Analyzing Completed Jobs Without Prior Started Events

Vlad S. · 2026-01-07T18:00:12.364Z

I'm trying to build something like an inverted funnel, and I don't know how to do it: we have two events, one is Started Job and another one is Completed Job. Technically, you shouldn't be able to complete a job without starting one, but it's pretty clear that it's happening in our data. To debug/investigate it would super useful to ask "how many people (and who) have completed jobs without first starting them? what percentage of completed job events don't have a precurson started job event?" that sort of thing, but I don't know how to do it with the current tools

Analyzing Completed Jobs Without Prior Started Events | Mixpanel Community

11 comments

· Sorted by Oldest

Andrew S.
·
·
Could you do a cohort that identified users that had a count greater than one for completion, but equal to zero for started
🤔1
Andrew S.
·
·
In a separate but related note, I have noticed instances in our data where the timing of events does not coincide with the timing of the actions. In other words, B comes before A in Max panel, but in reality the user did A before B.
Vlad S.
·
·
yeah that's a separate problem that I've run into, especially when mixing frontend and backend events, but not at play here
Vlad S.
·
·
yeah, I could find the users who never fire the started event when they should, but I'd also like to find out the instances where it happens occasionally
Vlad S.
·
·
also, started and completed events have unique IDs that correspond to each job, so ideally I would look for completed jobs that never got started (even if other jobs got started during that time period) by using "hold property constant" and setting it to the unique job id
Vlad S.
·
·
easily done going forwards, have no idea how to do it going "backwards"
Vlad S.
·
·
might just need to download all the events and parse them offline
Andrew S.
·
·
How many times does an average user do this event? It would be very annoying, but you could do a cohort where event one equals zero and event two is greater than zero. Then a separate cohort where event one equals one and event two is greater than one. And so on. Then you could look for situations where any of those cohorts are involved.
Vlad S.
·
·
How many times does an average user do this event?
it's a very long tailed distribution... the average is probably once per week, with a LOT of users not doing it ever and several doing it hundred times a day
Vlad S.
·
·
but that's a good idea... it doesn't fully account for the "started job a" -> "completed job b" scenario, but it's close
Andrew S.
·
·
Ugh. If I needed more specifically the “tie in” I'd probably do the export, import to Sql and then do a group by ID to see where ids have 1 event vs 2.
👍1

Andrew S.
·
·
Could you do a cohort that identified users that had a count greater than one for completion, but equal to zero for started
🤔1
Andrew S.
·
·
In a separate but related note, I have noticed instances in our data where the timing of events does not coincide with the timing of the actions. In other words, B comes before A in Max panel, but in reality the user did A before B.
Vlad S.
·
·
yeah that's a separate problem that I've run into, especially when mixing frontend and backend events, but not at play here
Vlad S.
·
·
yeah, I could find the users who never fire the started event when they should, but I'd also like to find out the instances where it happens occasionally
Vlad S.
·
·
also, started and completed events have unique IDs that correspond to each job, so ideally I would look for completed jobs that never got started (even if other jobs got started during that time period) by using "hold property constant" and setting it to the unique job id
Vlad S.
·
·
easily done going forwards, have no idea how to do it going "backwards"
Vlad S.
·
·
might just need to download all the events and parse them offline
Andrew S.
·
·
How many times does an average user do this event? It would be very annoying, but you could do a cohort where event one equals zero and event two is greater than zero. Then a separate cohort where event one equals one and event two is greater than one. And so on. Then you could look for situations where any of those cohorts are involved.
Vlad S.
·
·
How many times does an average user do this event?
it's a very long tailed distribution... the average is probably once per week, with a LOT of users not doing it ever and several doing it hundred times a day
Vlad S.
·
·
but that's a good idea... it doesn't fully account for the "started job a" -> "completed job b" scenario, but it's close
Andrew S.
·
·
Ugh. If I needed more specifically the “tie in” I'd probably do the export, import to Sql and then do a group by ID to see where ids have 1 event vs 2.
👍1