Ally Financial Embraces AI in Contact Centers for Enhanced Efficiency
Ally Financial has embarked on a transformative journey by embracing generative artificial intelligence (AI) technology. The Detroit-based bank has accelerated the adoption of AI, revolutionizing the efficiency of its contact centers. This move comes after the success of ChatGPT in late 2022, which prompted Ally Financial to form a dedicated working group around generative AI in early 2023.
To kickstart their AI initiative, Ally Financial engaged in discussions with tech giants Microsoft and Amazon in Seattle back in February. Subsequently, the bank sealed a contract with Microsoft to utilize its enterprise-grade generative AI software in April. The team at Ally Financial wasted no time in developing their proprietary cloud-based platform, Ally.ai, specifically designed for AI-related projects. By June, they had started building this cutting-edge platform, and at the end of the same month, they launched a pilot to test its efficacy. The successful pilot moved to production on July 31.
Sathish Muthukrishnan, the Chief Information, Data, and Digital Officer at Ally Bank, highlighted their commitment to constant progress, stating, We do not want to stand still. Ally.ai serves as the bridge between external large language models (currently Microsoft’s GPT 3.5), generative AI technology, Ally’s internal applications and data, data security measures, and human intervention, at least for now. The bank’s priority lies in managing risks such as hallucinations while ensuring the protection of customer information.
Ian Watson, Head of Risk at Celent, explains that banks are currently focusing on three key areas: refining their data foundations, selecting suitable large language models, and experimenting with various use cases. Ally Financial chose to begin their generative AI journey with their contact center as the first use case. Traditionally, contact center agents are burdened with the task of recording notes and summarizing call details for regulatory purposes and quality customer service. In the pilot phase, AI technology was introduced to transcribe conversations in real-time using the Ally.ai platform and create accurate call summaries. This approach aimed to relieve agents of multitasking, allowing them to focus more intensively on customer interactions.
Muthukrishnan underscores the potential for increased productivity, stating, This has the potential to unleash productivity for us. While the generated summaries are currently reviewed manually by agents to ensure accuracy, the system has shown great promise. Starting with a low approval rate in the early stages of the pilot, it quickly rose to 78% by the end of July. Now fully deployed to over 700 agents, Ally’s generative AI initiative is yielding positive results.
Ally Financial acknowledges the need for human intervention as they continue to refine their models. This precaution is one of three principles the bank adopted before implementing generative AI. The others include learning and testing on internal customers (employees) prior to external deployment and the strict maintenance of personally identifiable information (PII) within Ally’s firewalls.
The risks associated with generative AI fall into two categories, according to Celent. Adverse outcomes such as bias, hallucinations, and false outputs constitute one group, while external threats like regulatory violations and cybersecurity breaches make up the other. There is a real danger of the models developing complete falsehoods, warns Watson.
Ally’s team encountered hallucinations during calls lasting less than a minute or when the line quality was poor, prompting them to refine prompts and prevent recurring instances. To enhance security, the bank has established a secure pipeline between itself and Microsoft and employs a dedicated GPT 3.5 model. Ally ensures that personally identifiable information (PII) remains within its firewalls. Furthermore, the foundational models are prevented from learning from Ally’s data, and personal data is forgotten after each session with a customer associate ends. The team conducts thorough tests and evaluations to guard against model drift and the introduction of bias.
Watson emphasizes that despite the risks, the most significant danger lies in not utilizing generative AI. The transformative potential of this technology extends to transforming business models, reducing mundane tasks, minimizing employee turnover, and providing customized marketing materials.
Ally Financial is actively exploring other potential use cases for generative AI, such as assisting in the creation of user stories for software features and answering basic questions about human resources benefits. Looking ahead, they envision leveraging this technology to gain a deeper understanding of customer needs and provide personalized experiences tailored to individual financial requirements.
The debut of Ally.ai aligns with Ally Financial’s transition to the cloud, as more than two-thirds of their applications are now cloud-enabled. Sathish Muthukrishnan highlights the significance of cloud infrastructure for AI, explaining that it requires substantial computing resources for tasks, memory, and storage. Running applications on the cloud facilitates horizontal scaling and on-demand infrastructure.
As Ally Financial enters an era of AI-driven contact centers, its adoption of generative AI stands as a testament to its commitment to innovation. With the successful implementation of Ally.ai, the bank is primed to achieve greater efficiency, improved customer service, and enhanced productivity.
