Analytics Modeling Time is Crazy

S1mon

I can imagine a bunch of reasons why it's hard to accurately measure "modeling time" per user or project, but when one user racks up 31 hours of modeling time in a - last I checked - 24 hour day, there's something wrong. This is from a chart which is filtered to a single user.

There's a user activity (~17 hours) which seems somewhat rational, but project activity shouldn't be double or triple counting things just because there are multiple windows open.

Evan_Reese

uh oh. I really want to know exactly how this is being calculated. Having visibility into what the heuristics are for "active time" would help me understand what the analytics can be relied on for, and what they shouldn't be. I don't currently have a good sense for whether I should treat the analytics as if they represent actual work time ±5%, 25% or 50%, which would change how I use them. For example, if we forgot to track time on an hourly project is it accurate enough to ethically bill from, or should be buffer it down to be fully on the up-and-up about it? For project estimation I don't need that level of accuracy, so I could live with ±10-20% and could still usefully compare a new project to a few of similar scope and get a good estimate of effort. If it's 50% off, but usually always in one direction (i.e. always errs high), then I can really only use it to compare past projects to one another, but not to estimate new ones. I know that to make data understandable, it has to be simplified, which means you lose precision. I'm okay with that. I'd just like to better understand how precise I should assume it is.

S1mon

@Evan_Reese
Thankfully I generally don't need to track my hours these days, but having spent half my career in consulting, I can see where correct hours are important for billing. I haven't checked with the engineer in this particular case, but I kinda believe that there were ~17hours of work done that day (the user activity), but the project hours are clearly useless for billing purposes. I don't even know if there were multiple windows open at the same time which could possibly account for how this algorithm works, or if they were merely open tabs. 

Also if I'm staring at an open CAD window, am I working or am I daydreaming about something completely different, or did I go for coffee and the system hasn't timed out yet?

A while back I tried to do my own analytics with my personal Pro account so I could reconstruct hours. I did a lot of ridiculousness with expanding all the document change history, opening the HTML to extract the date/time info which normally only shows on hover, and then performing a bunch of text processing to put it into a GSheet. Then there was one key thing to estimate - how long on average after each micro-version do I count the time as working? I created a simple thing that said if it had been more than X amount of time since the last timestamp, then put in Y time as the estimate. The idea being that I might have been thinking, checking work email or whatever that was billable, but after X amount of time delta, I could assume that on average only Y work was done. I played with X and Y until I got results which made sense to me. I ended up with X=1hr and Y= 50 minutes, but there were very different results depending on these values.

Evan_Reese

We're moving away from hourly for a lot of reasons, thank goodness.

You're super right that choosing the right value for what is considered "inactive time" when no action is taken is really impossible to nail. You can only get close.