How an A.I. chatbot is helping Genpact retain employees and predict attrition: ‘Someone who responds to it is two times as likely to stay with the organization’

Genpact cares about how its employees feel. So much so that it’s linked top leaders’ compensation to it. 

Thirty-year HR veteran Piyush Mehta, the CHRO of Genpact, is leading the company’s HR practice with the help of a suite of internal tech tools that track and detect employees’ moods and sentiments. The goal: increased retention. 

The professional services company has a global workforce spanning India, the U.S., and the U.K. and over 115,000 global employees. With an average employee age of 28, Mehta is drilling down on how to better connect and understand such a large and dynamic organization. He spoke with Fortune about how Genpact is utilizing an A.I. chatbot and organizational network analysis to better serve and engage employees. 

This interview has been edited and condensed for clarity.

---

Fortune: Before I jump into the tech of it all, tell me how you’ve seen your job in HR change over the past 10 years.

I’m admittedly a cynic about many things said about what has changed, having started in the profession 30 years back. The reason people came to work even then, though it may not have been articulated as such, was the company’s purpose, growth opportunities, and the quality of their job and manager. That continues to be the most important reason why people come to work today.

What has changed is the patience people have for these things. What has changed is the experience that people want. What has changed is the technology that enables this. What has changed is the data and the analytics that have become a catalyst to better take care of the needs of our people. But fundamentally, the needs of why people come to work have not changed.

Can you share more about how Genpact is leveraging technology to assess employee engagement and connection? 

Every company has a purpose and culture to some extent. But they’re just empty words if you don’t bring them to action. Pre-pandemic, absorbing the corporate culture happened by osmosis because people were coming to work, interacting with others, and picking it up in an unintentional and unorchestrated way relative to what we do now. During the pandemic, there needed to be more intentionality in how we do this.

So there were three or four things we looked at. The first was, are people engaged? To measure engagement, we created an A.I. chatbot. To be fair, we did this before the pandemic, but we upped the ante to a different level during and after the pandemic because it was hugely useful for us. It’s called “Amber.”

How does it work?

So it is an A.I.-driven chatbot that learns as it interacts with people. A regular survey didn’t work for us because you have to orchestrate it each time, and there is a fixed interval by which it would act, which would then interact with the population and create results. Those results would then be run through data and analytics, provide insights, and then action. But we needed to measure mood and organization sentiment on an ongoing basis. Amber reaches out to employees in real-time, at least four times throughout the year for the average employee and eight to 10 times in the first six months for new employees.

Amber asks employees questions. For example, it reaches out to me as I’m working on my laptop and says, ‘Hey Piyush, can I talk to you for a couple of minutes? Can we chat?’ And I chat. It asks, ‘How are you doing? What’s working for you, and what’s not working?’ At the end of the conversation, it generates a mood score for me based on learnings from previous discussions. The reason I say it’s driven by A.I. is that Amber asks questions based on my earlier responses, learns, and becomes more intelligent as it goes along.

What do you look for in the data and mood scores you receive?

Amber provides me with real-time intelligent analytics on the organization’s mood at any point and the most micro level possible. I can look at data by individual and across the enterprise. That was huge during the pandemic. Response rates are about 65%. It’s phenomenal, considering it asks you questions four times a year. Someone who responds to it is two times as likely to stay with the organization as someone who doesn’t. We love measuring the hell out of everything.

We also said during the pandemic that employee mood and sentiment would be hugely important, so we linked 10% of our CEO and top 150 people’s bonuses to Amber’s mood score. We changed our bonus pool matrix when the pandemic hit—in the middle of the year—and said this is so important that it will be the only non-financial metric in our bonus pool.

Do employees still have traditional check-in meetings with supervisors, or is this the new, more transparent way of doing things?

Fundamentally, that is still an employee’s most important touch point, and Amber supplements it. Think about our organization: 115,000 people and 10,000 supervisors. By definition, you’re going to have 5,000 supervisors above the median in their ability to manage their people and 5,000 below. Amber is enabling them to better manage their people.

The responses go to managers as well. How do you ensure employees are answering honestly and not putting on a facade that they’re happier than they really are?

There is no pressure on people to respond, and about 35% of people don’t respond. If I’m not happy, I’m not responding. Because our positive mood score is upward of 85%, we see a huge variation in how we get there. Some are at 40%, and others at 95%. As long as we keep seeing the variation, we know people are being transparent.

We also have strict guidelines. You cannot victimize based on Amber’s feedback because if you have a team of 10 and one person feels victimized for their responses, she’s going to talk to the other people in her peer group and tell them not to respond, causing the response rate of that group to fall. The manager is accountable if their response rates are lower than others.

What are some actions or policy changes your team has made in response to Amber findings?

We get a lot of feedback from Amber about career progression. Employees felt that when they were placed in a role, they were forced to stay there for 24 months before applying for other company positions. We didn’t think 24 months was a long time to be put in one role, but many people were leaving us. So rather than lose those people to the external world, we decided to allow them the opportunity to move through roles internally in less than 24 months.

Genpact notably utilizes organizational network analysis in evaluating people. Can you expand upon what that is? 

Organizational network analysis looks at the metadata of networks within organizations. So it reads the subject—not the content—of emails, looks at patterns, and draws an analysis.

One application, which is profound because we are a high attrition business, is the ability to predict attrition before it happens based on specific metrics. That provides me an opportunity to jump in and say individual X has a higher proclivity to resign than individual Y, and I need to do the following things. There are data privacy implications, especially in markets like Europe. But we can do it in other areas because it’s within legal constraints.

What does your team do when Amber flags someone as likely to leave?

We tell managers to get those employees on to Genpact’s internal learning platform, so they can grow. We also look at reward and recognition data; when was this person rewarded last, especially if they’re a reasonable performer? The third thing we look at is how long the person has been in the role. If they’ve been in the position for a few years and are bored doing the same thing, we move the person to another function.

Amber Burton
amber.burton@fortune.com
@amberbburton

Around the Table

- The guy who inspired the phrase “quiet quitting” is back to working 50 hours a week. Business Insider

- Emotional intelligence was once considered unimportant for running a successful business, now people leaders obsess over how to infuse it across their organizations. Listen time: 47 min. HBR

- The National Labor Review Board filed a complaint on Wednesday alleging that Amazon CEO Andy Jassy’s comments against unions may have violated labor laws. The Washington Post

- Employers have been ghosting job candidates at almost double the rate they were pre-pandemic, according to data from Glassdoor. Fast Company

Watercooler

Everything you need to know from Fortune. 

401(k) going up. Employees may want to adjust their 401(k) contributions in 2023 as the limit will increase from $20,500 to $22,500 in 2023. —Alicia Adamczyk

TikTok death. A judge in Philadelphia ruled TikTok wasn’t liable in the death of a 10-year-old girl who died recreating a viral video the platform recommended to her. —Robert Burnson

Rejected, again. A railroad union representing 6,000 workers rejected the latest collective bargaining agreement, making it the second to do so since September. —Josh Funk