IBM's human resource AI can predict when employees are about to leave the company

Polycount

TS Evangelist
Staff member

The corporate world seems to get stranger by the day. With the growing popularity of artificial intelligence, it's no surprise to see companies use it to boost productivity and automate tasks.

However, one of IBM's internal AIs can now do something a little bit creepier. According to a report from CNBC, IBM's patented "predictive attrition program" -- revealed to the outlet by CEO Ginni Rometty -- uses machine learning to analyze employees and determine when they might be about to quit.

Rometty says the accuracy of these predictions is around the 95 percent mark, adding that the program has saved IBM roughly $300 million in "retention costs." Those are impressive numbers if they can be trusted.

Speaking of retention, Rometty said at CNBC's "@ Work Talent + HR Summit" that "the best time" to reach an employee is before they go.

IBM's CEO, Ginni Rometty

That statement may seem a bit obvious at first, but she likely means it's easier to convince an employee to stay than it is to persuade them to come back after they're already gone.

So, what does "reaching" an employee entail? According to Rometty, it may mean promotions, education benefits, or even financial incentives.

The CEO isn't divulging any of the specifics behind how this AI works, probably to prevent IBM's competitors from swiping the secrets for themselves.

However, CNBC claims the AI is successful because it "[analyzes] many data points," which could mean it looks at things like job satisfaction (likely determined through regular performance reviews) and the frequency of arguments or disagreements with other employees and supervisors. Of course, that's pure speculation on our part.

What do you think about IBM's semi-AI-powered HR system? Let us know in the comments.

Permalink to story.

 

brucek

TS Maniac
I'm really curious what the inputs to this are.

At least at companies I've worked for, formal entries into an employees "HR file" are pretty infrequent - perhaps no more than one performance review per year for many. It is hard to see how an AI could give a timely prediction working from that. (Unless it's as dumb as saying hey when we see negative entries in the file we predict they may be departing, which is hardly rocket science.)

So what are they looking at? Are they scanning emails? Key card swipes? Voicemails? Phone calls made and received? Network usage? Curious minds want to know...

Edit: and furthermore, even if they had a NSA style analyze everything system in the works, I still can't fathom how it could possibly get to 95%. You'd think there'd be at least 5% of departures that happened in response to unpredictable external events -- I.e., need to move to support family after illness, non-solicited recruitment, spouse needs to move, etc etc.
 
Last edited:

Manrubio

TS Rookie
If only we could predict when "that next guy" is loading his AR-15 and getting ready to do something terrible.
Yes Sir! Hopefully the Artificial Inteligence gods will hear you .

Thus, now that I think about it... Hydra already did that but the avengers stopped it. LOL
 

psycros

TS Evangelist
If only we could predict when "that next guy" is loading his AR-15 and getting ready to do something terrible.
That would the 1 of 5 guys the company just laid off because they thought (incorrectly) he was about the leave the company ..........
I admit that was darn good but I can't help spoiling it by stating the obvious. If you quit a job that has "at-will" employment with no severance agreement then they owe you nothing, not even unemployment payments.
 

mbrowne5061

TS Evangelist
If only we could predict when "that next guy" is loading his AR-15 and getting ready to do something terrible.
That would the 1 of 5 guys the company just laid off because they thought (incorrectly) he was about the leave the company ..........
Sounds like this algorithm is more to identify who to give a bump in pay or title to because they're planning on jumping ship.

As for what this AI is looking at, they say people don't leave companies, they leave bosses. If I wanted to predict attrition, I would be looking at managers, not so much the managed. Identify the bosses that are pissing off their employees, target the employees you want to keep, swoop in with a raise and promotion (to work for a different boss) when they're getting ready to leave (which can probably be predicted by present market demand for skills - who is going to get recruited first).
 
  • Like
Reactions: Mr MR

elementalSG

TS Addict
I'm really curious what the inputs to this are.

At least at companies I've worked for, formal entries into an employees "HR file" are pretty infrequent - perhaps no more than one performance review per year for many. It is hard to see how an AI could give a timely prediction working from that. (Unless it's as dumb as saying hey when we see negative entries in the file we predict they may be departing, which is hardly rocket science.)

So what are they looking at? Are they scanning emails? Key card swipes? Voicemails? Phone calls made and received? Network usage? Curious minds want to know...

Edit: and furthermore, even if they had a NSA style analyze everything system in the works, I still can't fathom how it could possibly get to 95%. You'd think there'd be at least 5% of departures that happened in response to unpredictable external events -- I.e., need to move to support family after illness, non-solicited recruitment, spouse needs to move, etc etc.
Just thinking about it, scanning emails for things like these might reveal a worker about to leave:
1) Employee stops receiving more personal email-type items (like class grades for kids, school notices, etc.)
2) A decrease of direct emails to coworkers (no cc's) than before (would indicate friendships between coworkers). Bosses don't count, of course.
3) A decrease in group emails coworkers are on (including business related ones). Being cc'd less. This worker is probably becoming less and less involved in company work.

For keycard access, if you see things like:
1) Time when coming to work keeps creeping later and later
2) Lunch re-key-in times changing

For computer, well there's so many obvious things you could be looking for if the corporation uses software to screen capture every few minutes and send to AI to identify from the image what the employee is looking at. If the amount of time where screen captures of non-work related things rise, most likely something is up.

Definitely creepy, but then all the big companies are monitoring every way they can anyways these days. Looks like the IBM AI is just letting them make good use of all the data coming from the monitoring.
 

McMurdeR

TS Addict
Most modern HR functions would be trying to understand what's driving attrition in their organisations, especially high performers. AI usually wouldn't be a great fit though - do IBM really loose enough people to provide them with enough data to train and validate a predictive model using AI techniques?
If they do than they definitely need it!
 

Zorak

TS Rookie
As a CEO of our company I can predict with 100% accuracy when somebody is going to leave us.
 

paul s2

TS Member
I think like used up all sick days were has not before
looked into 401k transfers
and used all vacation days
 

lazer

TS Addict
If you have a good boss, he will realize that you are not happy and try to keep you. If you have a boss like the PHB of Dilbert, you are better off leaving.....
 

woofer

TS Enthusiast
Now it would be nice if employees also could use AI to predict when they are going to be laid off as IBM did to me and a few thousand others back in 2007. Actually, that was the best outcome for me ("free at last") as I took skills I learned there to go back to my previous employer, in dire need of that skill set just then, who was a LOT more generous in laying me off 8 years later just as I turned 65, and was going to retire anyway, but got a very generous 18-year-based layoff package to send me on my way into retirement. IBM does not offer such deals of service-based layoff packages these days.