OpenAI Tech Helps Increase Productivity Of Philippine Contact Center Agents by 13.8%
-
Loading... - Salazar Liezel
- 19 May 2023
- 73 Views
- 0 Like
- 0 Comment
OpenAI's GPT, the expansive language model that powers ChatGPT, significantly enhanced the productivity of contact center agents in the Philippines, resulting in an average increase of 13.8% in resolving customer issues per hour.
Conducted by researchers from Stanford Digital Economy Laboratory and MIT Sloan School of Management, the study was published in April 2023 by the National Bureau of Economic Research, a US nonprofit research organization.
The research team examined 3 million chat interactions involving 5,179 agents employed by a Fortune 500 software firm specializing in business process software. Approximately 83% of these agents were based outside the US, primarily in the Philippines, where they provided technical support via chat to small business owners in the US who used the company's software.
The introduction of the AI assistant was a gradual process, primarily implemented between November 2020 and February 2021. This assistant monitored customer chats and offered real-time suggestions to agents on how to respond. However, the agents retained control of the conversation and were free to disregard the assistant's suggestions.
According to the researchers, this study represents the first large-scale examination of the impact of generative AI in a workplace setting. They also highlighted the high adoption rates of AI technologies in the customer service sector.
The observed 13.8% increase in productivity encompassed three key aspects:
1. Agents reduced the time required to handle individual chats.
2. Agents were able to handle a higher number of chats per hour.
3. Agents slightly improved the number of successfully resolved chats.
Notably, the productivity improvement was more pronounced among agents with lower skill levels and less experience, while higher-skilled and more experienced agents exhibited less improvement.
To assess agent skill levels, the research team utilized metrics such as average call efficiency, resolution rate, and surveyed customer satisfaction in the quarter preceding the AI system's adoption. They found that agents in the lowest tier experienced the greatest productivity improvement, reaching 35%.
Moreover, the AI assistant facilitated accelerated learning for agents. The researchers observed that agents with two months of experience using the assistant performed as effectively as agents with six months of experience who did not use it.
The researchers postulated that higher-skilled workers had less to gain from the AI recommendations since their own behaviors already embodied the tacit knowledge. AI recommendations, derived from data collected from experienced workers, were inherently embedded within the wisdom and experience of these higher-skilled, tenured agents. The researchers discovered "suggestive evidence" that the AI assistance influenced lower-skilled agents to communicate more like their high-skill counterparts, potentially leading to reduced attrition rates in the industry.
The researchers emphasized that AI systems could effectively share knowledge among workers by capturing tacit knowledge that was challenging for managers to articulate and by providing real-time recommendations. They noted an improvement in the way customers interacted with agents, as agents acquired job proficiency more swiftly with the aid of the AI assistant.
In conclusion, the researchers asserted that the collaboration between generative AI and human agents can significantly enhance productivity and retention rates among individual workers.
However, the study does not forecast future employment trends affected by AI. The researchers clarified that their findings do not encompass potential longer-term impacts on skill demand, job design, wages, or customer demand.
For instance, although more efficient technical support might lead to contact center agents assuming more complex customer responsibilities, thus increasing aggregate demand, it does not guarantee that agents will become more productive in every scenario.
The researchers acknowledged that the efficacy of AI recommendations, which are trained on historical data, may vary in rapidly changing product or environmental contexts.
Finally, just as creators of generative AI demand compensation for their training data, the researchers raised questions about whether highly skilled workers should also be rewarded for providing high-quality training data to AI systems through their exemplary work.
