# The impact of AI on the customer support industry

This <mark style="color:blue;">**April 2023**</mark> paper explores the impacts of a AI-based conversational assistant on customer support agents' productivity and learning, using data from 5,179 agents.&#x20;

{% embed url="<https://arxiv.org/abs/2304.11771>" %}
Erik Brynjolfsson, Danielle Li, and Lindsey R. Raymond
{% endembed %}

### <mark style="color:purple;">Key findings include</mark>

<mark style="color:green;">**Productivity Gains**</mark>

The introduction of the AI tool *<mark style="color:yellow;">**resulted in a 14% increase in productivity across the board**</mark>*.

However, the gains were not uniformly distributed; *<mark style="color:yellow;">**novice and low-skilled workers saw a 34% increase in issues resolved per hour**</mark>*, whereas experienced and high-skilled workers experienced minimal impact.&#x20;

This suggests that the AI tool effectively disseminates best practices and accelerates the learning curve for less experienced workers.

<mark style="color:green;">**Customer Sentiment and Employee Retention**</mark>

Besides boosting productivity, the *<mark style="color:yellow;">**AI assistance improved customer sentiment toward agents and increased employee retention rates**</mark>*, particularly among newer workers.&#x20;

This implies that AI tools can enhance the quality of customer service and the workplace experience for agents.

<mark style="color:green;">**Worker Learning**</mark>

There is suggestive evidence that *<mark style="color:yellow;">**interaction with the AI tool contributes to worker learning**</mark>*.&#x20;

Even during software outages when the AI was unavailable, agents who had more exposure to the AI assistance continued to show improved productivity compared to their pre-AI performance levels.

<mark style="color:green;">**Learning Mechanism**</mark>

The paper posits that *<mark style="color:yellow;">**generative AI systems capture and distribute the tacit knowledge and successful patterns of top-performing agents**</mark>*, making these insights accessible to less experienced or skilled workers.

<mark style="color:green;">**Impact on High-Skill Workers**</mark>

Contrary to the common narrative of skill-biased technical change, where new technologies tend to benefit higher-skilled workers more, this study finds that *<mark style="color:yellow;">**generative AI tools can disproportionately aid less-skilled workers**</mark>*, potentially flattening the productivity distribution within firms.

<mark style="color:green;">**Implications for Future Research**</mark>

The paper calls for further investigation into the broader economic implications of AI adoption, including its effects on wages, labour demand, and skill requirements. It also raises questions about compensating workers for their contributions to the training data that power AI systems.

### <mark style="color:purple;">Economic Impacts</mark>

The economic impacts of generative AI, as discussed in the paper, highlight a significant shift in how computers interact with the workforce.&#x20;

Historically, automation and computerisation have primarily affected routine tasks, leading to a decrease in demand for workers in roles like data entry and an increase in demand for roles requiring complementary skills, such as programming.&#x20;

This shift has contributed to rising wage inequality.

#### <mark style="color:green;">Generative AI tools, however, operate differently</mark>

Generative AI tools do not require explicit instructions to perform tasks but learn from examples, allowing them to undertake activities that involve tacit knowledge, which is knowledge gained through experience and difficult to articulate.&#x20;

This ability *<mark style="color:yellow;">**enables generative AI to perform non-routine tasks that rely on judgment and experience**</mark>*, expanding the scope of tasks that machines can handle and potentially affecting jobs that have been less impacted by previous waves of automation.

However, the deployment of AI in the workplace faces challenges, such as the generation of false or misleading information and the unpredictability of real-world problems compared to controlled laboratory settings.

In addition, the effectiveness of these AI tools in the workplace will likely depend on the interaction with existing organisational structures and may require additional investments or redesigns in business processes.

### <mark style="color:purple;">Customer Support</mark>

The section focuses on the application of Large Language Models (LLMs) for customer support within a Fortune 500 enterprise software company, emphasising the context and potential of generative AI in this domain.

<mark style="color:green;">**Context of Customer Support**</mark>

The customer service industry, marked by high turnover and significant training costs, is increasingly turning to AI tools to enhance efficiency and address workforce challenges. In this setting, customer-agent interactions, critical for maintaining company reputation and customer relationships, vary widely in productivity.

#### <mark style="color:green;">**Generative AI in Customer Support**</mark>

Customer support is an ideal application for generative AI, where conversations can be seen as pattern-matching problems.&#x20;

AI tools can help agents identify and resolve customer issues more effectively by providing real-time suggestions based on past successful and unsuccessful interactions, enhancing the overall customer experience.

#### <mark style="color:green;">**AI Firm and Data Firm Background**</mark>

The study is conducted in collaboration with an AI firm providing AI-based customer support software, focusing on its deployment at a data firm - a Fortune 500 company specializing in business process software.&#x20;

The data firm employs numerous chat-based technical support agents, primarily in the Philippines, to assist U.S.-based small business owners.

Initial observations indicate that agents with AI assistance resolve more chats per hour and exhibit improved performance metrics compared to their pre-treatment and never-treated counterparts. &#x20;

While customer satisfaction remains consistent across groups, post-treatment agents show a notable decrease in average handle time.

The section sets the stage for a detailed examination of how AI deployment affects agent productivity and customer interactions in customer service settings, highlighting the potential for AI to disseminate best practices and improve performance, particularly among newer and less experienced workers.

### <mark style="color:purple;">Conclusion</mark>

The conclusion of the paper emphasises the transformative potential of AI technologies in the workforce, presenting empirical evidence from a real-world setting where a generative AI tool was deployed in customer service. Key findings include:

<mark style="color:green;">**Productivity and Customer Sentiment:**</mark> Access to AI-generated recommendations significantly boosts worker productivity, enhances customer satisfaction, and is linked to lower employee turnover.

<mark style="color:green;">**Dissemination of Best Practices:**</mark> The AI tool appears to capture and distribute the tacit knowledge of high-skill workers, making it accessible to newer and less-skilled employees, thereby democratizing expert knowledge within the organization.

<mark style="color:green;">**Impact on Different Skill Levels**</mark><mark style="color:green;">:</mark> While the AI tool substantially benefits newer and lower-skilled workers by improving their problem resolution and communication styles, it doesn't offer similar advantages to the most skilled or experienced workers.

<mark style="color:green;">**Long-term Implications:**</mark> The study opens up questions about the long-term effects of generative AI on job design, skill requirements, wages, and overall employment in customer service, with potential varying outcomes based on the demand elasticity for customer support.

<mark style="color:green;">**Worker Contribution to AI Training**</mark><mark style="color:green;">:</mark> High-performing workers contribute significantly to the AI's training data but do not see proportional benefits in their productivity, raising questions about appropriate compensation mechanisms for their contributions.

<mark style="color:green;">**Generalisability of Findings**</mark><mark style="color:green;">:</mark> The effects observed in this study might vary in different settings, particularly where product offerings or technical questions are more dynamic, affecting the utility and impact of AI recommendations.

In essence, the study highlights the nuanced and multi-faceted implications of integrating generative AI tools in the workplace, underscoring the need for further research to fully understand their broader economic and organisational impacts.


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://training.continuumlabs.ai/continuum-applications/discussion-and-use-cases/the-impact-of-ai-on-the-customer-support-industry.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.