Customer service was one of the earliest functions to yield tangible applications for generative AI.
From summarising cases behind the scenes and assisting agents on the fly to directly interfacing with customers, the use cases for generative AI in customer service are numerous, and many businesses are either experimenting with or have already deployed GenAI in this way.
But which ones are seeing real results, how are they benefiting, and what lessons can we take from their experience?
Verizon: AI-assisted customer service agents lift sales
Telecoms giant Verizon has been progressively ramping up its use of generative AI in a customer service context.
In May 2024, the company announced a suite of new customer- and employee-facing applications including ‘Personal Research Assistant’, a conversational AI assistant that can suggest context-based answers to customers’ enquiries, and ‘Personal Shopper/Problem Solver’, which analyses a customer’s profile to suggest who they might be and why they could be calling.
Since then, the CEO of Verizon’s consumer group, Sampath Sowmyanarayan, has revealed that sales at Verizon rose by 40% thanks to the new technology freeing up customer service agents to sell to customers.
“We are doing reskilling in real time from customer care agents to selling agents,” he told Reuters.
Verizon has implemented the GenAI technology in partnership with Google, and has stated that it enables customer service representatives to comprehensively answer 95% of queries.
The company trained a version of Google’s Gemini large language model on nearly 15,000 internal documents, which form the basis of the responses used by customer service agents, saving them time and effort that would otherwise be spent searching for the right details.
To guide its implementation of generative AI, Verizon convened an AI council, Chief Technology Solutions Officer Debika Bhattacharya told Business Insider last year. It also released a set of AI principles to ensure the responsible use of AI.
According to Bhattacharya, Verizon’s aim is to enable hyper-personalisation through generative AI, “operat[ing] at scale but view[ing] each caller as a segment of one”. Everything the company does with GenAI, she said, is done with the goal of making customer interactions as seamless as possible.
Takeaway:
Verizon’s choice to reskill customer care agents as sales specialists and have them devote more attention to selling is an interesting one, but reflects the priorities that Verizon has for its business while illustrating the extent to which generative AI has freed up customer support agents to do other things.
This is not a use case that will apply to every business, many of whom will likely prefer to double down on the level of service provided. Either way, though, it demonstrates that with automation taking care of more time-consuming legwork, human know-how can be directed to the areas where it’s most valuable.
The most intriguing aspects of Verizon’s use of generative AI are the predictive elements, with AI analysis able to identify the nature and type of a customer inquiry ahead of time. While this is doubtless not foolproof, when it works, it surely helps customer support agents to respond proactively and not just reactively – while saving time that would otherwise be spent establishing context.
ING: Conversational AI ups customer satisfaction and loyalty while maintaining trust
“First nail it, then scale it” is the way that Ayush Mittal, IT Chapter Lead at Dutch multinational bank ING describes the company’s approach to generative AI.
ING used a generative AI chat solution to tackle some of its key problems with customer service, such as high volume – ING receives queries from 85,000 customers per week – and the need to offer support to customers outside of traditional working hours.
The company is no stranger to automated customer service, having experimented with the use of chatbots since 2017; however, these early chatbot experiences were rigid and only allowed customers to select from a range of pre-defined options. Using large language models and natural language understanding (NLU), ING could create a more fluid conversation: as Mittal puts it, “Virtual agents have the potential to answer any question and provide a more natural conversational flow.”
Working with QuantumBlack, the AI arm of McKinsey, ING piloted a new GenAI assistant with 10% of customers in the Netherlands who were using the mobile app’s support chat function.
Within the first seven weeks of use, the chat agent had served 20% more customers than the bank would typically serve with customer support, providing a blueprint that ING could progressively scale across ten markets. Mittal stated that conversational AI also enables personalised, efficient interactions and has improved customer satisfaction and loyalty.
Generative AI is of course not without its shortcomings, such as a propensity to confidently provide wrong information – something that could be highly dangerous for a financial institution and would be disastrous for customer trust. As such, ING has implemented strict guardrails: ING risk stakeholders were brought in from the beginning of the process, real-time monitoring and auditing are in place, and a low-confidence AI response will trigger human intervention.
ING’s conversational agent also cannot give advice on specific topics, such as mortgages and investment products. As Bahadir Yilmaz, Chief Analytics Officer at ING, told Fintech Finance, “Introducing generative AI techniques to a business problem is only five percent of the job.
“Ninety-five percent of the job starts after that. It is important to build systems around AI tools and that takes a lot of effort.”
Takeaway:
Many an executive has lamented the need for placing strict compliance guardrails around the use of generative AI, fearing that it will hamper innovation. If anything, however, ING has leaned into these guardrails, making customer trust in its GenAI implementation the utmost priority.
This also provides peace of mind for company executives, who know that they have already done due diligence and don’t have to worry about GenAI introducing risk elements or creating a sudden reputational crisis.
As McKinsey senior partner Andrea Del Miglio said, “There’s a huge challenge for the banking industry to improve the customer experience, but there’s also a huge amount of risk. You cannot just take technology coming out of a box to do this.
“Leaders must ask themselves: What is the value you are adding to the technology? What is the edge you can add … to make it more helpful for customers and better meet their needs?”
United Airlines: GenAI-enhanced flight stories boost customer satisfaction
United Airlines’ “Every Flight Has a Story” initiative uses SMS messaging and app notifications to provide customers with detailed reasoning for flight delays, leading to an improved experience on the part of the waiting customers.
The airline began this initiative several years ago, but has recently introduced generative AI into the process of creating the messages, enabling greater scale and freeing up staff to solve more challenging issues in lieu of editing templates.
As a result of introducing GenAI into “Every Flight Has a Story”, customer satisfaction has risen by 6%, as United Airlines CIO Jason Birnbaum told CIO.com.
United Airlines’ human “storytellers” still look over the messages to ensure that they’re right for the brand, but Birnbaum said that United is becoming much more comfortable with using generative AI.
“We worked hard to fine-tune this model to take operational feeds, notes from our operations teams, the crew, and all these different sources of data,” he said, “and have AI take all this data and create a narrative that is more transparent, empathetic, decisive, and clear as we can be.”
The flight delay notifications might include details of an incoming aircraft arriving late due to runway construction, or an early heads up about crowded security due to an NBA All-Star game, complete with advice for travellers to arrive early and use the United Airlines app to streamline their experience.
Birnbaum told CIO.com that United has also developed LLMs for use in procurement and manager-employee communications and is beta testing the use of GenAI to improve operations summaries for shift handovers.
The airline’s building and training of generative AI models is facilitated by United Data Hub, a centralised data lake that consolidates United’s data sources and enables real-time access to data.
Takeaway:
United Airlines’ use case here shows how generative AI, when implemented in the right way, can enhance and scale up an initiative that is already effective.
“Every Flight Has a Story” wasn’t conceived for generative AI – but it is being widely expanded as a result of its introduction. In February 2024, Inc. reported that United was only providing the more detailed backstory for 15% of flights, but that it hoped to scale this up to 50% thanks to the introduction of GenAI.
United Airlines began the initiative as a way to add humanity and nuance to delayed flight notices, and the addition of GenAI hasn’t reduced that humanity – thanks in part to the continued oversight of human ‘storytellers’.
The case study from United also shows how work by companies to get their data ‘house’ in order can pay dividends when it comes to implementing generative AI. The work to create United Data Hub has provided a solid data foundation for United’s introduction of both external and internal-facing LLMs, which are now being used for flight status updates, procurement, operations, and more.
