Generative AI

AI has been deployed for years in customer experience settings, but it was an expensive, time-consuming exercise to train the models and required ongoing maintenance to analyze sentiment and other signals, according to Jonathan Rosenberg, CTO and head of AI at Five9.

“Then generative AI came along and changed all of this,” Rosenberg said.

Many business leaders expect the technology to transform operations in the coming years, improving efficiency and productivity and reducing costs, according to a 2024 Deloitte Digital report. Customer service is one of the top use cases for generative AI, helping agents rapidly build rapport and elevate experiences and acting as a force multiplier.

“It’s possible to analyze every conversation in the contact center by feeding the content of the conversation to a generative AI model and asking it to assess sentiment, customer satisfaction, issue resolution and many other metrics,” Rosenberg said.

But it doesn’t stop there: The technology can then aggregate all this information, providing the business with a bird’s eye view of what’s happening in the contact center.

“We can now automatically detect new topics of conversation in the contact center and alert managers to these topics so they can proactively take action,” Rosenberg said.

Alaska Airlines has deployed Five9’s Intelligent CX Platform for automating calls routing based on channel volume, agent skill set, and availability. The airline was able to lower wait times, even during call spikes.

“Real-time call transcriptions let agents read what a guest is saying in real time, improving their comprehension and engagement,” Rosenberg said.

Aeroflow Health, a medical equipment and supply provider, combines multiple AI tools to streamline customer inquiries. 

This approach has increased self-service options, decreased average handle times and reduced call volume, while AI-generated summaries enable Aeroflow to test, monitor, and improve its models to optimize the output of its call summaries.

From chatbots to agent assistants

Human intervention remains essential, and generative AI can enhance the traditional 360-degree customer view, prioritizing key insights to guide agent interactions.

In addition to outward-facing tools such as chatbots, internal assistants can support customer service teams and show the potential of the force multiplier effect.

“Conversational AI, especially in live chat and messaging, has matured significantly, but with generative AI, it’s shifted inward, focusing on augmenting employees who interact with customers,” said Jason Valdina, senior director of go-to-market strategy for digital engagement channels at Verint.

When it comes to customer support, these tools excel at summarizing interactions and extracting useful data points, such as language and caller history.

“Summaries help agents quickly grasp the context, reducing average handling times by approximately 50 seconds per interaction,” Valdina said.

By combining tools, businesses can realize more significant results. For example, Volaris airlines, a major Latin American carrier, deployed knowledge automation and coaching tools, freeing up agent capacity and redirecting agents’ attention.

“Agents, with newfound bandwidth, could upsell services, such as discounted seat upgrades, and they were able to offset the entire cost of their contact center,” Valdina said.

Sentiment analysis is another area of customer support that can be improved by applying generative AI tools. A coaching tool that operates as a copilot alongside an agent’s workflow can intervene with prompts that provide guidance, from improving communication etiquette to ensuring accurate responses and gauge customer and employee sentiment in real time.

“One healthcare provider saw a 16% increase in net promoter score after implementing a scoring tool that monitored and intervened in conversations with declining sentiment,” Valdina said.

Other tools can create summaries of ongoing conversations when transferring interactions between agents and at the end of an interaction to make a record of the call, saving on average 30 seconds and 60 seconds, respectively, according to Valdina.

However, consumers are concerned about not being able to speak to a human. While generative AI tools can help save time, they are not a replacement for human touch.

“The emphasis is not on replacing human agents, but on enhancing their efficiency, enabling them to handle more tasks effectively,” Valdina said.

 

Walmart is accelerating its Adaptive Retail strategy, aiming to combine generative AI, augmented reality and personalization to create immersive experiences in Walmart and Sam’s Club stores, apps and websites, the company said in October.

The retailer has developed a series of LLMs called Wallaby and plans to use the technology to support customer-facing assistants and experiences. Walmart has trained Wallaby on decades of internal data, which lets it work with other LLMs to create responses that are highly tailored to the Walmart environment.

Walmart also created an AI-powered Content Decision Platform that tries to predict what customers want to see on its website to create unique homepages for every customer based on their interests. The updated website is expected to launch in the United States by the end of 2025.

Walmart is continuing to invest in AI and personalization technology aimed at creating experiences that help the retailer stand out from the competition.

The retailer has been rolling out generative AI tools throughout the year, with an emphasis on improving search and discovery. Walmart added enhanced search to its app in January, and began testing an AI shopping assistant with select customers in June.

“A standard search bar is no longer the fastest path to purchase, rather we must use technology to adapt to customers’ individual preferences and needs,” Suresh Kumar, global chief technology officer and chief development officer at Walmart, said in a prepared statement.

The average customer spends 6 hours searching and browsing online every week, Anshu Bhardwaj, SVP and COO at Walmart Global Technology and Walmart Commerce Technologies, told CX Dive at National Retail Federation’s conference in January 2024. The generative AI updates aim to cut this time by letting customers use natural language queries and receive more accurate responses in return.

Walmart isn’t the only one updating its search with AI. Google rolled out AI-powered search updates in October. They include improvements to the Google app’s Lens feature, which lets customers take photos of items and then search for similar products, and curated results pages for recipe and meal inspiration searches.

Walmart’s tech investments are aimed at enhancing in-store experiences as well. The company used generative AI to create or improve the quality of more than 850 million pieces of data across its product catalog. This information connects to store and warehouse associates’ mobile devices, helping them find products in the backroom or put together deliveries faster. 

Walmart credited associate-facing tools for a successful holiday season in 2023.

Walmart is expanding shopping experiences to new channels too. The company developed an augmented reality platform called Retina, which will let customers shop in virtual social environments, though Walmart has yet to provide specifics on the experience or a launch date.

In a bid to improve efficiency and quality in its customer support, streaming services company Brightcove became an early adopter of Amazon Q, a generative AI–powered assistant for its in-house “Expert Bot.”

When approached by Amazon, Brightcove executives immediately saw the potential to better handle technical queries from its diverse range of global clients that include the SXSW conference, Consumer Technology Association, Cancer Treatment Centers of America and the Academy Awards.

Brightcove support agents field thousands of queries, from simple instructions to reset a password to more complex help like embedding video in a WordPress site that can require significant time.

“In the past, answering this question would require a support agent to manually search through documentation and provide a detailed response,” said Scott Levine, chief product officer at Brightcove.

With the new tool, the company says it is saving time and improving its already highly regarded customer service.

“The starting point for us was ‘Can we make it easier for us to answer every single one of those questions and make ourselves do better?’” Levine said.

Delivering value in the rush to adopt generative AI tools

The Brightcove example reflects a wider trend of organizations adopting generative AI across a range of business functions, showing they can deliver real business value, a 2024 McKinsey survey found.

To gain a genuine competitive advantage, however, businesses should focus on helping people do their current jobs better and lift productivity, according to the research firm.

If generative AI tools improve the way employees provide customer service and support, it can have a real impact on the business, according to some experts.

“How you service and support customers, whether your business offers a product or an interaction, is becoming more of a differentiator,” said Melissa Copeland, founder and principal of Blue Orbit Consulting, which specializes in customer experience design and technology adoption.

The challenge is applying the technology where it delivers a dividend, not just for automation or cost cutting.

“Organizations must find the opportunity to use technology to reduce friction for customers and reduce friction for agents,” Copeland said.

Copeland cites examples of businesses that have seen a minimum 25% to 35% return on investment in the form of decreased handling time or first call resolution when deploying generative AI customer support tools. The gains could expand beyond, too.

“Then there’s thinking about new metrics beyond the traditional ways. That’s what’s really exciting about the possibilities of these tools,” she said.

Build generative AI tools with guardrails to foster trust

To deliver on the promises of generative AI, organizations must ensure the accuracy of the model to build trust, according to McKinsey.

In Brightcove’s case, it took a cautious approach, setting well-defined guardrails around the use of customer information that require permissions, rigorously testing the system, and having in-house technical experts scrutinizing the output.

“It's not one person looking at it; it's 50 people looking at it,” Levine said.

The Amazon team worked closely with Brightcove staff, teaching them how to craft questions to get the most suitable answers and taking time to train the bot on its extensive library of documentation as the source for responses.

“Streaming is complex, and that creates a lot of information about how to do it the best way across thousands of customers,” Levine said.

The bot-generated answers also include citations that link to the Brightcove source information so if the answer doesn’t look right, human experts can verify the response and feed that back to the tool. It adds an important layer of trust and teaches the model to help improve its output as it processes more queries.

“We have a lot of very technical people here at Brightcove, and when we see answers that aren’t right, we call it on it,” he said.

The end result is that the bot provides customer and product support teams with responses to customer questions. Brightcove has found it’s not only reduced the response time but also improved the overall efficiency of the support team.

“It’s about giving our customers the information they need quickly and accurately, and continuously improving that process,” he said.

Avoiding the trap of one-off tools

To maximize the potential of generative AI, McKinsey recommends that organizations avoid one-off solutions and invest the time and resources into tools that will scale.

Copeland cautioned that organizations must think through the longevity of the platform they’ll be working with.

“Use as your base platform a known entity, such as a large outfit that already offers customer experience tools you’re using across the business,” Copeland said.

While organizations might want to experiment, they also need to avoid different internal groups building their own tools independently to ensure widest application of new tools.

“If you get one-off customizations, it’s difficult to pull together, and when things are done in bits and pieces without any kind of central coordination, it becomes a nightmare to centralize and scale,” Copeland said.

In the case of Brightcove, Levine sees ways to expand the company’s use of generative AI to address more complex tasks such as enhancing metadata management on video streams or automating translation services to convert a video into multiple languages.

“Our customers have footprints all around the world, and if we can enable them to translate content on the fly, that's another use case for generative AI,” Levine said.

Above all, Brightcove’s use of Amazon Q is meant to not only improving operational efficiency but also add value to its service for customers.

“Everything we do starts with that mindset and driving business value,” he said.

Generative AI will result in fewer agents taking on more complicated work, according to a Forrester report released in June. Members of these smaller teams will need a combination of emotional intelligence and technological know-how, the report found.

Next-generation call center work will require leaders to rethink their hiring and retention strategies, according to Christina McAllister, senior analyst at Forrester. The call center of the future will need agents with relevant training and advanced skill sets, making individual workers more valuable.

"I believe that it's no longer going to be a race to the bottom on how cheap you can get your labor," McAllister said. "There's going to need to be a focus on retention."

It’s still the early days of the generative AI revolution, but it’s as good a time as any for call center leaders to future-proof their workforce and technology strategies.

One way to prepare agents for their changing roles is investing in their artificial intelligence quotient, or AIQ, according to McAllister. Having a human in the loop can be valuable, particularly during early rollouts, but few workers understand what that really entails.

“You need to educate folks on what it means to be the human in that loop and how best to work alongside AI,” McAllister said. “When do you trust the output of an AI solution? When do you question the output of an AI solution?”

Contact centers need to ensure their knowledge management systems are up to the task of supporting these agents as well, according to McAllister.

Generative AI is only as good as the data that powers it, and the technology can’t fix a poorly maintained or outdated knowledge base. A trustworthy copilot that helps agents deliver better experiences needs clean, trustworthy data it can draw from.

However, in the past, many call centers underinvested in their knowledge management solutions, according to McAllister. Generative AI projects often receive bigger budgets, and modernizing the data backbone is a worthwhile use for this funding.

"Folks are coming face-to-face with the need for data,” McAllister said. “It’s one of the investments they need if they pursue generative AI projects. If you don't have the data from which you can generate that information or knowledge, what are you doing?”

Generative AI was introduced to technology leaders as a transformative addition to their company’s toolbelt. Yet organizations are still struggling to reap the technology's full benefits and accurately assess its impact.

Organizations are predicted to abandon nearly one-third of generative AI projects after the proof of concept stage by the end of next year, according to Gartner research, which blamed rising costs and unclear business value as contributing culprits. 

Tuning, customizing or simply deploying generative AI models costs businesses an average of at least $5 million. Naturally, companies are expecting a return on their investment.

“We’re a pretty well-funded IT department, so we have access to the resources and tools we need,” Brent Harris, CIO at Tulsa, Oklahoma-based community mental health center Family & Children’s Services, said. “But we’re also nonprofit, and we try to be responsible with the money we’re given.”

The organization is assessing generative AI’s ability to improve workflows, provide better data-driven decisions and reduce repetitive tasks by creating cheat sheets for frequently-asked-questions and step-by-step guides. 

Tech leaders want to ensure generative AI use cases are set up to succeed. To avoid pouring money into failed pilots, executives are weighing implementation costs against the potential value and risks the technology could bring.

“We certainly imagine that list [of use cases] is going to grow,” Harris said. “We’ve tried to be mindful of being on the leading edge, but not the bleeding edge. We don’t want to be the guinea pigs, but we don’t want to just keep up.” 

Where generative AI pays off

Businesses land on a wide spectrum when it comes to AI readiness. This can impact their ability to determine ROI.

Poor data quality and management stifle aspirations. Caution around cybersecurity and protecting sensitive information presents additional hurdles. Plus, generative AI tools can open the business up to risks related to the tech’s inaccuracy and bias. 

Despite potential risks, one of the leading adoption barriers for enterprises is an inability to estimate and demonstrate the value of AI projects, according to a Gartner survey published in May. Less than half of AI projects ever make it to the production stage. 

“A lot of companies plunged in with hype, converting everybody and any use case that came up, they tried it,” Shohreh Abedi, EVP and chief operations and technology officer and membership experience at AAA - The Auto Club Group, told CIO Dive. “We didn’t do that. I set up a structure with a methodology and a process and put guardrails around it.”

Businesses that rush into generative AI adoption have reported buyer’s remorse. Around 1 in 4 IT leaders said they regretted investing in AI so quickly, according to a March Asana survey. 

Still, tech leaders are trying to keep pace with broader business goals, priorities and responsibilities. 

At Family and Children's Services, Harris worked to find a balance between innovation and performing the “basic blocking and tackling,” as finding staff, therapists, clinicians and other experts has become more difficult. 

“The more time we can spend automating what they do and making documentation easier for them it helps with retention,” Harris said. “If, relative to our peers, it’s much more difficult to use our systems, we’ll have a hard time bringing on and retaining quality staff, so there’s a balance but nothing is done at the risk of security.”

Eager adopters have found value in taking a measured approach to piloting the technology, explicitly aligning use cases with business goals and success metrics.  

AAA’s second-largest club focused its own generative AI push on content management, internal support services and coding tests as it laid out its strategy last year. The organization has a dedicated team sifting through use case proposals, led by Abedi alongside the group’s chief architect and CISO.

“I give the final approval on what use cases we’re going to look at, and they have to meet certain criteria,” Abedi said. “The use cases have to meet a business goal, either growth, cost containment or efficiency from a productivity increase.”

When a use case meets the standards, and only then, does the organization move forward with a proof of concept.

Speed is an important element of self-service, but it’s not the only thing that matters. 

Balancing speed and quality is particularly important for developing trust in generative AI. Customers are open to chatbots, but they’re not willing to risk traditional service for a theoretically faster experience, experts said at an August webinar hosted by CCW Digital and Zendesk.

“If they feel as if they have to trade some semblance of empathy, if they have to trade a better resolution, if they have to trade relevance of what they're getting just to save some time, it's going to feel like a choice they don't want to make,” Brian Cantor, managing director and principal analyst at CCW Digital, said during the presentation.

Generative AI chatbots aren’t suited for offering the human touch, but they can excel at offering a personal touch. The right approach to self-service, backed by the right data, can make customers feel like chatbots understand their needs.

Personalization is a significant pain point for modern contact center operations — only 20% of consumers say they feel the typical customer service interaction is highly personalized, according to Cantor.

Simply showing off what you know about the customer isn’t proper personalization, Cantor said. Companies need to tap into that knowledge to change a customer’s experience, and AI is well-suited for taking actionable consumer data and automating interactions based on that information.

No chatbot is going to ask about a customer’s pets or vacation plans like a good customer service agent might, but the technology can look at prior interactions to inform possible solutions. A well-informed chatbot doesn’t make customers repeat information and can tailor its responses based on past interactions.

However, creating a personalized experience requires careful data management, according to Jason Maynard, chief technology officer at Zendesk. 

Businesses need to ensure all relevant data is properly integrated into the tech stack before it will be useful for powering chatbots. This includes ensuring customer metadata, the answers to common customer inquiries, and knowledge about the company’s products and services are readily accessible to chatbots.

“Make sure your content is complete, robust and kept up to date,” Manyard said. “Those are the big areas that I think are really going to be an enabler for this technology.”