Thread regarding Anthem Inc. layoffs

AI Tracking at Anthem Inc

Does anyone know about the AI algorithms used for tracking our buy-in to the new processes?

by
| 1126 views | |
Post ID: @Z2I37J2

5 replies (most recent on top)

So if we haven’t been notified of our score by now, does that mean we are okay?

by
|
Post ID: @Z2I37J2-Ltvz

Wonder why it's cutting off.

The formula gives a very straightforward way of putting everyone in list order in an objective way. It will minimize lawsuits which cost the company over $30M annually. You may have noticed new software pushed to your laptops/desktops/HVDs which tracks and graphs activity to the keystroke.

The legend is beneath the formula, which is current as of April 3. Here is the formula:

(75A + 15B)/60 + (185C)/(D) + 1.375E + 2F + G/180 + 20H + 45*I + J

where:

A = number of telecommute days as determined by VPN login (one per calendar day max). If the employee badges in to the work location that day at least twice and at least four hours apart (e.g. when entering in the morning and when coming back from lunch), a subsequent VPN login is assumed to be working from home after hours and will NOT count against the employee for that day.

B = number of "mostly idle" hours per day at desktop; this is calculated by the new software (check your system tray on Windows or Mac) by making one random sample per minute on average; more than 30 random samples in one hour that determine an "idle" (non moving/non typing) state per minute flags that hour as "mostly idle". This prevents employees from using a "drinky bird" device to hit keys, which generally do so at a much more regular interval; random sampling defeats that. (half kidding about that; just pointing out that any sort of attempt to circumvent it will likely not work)

by
|
Post ID: @Z2I37J2-1isw

This is the first part of the AT this is calculated by the new software (check your system tray on Windows or Mac) by making one random sample per minute on average; more than 30 random samples in one hour that determine an "idle" (non moving/non typing) state per minute flags that hour as "mostly idle". This prevents employees from using a "drinky bird" device to hit keys, which generally do so at a much more regular interval; random sampling defeats that. (half kidding about that; just pointing out that any sort of attempt to circumvent it will likely not work)

by
|
Post ID: @Z2I37J2-1gtr

Previous post was too long and the first line got cut off.

"Found an AI algorithm on an AT&T post. "

by
|
Post ID: @Z2I37J2-1liz

Found this on an *AT this is calculated by the new software (check your system tray on Windows or Mac) by making one random sample per minute on average; more than 30 random samples in one hour that determine an "idle" (non moving/non typing) state per minute flags that hour as "mostly idle". This prevents employees from using a "drinky bird" device to hit keys, which generally do so at a much more regular interval; random sampling defeats that. (half kidding about that; just pointing out that any sort of attempt to circumvent it will likely not work)

C = using internally developed voice technology via webex (this is why they're pushing webex), C is the number of times the employee's line is transmitting noise above a 45dB threshold just prior to the meeting host asking everyone to "please mute" or "We're getting a lot of voices on the line" (it uses AI to determine the phrases); If there is one piece of advice this shows, it's to mute your phones/mics when on a call. Loophole here: The voice tracking software is not integrated with Skype, so if you primarily use Skype, it will not affect you.

D = the number of hours on which the employee spent calls (just here to normalize this factor; however, if you have been the target of a lot of "please mute" events, notice you can dilute this factor by scheduling and attending a lot of extra calls, even if just for 15 minutes).

E = the number of pages of personal printing the employee does on workplace 2020 printers; the "print release" centralized software efficiently scans all print jobs and has been making this determination since October 2016. It does not matter how much material (i.e. toner) is on each page; it's simply number of pages.

F = the number of personal (non work related) websites visited. This is a tricky one; it will look at the domain name primarily; subsequent visits to pages under the same domain will not count further against you as long as they are made within forty five seconds of the initial hit. There are some loopholes here, the best I can tell: sites determined by CSO software to be work-related include things like stackoverflow for developers. Google searches by themselves are likewise NOT counted against you. However, just after glancing at a brief list, it appears that CNN is NOT considered work related (just pointing it out for being ironic). That may change in the formula over time.

G = your base salary in U.S. dollars. Simple enough. Does not count team award/KCA or any other special awards

H = a factor based on your geo zone. This seems to me to double-dip a bit with G, above. However, everyone as you may know is in a geo zone tied to their permanent work location, N1 through N4. The higher the cost of labor in that zone, the more expensive the employees is not only by salary but also for things like employment taxes, cost of office space, utilities etc. So this factor is N1=1, N2=2.5, N3=4, N4=6. So this factor really hurts people in high-cost cities like New York, San Francisco, etc. If you're in one of those, you might consider getting out if you think your job is in jeopardy. The geo zone is only sampled once; it was taken November 1 already, so you can't affect it this round.

I = a factor based on your appraisal rating ("impact") for the previous calendar year. Extraordinary Impact = 0, Meaningful Impact=1, Limited impact=4. You can see the outsized effect they are placing on this one; Limited Impact folks are getting hammered by this factor.

J = this allows GM level (director) and above to set a fudge factor for extraordinary circumstances. It can be negative or positive and could help push the employee up or down a little bit. It's rather limited though in that the range that they can select in Amplify) is only -20 to +20. This tool is available the Supervisor area in eCorp right now; however, it's grayed out and won't be available for managers until the scores are calculated company-wide.

It's estimated that the calculations will be run as soon as one year of data is gathered across all the above factors, which will be completed on May 1. On May 6, scores will be released to supervisors, who can then work with their GMs to input the fudge factor. Then top "scorers" will be ordered to determine the impact to the business. If there is too much impact in one area, they may cut it off for one business unit and hit another business unit harder. But it's expected that those affected will be automatically notified via email by May 15.

by
|
Post ID: @Z2I37J2-1kzt

Post a reply

: