Mixpanel and Intercom told different versions of product usage.
I spent the better part of a day going through both. The differences were not random, and I understood most of the reasons for them, but the two systems were not identical enough to treat their numbers as interchangeable.
I had more confidence in Mixpanel. The problem was getting the useful data back out.
The code I wrote could dump JSON, but the result still required a lot of work before anybody could analyze it. Mixpanel was good for checking an account directly and confirming when a user had last been seen. It was not the easy export the client-success team needed for a larger scoring spreadsheet.
Intercom was easier to export to CSV. Its usage data looked directionally useful from the point when its integration began, so it could feed the more complicated scoring work. It just needed a sanity check against the source I trusted more.
The short-term plan used both tools. Run the reliable report in Mixpanel and inspect the returned accounts there. Run a similar report in Intercom, export the CSV, and use that in the scoring sheet. I would help train the team and stay directly involved until the numbers were solid enough for the immediate work.
We had handled product reports the same way a few weeks earlier when several urgent issues needed to be separated and validated. A report arriving with confidence did not make it correct.
The team started updating the information manually while we investigated better export options. It was not elegant. It was a workable answer for the next report.
Mixpanel for confidence. Intercom for the export. Human comparison in the middle until the two stories made sense.