Sparse function in MDA for nulls
None
We're running into an issue with reports of usage data, where if there is no data for a given customer for a week, that week is just skipped in the report, condensing the X axis and making trend lines inaccurate. In order to remedy this, we're starting to write 0s into the MDA for any data points that are null, which exponentially increases the amount of data we're writing. It would be nice to have a "sparse" function that treats nulls as 0s for reporting purposes.
Sign up
If you ever had a profile with us, there's no need to create another one.
Don't worry if your email address has since changed, or you can't remember your login, just let us know at community@gainsight.com and we'll help you get started from where you left.
Else, please continue with the registration below.
Welcome to the Gainsight Community
Enter your username or e-mail address. We'll send you an e-mail with instructions to reset your password.
Haven't seen a use case for assuming a null would be same as the previous week - always means either no occurrences (if aggregating records - so a zero) or missing data (in which case they don't want to mask it by carrying the previous week's data along).
Here is an example of how two charts that represent two sides of the same coin (risks and opportunities) over time can get out of alignment if nulls are not treated as zero (and given space on the timeline for that zero).
The arrows show where the events from the lower timeline appear on the upper timeline. For the most current 3 weeks the correlation is perfect, but then things get skewed with missing dates.
However, since many of the reports (especially on usage) are often driven from imported data tables, the new rules function to set nulls equal to zero will allow you to manage this on the data source side vs. the reporting side.
When running a report with a date field for the X-axis, the data needs to be properly spaced along the axis. A key benefit of that report is to let the distance between data points indicate elapsed time. But I'm seeing 4 days between dates look exactly like 40 days between dates.
There are times when the underlying data can be filled in so that the chart effectively looks "right" (i.e. put in a 0 value with the right attributes for the days with no value) but that doesn't always work.
I think I remember an earlier conversation about this topic, but I can't find it. Sorry if I'm duplicating -- if anyone knows of the other thread I'll move this there!
The first is how a timeline axis should function. The confusion around this seems to come about because it's an X-axis instead of a Y-axis. If we plot "number of accounts by CSM" we expect the column for 10 to be twice the height of the one for 5 even if no CSM has 6, 7, 8, or 9 accounts. The distance between 5 and 10 is fixed and needs to be shown that way. In the same way, the distance between today and 10 days ago is fixed, even if nothing happened between them.
Once we've agreed that a timeline needs to be continuous there's a second question of how to show trended data (lines not columns) across those periods. But "filling with 0s" is a workaround for the continuity issue, rather than a response to how this data is plotted. If we resolve how data over time is plotted, then the question of whether or not you need to insert a value can be addressed based on the specific business need.
Lastly -- this isn't really a topic that needs much discussion or debate. The most common charting tools like Excel and Tableau have well-established patterns for how to display this data. If we were in alignment with those tools then any resolutions would sit 100% with the individual CS teams.