AWS Partner Network (APN) Blog
Redefine Crisis Management with Infosys’ Solution powered by AWS Generative AI
By Saibal Samaddar — Infosys Consulting
By Tanushree Halder — Infosys Consulting
By Supriya Pandey — Infosys Consulting
By Sanchit Jain — Infosys Consulting
By Dhiraj Thakur — AWS
 |
| Infosys |
 |
Critical infrastructure like consistent and predictable electricity supply serves as a foundation for a country’s development and advancement. India’s development pace relies heavily on a consistent energy supply. Any interruptions in this vital resource can significantly hamper progress across industry sectors and frustrate citizens who depend on the electricity supply for everyday work making customers raise complaints. Indian customers use multiple channels like voice calling, social media and word-of-mouth to show their frustration over energy crisis. This needs a redefined approach to manage the crisis using the AWS powered Generative AI solution built by Infosys.
Indian users raised 76,321 complaints across social media platforms in June 2023 as per the transparency reports of four platforms. Additionally, around 80% of the youth population actively uses social media in India1. Social media has become a medium for expressing one’s views openly. When citizens share their experience of electricity cuts or shortages on a social media platform like Twitter, Facebook, or Instagram, they not only invite attention to the issue but also escalate it for further discussions. Hence, catering to public concerns on social media becomes important. It helps maintaining public trust, addressing immediate needs and emergencies, and gathering valuable feedback for experience and process improvements.
In this blog post, Infosys provides insights on the solution and how it helped one of the largest electricity distribution companies in India to manage their customer complaints and ensured continuity of trust with its customers. Infosys-led Generative AI-based Social Media Response Management solution was built in collaboration with AWS by utilizing Amazon Bedrock. As Generative AI evolved even during the course of the development cycle, AWS stepped in to provide support and guidance in utilizing all latest Generative AI services which produced better results and gained client satisfaction.
The crisis
The electricity distribution company follows the conventional process of responding to the customers on social media, manually through a third party. This approach is effort-intensive and slow. During the surge times like storms, heatwaves, or natural disasters the volume of customer complaints increase dramatically. The company faces a dilemma – whether to focus on cost reduction through optimization of the number of agents handling customer complaints or to focus on improvement of customer satisfaction by increasing the number of agents. The inefficiency of manual processes often invites more frustration and dissatisfaction from customers.
Additionally, electricity supply and its availability are closely monitored by political parties. A few unsatisfied customers or those who see it as political tool, use this opportunity to malign the image of the government. They often resort to vulgarity, tagging the political leaders of the state onto social media and blame-shifting onto them. This behavior causes undue political and social pressure on the company further complicating its efforts to address customer concerns. Hence, satisfying the customers by promptly attending to concerns and raising tickets for immediate resolution has become crucial. The company needed to find a sustainable solution that met both financial and customer service goals.
This is where the advent of AWS Generative AI has proven to be a boon to the organization.
The solution employs understanding of customer posts on various channels for negative, positive, and neutral sentiments and categorizing these into various services – supply related issues, voltage fluctuations, billing related issues, emergency situations and few other broad categories. The solution is fine-tuned to understand the local vernacular language and the aspects of sarcasm present in the social media posts. The Generative AI solution integrates with the CRM system of the organization to raise complain tickets. Once the complaints have been raised, a personalized response along with a complaint number is generated and posted for each customer under specific categories in near real-time.
Infosys combines the solution with aspects of ethical AI powered by a comprehensive framework. Each of the generated responses are checked across multiple frameworks of “customer justice” to ensure that customer is being provided with appropriate information to rebuild the trust with the firm.
Figure 1: Process flow diagram
Solution overview
This Infosys Topaz solution leverages a tech stack for deploying the AI solution that is designed to be robust, and scalable, utilizing a variety of AWS services and components. At its core, the architecture has an Developer portal API as the entry and exit point for handling complaints and ensuring communication, facilitating interaction between customer and bot on behalf of the organization. AWS Lambda plays a crucial role here by serving as a code deployment platform. This allows for the execution of the code in response to the events without the need for server provisioning or management. Comprehensive monitoring and error tracking are provided by Amazon CloudWatch, enabling proactive identification and resolution of issues that may arise within the system. Amazon EventBridge Scheduler is used to create, run, manage and schedule tasks at scale and reduce maintenance costs. Amazon Sagemaker studio utilized here, hence emerges as a comprehensive toolset for building, training, and deploying machine learning models. Amazon EC2 instances act as the virtual motherboard of the architecture, providing a virtual computing environment for running applications and services with flexibility and scalability. By leveraging AWS services and components, the system is well-equipped to handle various tasks and workflows associated with filtering, analyzing, and responding to consumer queries and complaints.
Figure 2: Architecture diagram
- The solution utilizes social media APIs like X API, and Facebook API to extract mentions of the client’s organization.
- AWS Translate helps in translating the posts into English as most of the customers use their native language to show their frustration
- The AWS Bedrock Anthropic Haiku model is used to analyze the sentiment of the posts and then categorize them into supply related, voltage-related, and brand-related topics.
- The solution uses few-shot prompting techniques to provide these tags to posts – sarcasm, political, abusive, supply related and so on.
- It also includes models that are fine-tuned on the client’s response guidelines to customers and the customer justice framework.
- A custom score is done using Tableau reporting the improvement of customer satisfaction using “Customer Mood Index”. This is done to provide insights on how the solution over time develops trust in customers by identifying their problems, responding to them immediately, redirecting the issues to the right department and providing appropriate customer justice.
It also includes models that are fine-tuned on the client’s response guidelines to customers and the customer justice framework.
Reporting the improvement of customer satisfaction using “Customer Mood Index” a custom score is also done using Tableau. This is done to provide insights on how the solution over time develops trust in customers by identifying their problems, responding to them immediately, redirecting the issues to the right department and providing appropriate customer justice.
Largely, there are two customer needs. One is the need for resolution of their individual problems, and the other is associated with engaging in state or country-level public issues. While the customer aims to tackle both efficiently, they have to be separated first.
Using AWS services, we distinguished between these by using classification and close analysis of typical words, emotions, and context helping in differentiating problems from complaint posts. Sentiment analysis is done to gauge the emotional tone to identify complaints. Identifying a politically motivated post, emotionally charged cases, vulgarity, fake news, or hate speech disguised as electricity complaints on social media involves multi-layered analysis and intervention. Such cases are to be handled with more care. Thus, the solution is not aimed at eliminating, but rather involving the human in the loop for more critical cases. The flagged instances are then reviewed by humans, who apply their judgment. This ensures that legitimate complaints are addressed while inappropriate contents are managed effectively.
This solution helps in safeguarding public discourse by preventing the spread of misinformation. By integrating AI and human oversight, this approach ensures a balanced, efficient, and socially responsible management of online interactions. This enhances trust and engagement between utility providers and the public. Ethical AI caters to customer’s perceived fairness of the complaint handling process. It ensures their integrity is protected by the newly implemented AI-based complaint handling system. We have codified the ethical AI framework to ensure that customers do not receive non-compliant responses on social media.
Customer benefits
The solution automated around 80% of the cases around supply and voltage concerns raised by customers. The response time including the time to generate complaint numbers in the CRM was brought down from 7-8 minutes to less than a minute. It increased the productivity of the customer relationship team of the organization by ~20% and helped them reduce operational costs. The costs that were related to agents spending time in negotiations, streamlining, and enhancing the efficiency of conflict resolution were reduced.
It also revolutionized user interactions by limiting the use of vulgarity, ensuring a more respectful communication environment.
The solution helps in bringing transparency to their customers, and immediate action is taken against the registered complaints.
Conclusion
Through its comprehensive design, this AI model emerges as a versatile tool. It is dedicated to enhancing customer satisfaction globally while upholding principles of fairness and respect in online engagements. At Infosys, we aim not only to meet but exceed our customers’ expectations, fostering trust, loyalty, and satisfaction towards the organization. The proactive approach – combined with insights from data analytics and sentiment analysis, powered by Infosys Topaz, demonstrates our commitment to customer satisfaction and also helps us allocate resources efficiently, maximizing the ability of the organization to address customer concerns promptly.
To align with the need for ethical AI, this solution model advocates a mix of approaches while handling complaints in an organization. The mechanistic approach emphasizes structured procedures, clear hierarchies, and standardized responses to address complaints while the organic approach focuses on prioritizing flexibility, empathy, and adaptive responses tailored to individual needs and contexts. The model has been trained on the same lines and these approaches can be traced in the responses generated while answering concerns of customers. By incorporating the mechanistic approach, it ensures consistency, transparency, and accountability in their process. Meanwhile, the organic elements lead to fostering trust and satisfaction. A striking balance between the two enables them to address complaints effectively while demonstrating understanding and care for their stakeholders’ concerns.
In the age of rapid innovation and AI, the focus has already shifted beyond mere process transformation and has moved towards experience transformation. By embracing Generative AI and empathy in equal measures this solution model is catering to customers in anguish in a balanced way – setting a benchmark across industries.
For more details and implementation, contact the Infosys team.
References:
- https://www.medianama.com/2023/08/223-india-social-media-user-grievances-june/
.
.
Infosys — AWS Partner Spotlight
Infosys is an AWS Premier Tier Services Partner and MSP that enables clients to outperform competition and stay ahead of the innovation curve.