AI in the Workplace
In this week’s lecture, Craig Martel from LinkedIn said that there are several billion inputs used in LinkedIn’s job-matching algorithms. These algorithms can create immense value for both organizations and individuals, as each has a high incentive to find ideal matches. Thinking about the amount of investment placed on recruiting, I wondered if companies are fully utilizing technology to evaluate their internal employee’s performance.
Are organizations using AI to evaluate and monitor their talent?
From my experience in consulting, many organizations do a poor job at understanding not only what competencies and behaviors lead to success, but also in knowing which employees are actually driving their success. If an organization doesn’t know which employees are driving their success, how can they predict—much less recruit for—the behaviors, competencies, and profiles they need from their future employees for continued success? While there is a wide spectrum of sophistication in the manner and success of evaluating talent, I have found that many organizations are using surrogate measures—instead of direct measures—paired with AI to gauge the success of their workforce.
For example, this past week, a news story broke which stated that “Walmart just won a patent for audio surveillance technology that measures workers’ performance, and could even listen to their conversations with customers at checkout [1].” This technology would help Walmart gain a deeper understanding in areas such as their employee’s workload, efficiency and customer service. It would not, however, help them have a clear idea into the underlying behaviors, competencies, and profiles of employees that predict success. Surrogate success measures abound in the workplace. A quick search for these measures yielded measures such as [2]:
- Clock-in and-out times and patterns
- Hours worked
- Content and volume of digital communication (e.g., email, text messages, Facebook, WhatsApp, and any other commonly used app)
- Web browsing patterns
- Number of keystrokes
- Employees pictures and screenshots (the company Crossover has a disturbing product called ‘WorkSmart’ that takes pictures of employees and their screens every 10 minutes to evaluate productivity)
- Location, through GPS tracking devices in cars, phones and occasionally employee badges
- Political and religious leanings
While these measures could be important and can act as a proxy for success, they will require a direct line of sight to their impact on the overall success of the company to determine real correlation between their measures and organizational success. With no direct line of sight, organizations tend to 1) benchmark the costs and efficiency of these individual activities, 2) evaluate each employee’s performance against those benchmark, and then 3) use the evaluation against the benchmark as the yardstick for success.
Regardless of the measures, in this new age of data, companies need to gather all direct and surrogate measures and “favor analytics over [their] gut instincts’ in their workforce decisions [3].” They need to determine what the good vs bad outcomes look like. They then need to classify employees as top vs bottom performers and rigorously assess their employees’ behaviors and psychometric profiles. By doing this, they will be able to create a description of the behaviors and profiles correlated with both success and failure. From this data, they can finally build job descriptions and assessment tools to attract, assess, and hire the candidates with the highest probability of success. While this work might not seem sexy, it can likely be automated with advanced algorithms and would fit Craig Martel’s description of AI as being “dirty, hard, sweaty, and boring.”
[2] https://www.theguardian.com/world/2017/nov/06/workplace-surveillance-big-brother-technology
[3] https://hbr.org/2010/10/competing-on-talent-analytics?referral=00134
2 comments on “AI in the Workplace”
Comments are closed.
That is very disturbing that Crossover takes pictures and screenshots every 10 minutes. That can’t be good for workplace community/environment. If anything, people would be working out of compliance and probably be less productive overall but still check all the boxes to look like they’re doing a good job. I could be wrong, but it would be interesting to see some of the results from a study on this if there is one.
I like your article and especially how you try to provide value for companies by creating a more accurate, AI-based employee evaluation. However, I have a different view about whether it will ultimately add value to the companies.
My opinion is backed by an interesting article I have found on CMS Wire (Morphy (2018): https://www.cmswire.com/digital-workplace/will-artificial-intelligence-write-performance-evaluations-one-day/). In this article, Jason Wisdom, CTO of an AI-powered startup, describes why he thinks that AI should not be applied to performance evaluation. Basically, he points out that an employee’s performance is more than just a pure combination of metrics. Even though such metrics might be useful, they can bias you without a review. So according to him, the only solution might be a hybrid approach (Wisdom, 2018, in: Morphy (2018)).
But let’s just think about a scenario in which a completely AI-based employee evaluation system is implemented: For every one of your actions you need to fear to be screened. Let’s face it: Screening an employee the whole day will theoretically result in a better, more accurate evaluation. But that is only the theory, in reality, the following will happen: Employees feel screened and the average performance will go down. There is an interesting article about that which shows that too much control is negatively correlated with performance: https://hbr.org/2010/05/why-controlling-bosses-have-un.html.
So in my view, the thought experiment is super interesting but we would never want such a “perfect” AI-based employee screening. Ultimately human beings should be the one saying whether they want to work with someone else. But this is a good example of the limitations AI potentially has!