throuput times between two activities in an OLAP table (ROUND_MONTH(“CEL_LBU_ACTIVITIES”.“EVENTTIME”) as Dimension) as follows
AVG(CALC_THROUGHPUT(FIRST_OCCURRENCE[‘Activity1’] TO LAST_OCCURRENCE[‘Activity2’], REMAP_TIMESTAMPS(“CEL_LBU_ACTIVITIES”.“EVENTTIME”, HOURS))) .

To check the results I used the Throuput Time Search Component months by month (and FIRST_OCCURRENCE[‘Activity1’] and LAST_OCCURRENCE[‘Activity2’]) .

However, I get different results and don’t really understand, which one of them ist correct. What I want in the end is to get the average thoughput time between Activity1 and Activity2 for every respective month.

I would be really happy if someone could help me.

Thank you!


#3

Hello Maria,

do the results seriously differ or could this be a result of differing rounding mechanisms? In essence both solutions should be correct.

Best,
Henry


#4

Yes, they actually differ a lot…for example 143 vs 161 hours for on of the month. I dont think it is due to the rounding.

Best regards,
Maria


#5

Here is the pic with the codes…I do think it is due to the “round months” as Dimension for the OLAP-table. Is there any other Option then to Display the months on the x-axis and what Kind of date Format the Throughput Time Search Component and the Single KPI uses?


#6

Hello Maria,

apparently this occures indeed because of the dimension. Because you use the Eventtime from the activity table, cases which flow through certain activities several times are weighted heavier than the cases where the activities occur only once. The solution would be to use the PU_FIRST function e.g.

ROUND_MONTH(PU_FIRST(CASE_TABLE,EVENTTIME))

In the dimension to ensure that only one eventtime is used to calculate the throughput time.

Best,
Henry


#7

Hello,
unfortunately, it doesnt work:

Best,
Maria


#8

Well, you have big variations in throughput time month by month, so total average can be easily 4 days. Also you can have different number of cases per month. Just do а quick check:

  1. Remove filter from the top
  2. Click on one column on your chart to select only cases which started on specific month. Values on all three widgets should match
  3. Remove filters again and select two continious months. I.e. 2016-04 and 2016-05. Two other widgets should show 17.5

If tests above works your formula is correct (and I believe it is correct).


#9

thank you, u r right!


#10

I have one more question: the analysis works fine (all three components coincide) if i pick the month from the Chart (like 01.2017) as @nicks.si suggested.

But if I use date-picker (for example, from 01-01-2017 to 31-01-2017) then the values of the Throughput Time Component differ from the Values in the Chart. Theoretically, it should be the same month and the same observations.

Could you explain, what is calculated by each of the components and where the difference comes from? Is it possible to enable the chart to show the values as in the Throughput Time Search component? It would be otherwise confusing for the users.

Thank you!


#11

Well, unfortunately date range filter name is a bit misleading here. There are two very different filters in your analysis even is they are called simulary:

  1. Date range filter from the top row will select all cases which has AT LEAST ONE ACTIVITY in January 2017
  2. Clicking on your chart you will select cases which STARTS in January 2017

If you have cases longer then one month it will result in different data subsets and as result different average values.


#12

Thank you!
Does it mean that the Throughput Time Search Component may count the cases ‘double’ (first time say in January 2017 when the activity1 occured and the second time lets say in February 2017 when the activity 2 happened)? Is it also the case that with the events lasting longer than one months, one observes a falling trend closer to the end of the observation period?

Thank you


#13

Nope, it ill be counted as one case.

You are right. This is kind of standard error people make when building some charts with trends. Positive trend can be not as positive as you expect :grinning:


#14

Thank you very much. Then I have some other problem. My current code gathers all the events with activity1 starting in some month K and looks whether those had activity 2 afterwards…if yes, it takes them into account for the average throughput time (the IS situation in my awesome drawing :rofl: ) . I am, however, rather interested in the situation, where the code looks for activity2 in some months K and calculates the throughput time from activity1 that happened some months/days before. The advantage of the second approach is that the numbers for the same months do not change over time (compared to the IS case, where new and new cases appear for the previous months as activity2 is executed). Is it possible to implement?
image

Best regards,
Maria


#15

Sure you can do it :grinning: If you want it on chart just use ROUND_MONTH(PU_LAST(“your cases table”, “your events table”.“your event time field”, “your events table”.“your activity name field” = ‘activity 2’)) to get month of activit2 as dimension and use PROCESS EQUALS “activity 1” TO ANY TO “activity 2” as component filter or as CASE WHEN condition to calculate your KPIs.
You can even check for multiple activity names using “your events table”.“your activity name field” IN (‘activity2’, ‘activity2a’, …)


#16

Awesome, thank you!!! Can you explain how the Filter with “TO ANY TO” would work for multiple Activities?