In one of the largest and longest studies of its kind, ongoing research from Stanford and MIT revealed nuanced differences in how generative AI can help different types of workers and the kinds of functions it assists with.
For context, the study involved over 5,000 customer support agents at a Fortune 500 enterprise handling software support for SMEs in the United States. These agents were primarily based in the Philippines and researchers tracked their performance over the course of a year.
Additionally, the AI model used was a recent version of GPT that had been fine tuned for customer service applications.
Here are the three main takeaways from the study:
1. Increased productivity across the board
When controlling for factors such as experience level and geographic location, the researchers still found a 14% increase in the number of calls handled per hour across the 5,000 workers. Workers also experienced a substantial increase in productivity in the first month of deployment, which increases slightly in the second month and thereafter remains stable until the end of the 12-month sample.
2. Least skilled workers benefit the most
The researchers separated the call agents into different categories based on their performance level in the quarter prior to the adoption of the AI system.
It was found that the lowest performing workers had the biggest increase in productivity at 35% more chats handled per hour. Similarly, newer workers who had only been with the firm for less than a month experienced a bigger jump in productivity compared to their colleagues who had been with the company for longer.
Finally, when comparing progress between new hires who used and did not use AI, the former were able to go from 2 calls per hour to 3 calls per hour within five months, while the latter took 8 to 10 months to go from 2 calls per hour to 2.5 calls per hour.
3. High skilled workers get distracted
While top performing agents using AI experienced smaller increases in number of calls handled per hour, they also had relatively greater reduced rates in customer satisfaction and resolution, which caused the researchers to suggest that the use of AI was a potential distraction for workers already doing their jobs effectively.
Implication for Malaysian SMEs
While we highly encourage you to read the study for full details, for those pressed for time, it would seem that use of AI in the workplace has the greatest potential for accelerating the onboarding of new hires and upskilling of low-performing workers.
In particular, the AI being used is known as a Large Language Model, which has been trained on a large corpus of language and can understand and produce grammatical and meaningful sentences. As the company in question exclusively served clients based in the US, It would not be unreasonable to assume that this was helpful for the agents with poor English proficiency. Considering Malaysia is ranked below the Philippines in global English proficiency level, this type of AI could be used to streamline communications in Malaysian workplaces.
Finally, it would seem that AI is still no replacement for high performing workers, and may even compromise their output. Application of AI in the workplace should therefore be situational rather than as a one-size-fits-all.