Agentic AI

Agentic AI, at its best, offers a significant improvement over older forms of automated self-service.

Early AI forms “sounded a bit more human-like than a static FAQ page,” Brian Cantor, managing director of CMP, said during a July CMP webinar. But agentic AI’s ability to process and adapt to context makes the experience feel more natural for the caller.

Despite its many promises, agentic AI too often falls short of expectations for businesses and customers alike. To ensure success, businesses need to get their data collection and storage operations in order, according to experts. Leaders also need to understand where agentic AI can replace live support — and where it should serve in a supporting role to human service.

The key to success is to approach agentic AI with the right mindset, according to Cantor.

The technology for creating amazing experiences is available, but a lack of a proper framework and vision is holding leaders back, Cantor said. “If you aren't thinking about it the correct way, that's really a very negative sign for how you're going to care about customers.” 

Data and people are at the core of agentic AI

Getting into the agentic AI mindset is a matter of information. Businesses need to keep customer data collected and organized, while leadership needs a strong understanding of their contact center workforce. 

Every bit of customer data helps when it comes to agentic AI, according to Merlin Bise, CTO at Inbenta, an AI solution provider. Even minor details help the experience add up to a journey that feels a lot more personal.

“Any amount of information you can gain, it's leverage,” Bise said. “So anything a customer will give you, you should be collecting it, whether it's addresses, emails, phone numbers. They need to be safeguarded. But the more information, the more power.” 

Once companies have the data, teams need to store it securely, ensuring compliance with privacy regulations. Relevant data also needs to be properly structured so AI can access it.

“Just connecting OpenAI or Anthropic or Google to a simplistic database within your organization, expecting to have these magical outcomes is never going to be the case,” Bise said. “If you have any desire to do a true agentic AI at some point, and you're going to turn a reasoning model loose on poor data, it's going to behave based on what you give it.”

Successfully deployed agentic AI will replace some live support, but the technology has its limitations, according to Bise. Leaders need to have a good idea of their contact center team’s members and their roles to understand what agentic AI can do to maintain or improve the brand’s customer service performance — and what it can’t.

“You're going to have some stuff you're not going to be able to do,” Bise said. “That's why it's critical to get buy-in from all the people that impact your business.”

Priceline updated Penny, its AI-powered travel assistant, with an agentic capabilities to help customers go from an idea to booking travel, the company said in June. Penny is powered by Anthropic's Claude and runs on Priceline’s proprietary AI stack and data.

Following the update, Penny can take complex questions about travel, evaluate real-time pricing and availability, and provide tradeoffs between options. Customers can purchase flights without having to leave the Priceline platform. 

Penny makes travel planning feel less fragmented, according to Cobus Kok, VP of AI experiences at Priceline. “Instead of opening multiple tabs, changing filters and trying to compare hotel options on their own, travelers can describe what they want in natural language and let Penny help them narrow the choices." 

Online travel companies are racing to go agentic. While some brands, like Airbnb, are focusing on customer support, others are focusing on search discovery. 

Priceline is looking to do it all. Penny can aid with destination planning, searching and selecting flights, hotels and car rental, and customer support for existing reservations, operating over 10 specialized agents as a coordinated system. 

“Many AI tools focus on the inspiration stage,” Kok said in an email to CX Dive. “Penny is designed to go further. It can help a traveler think through the trip, compare choices and move toward a booking in the same experience.”

That matters because making such travel decisions is rarely just about price or star rating, especially for hotels. 

“Travelers are weighing location, flexibility, loyalty, amenities, trip purpose and value,” Kok said. “Penny is designed to understand those signals and make the hotel decision easier.”

When a customer asks a question, the travel assistant provides Penny’s Picks, its top recommendations that weigh the traveler’s stated preferences, overall value, and the context of the conversation thus far. Penny’s Take, currently in beta for hotels, then provides a candid take on a single property and why it may be the best fit. 

Priceline plans on expanding Penny’s Take across the platform over time. That feature may help with choice overload and decision fatigue, which often leads customers to leave a platform without a purchase.

Preference is another important layer. 

“Penny can combine what a traveler has done before with what they tell us matters for this trip,” Kok said. “That helps the experience become more personal and more relevant over time, while giving travelers visibility and control over the data Penny can access.”

Early tests show that customers who engage with Penny show stronger engagement, higher conversion, and reduced customer support contacts, according to Priceline.

“Priceline also estimates that travelers who used Penny saved an average of nearly 10 minutes per trip compared with those who called customer support,” Kok said. “That is a meaningful customer experience benefit, not just an operational one.”

AI is changing the way people work in contact centers, leading to both job cuts and new responsibilities for existing workers. 

It’s also creating challenges for customer service leaders.

“Let's not confuse ourselves,” Neville Letzerich, CMO at Talkdesk, said during a panel at Customer Contact Week Las Vegas in June. “This isn't an operating model change, it's a paradigm shift. If you're saying, ‘I'll just throw AI and humans in there together, it's all going to be fine,’ I think you're looking at it the wrong way.”

Workers feel threatened, according to Letzerich. They are worried that AI will take their jobs, and leaders need to carefully consider how they are handling the introduction and expansion of AI in the call center.

A good strategy includes keeping employees informed about potential cuts or new job opportunities, experts say. Leaders will want to avoid looking at AI in all-or-nothing terms — there may be journeys where starting with AI and ending with human support is optimal.

Every leader needs to think about how their AI plans will affect their workforce regardless of how they plan to move forward, according to Nicole Kyle, managing director and co-founder of CMP Research. Navigating skepticism and keeping frontline morale high are essential for continued success. 

“Executives are pressed for time,” Kyle told CX Dive. “How do they balance transforming the tech roadmap of CX with handling people-related challenges? The best CXOs and the best customer contact leaders know you have to be aware of both.”

Be clear with workers from the start

Whether jobs are being lost or new opportunities are opening up, it’s essential that companies are transparent with workers about how AI will affect their call center operations, according to Jessica Gupta, COO at InfoPay, a data technology company.

InfoPay began informing its employees about its AI plans about two years before the company even knew whether there would be cuts, says Gupta. Despite this, its communication included acknowledgement that jobs may be affected.

“I was very upfront about the reality that I will not pretend that this won't change the head count that we need, that it won't change the way that we work,” Gupta said. “But we'll go through it together, we'll go through it carefully, and we'll be fair. We were very clear-eyed because we wanted people to have time to prepare.”

Infopay took a very methodical approach, according to Gupta. The company not only helped prepare its team for potential cuts, but took the people it knew would remain following the advent of AI and started having conversations about what their new roles might look like. 

For instance, agents who were particularly knowledgeable about certain subjects became part of an analytics project that helped get its AI program up and running, Gupta said. They came on board early in the process as experts on the aspects of the support experience that matter most. As InfoPay added AI features, those employees have continued working on development and testing.

“We did have to release folks — the cost savings aspect is reality — but we've been able to create teams that are now doing exciting things,” Gupta said. “We've given people careers that didn't exist before.”

A major benefit of this approach is that it helps keep morale high as AI becomes a larger part of contact center operations, according to Gupta. Agents know their jobs are changing, but they feel greater ownership over the technology that is causing the shift.

“I think involving them in that process makes it less scary,” Gupta said. “They're gonna be part of that journey with us as we evolve.”

Remember where people are the best fit

Technology is not automatically the answer, according to Bob Sacunas, VP and strategic industry executive at UiPath, an AI solution provider. Humans remain ideal in cases where judgement is necessary.

In some industries it makes sense to start by connecting customers to AI and handing them off to a human not as an alternative, but part of the standard operating protocol, according to Jonathan Rosenberg, CTO at contact center provider Five9.

“Self-service is not a black or white all-or-nothing thing,” Rosenberg said during a panel. “A lot of interesting use cases are actually to have an AI system do initial data collection, triage and then transfer the call purposefully — not because it failed.” 

One example is loan applications, according to Rosenberg. AI can efficiently collect the necessary customer information and then hand the call off to a human loan officer who has the judgement and empathy necessary to make the call less stressful for the customer.

“If you focus on the customer and put them at the forefront, it doesn't mean they don't talk to people anymore,” Rosenberg said. “You can realize cost savings, build great CX and make sure your agents still play a role in what matters.”

Older forms of AI have their application as well, according to Sacunas. Agentic AI is best when some level of interpretation is necessary, but simple deterministic AI is often going to be faster, cheaper and more reliable for simple, repeatable tasks.

“The way you avoid this false promise is really to make sure that you're picking the best tool for the job instead of just over-indexing and trying to apply AI to everything,” Sacunas said.

When it comes to customer service, organizations say agentic AI systems are beginning to deliver significant financial return on investment, according to a report by Deloitte Digital. That’s thanks to improved AI models and agentic systems that can resolve complex cases using natural language and the coordinated use of multiple agents.

Forty-three percent of organizations surveyed believe AI will allow them to lower contact center costs by at least 30% within the next three years.

End-to-end service models spanning contact centers, digital channels, field service and in-product support will soon transform customer service and support, according to Deloitte Digital. More than three-quarters of respondents expect AI to shift the contact center workforce toward higher value roles.

Customer service and support teams are achieving greater ROI as they gain more experience with AI-powered technologies such as chatbots and back-end solutions.

Advances in AI models and lower costs are also contributing to larger ROI. Nearly two-thirds of service leaders reported higher agent productivity, with 39% reporting lower cost per contact, according to Deloitte Digital.

The efficiency of AI is often cited by customer support teams. Just over half say faster response and resolution times is one of the top benefits of AI, according to an Intercom report from earlier this year. 

Other benefits include 24/7 support coverage, scaling support without increasing headcount and lower cost to serve, propelling more businesses to increase their investments in AI in contact center operations. 

But it’s also unclear whether AI will continue to remain cheaper than human support. Gartner predicts AI costs to rise to over $3 per resolution by 2030 — more than many offshore agents.

Mature service capabilities — defined service delivery models, comprehensive personalization capabilities, and low employee attrition — may be key to seeing ROI in AI in customer service. 

Deloitte Digital found that nearly half of the organizations with mature service capabilities are using agentic AI, compared with only one-quarter among their low-maturity peers.

Businesses with the most mature AI deployments tend to experience the most benefits, according to Intercom.

Agentic AI shopping journeys are going mainstream, and companies will now need to consider how any AI platform's access to data could affect the human shopping experience.

This will lead to the creation of a new approach to website design, according to Andrew Laudato, EVP and COO at The Vitamin Shoppe.

“For humans, you want to curate the assortment, curate that content, and just give them the things they need so they can transact,” Laudato said at the National Retail Federation Big Show in January. “On the flip side, AI wants really, really clean data on fulfillment and inventory in real time by location.”

Retailers will want to ensure their data is clear and accurate to have it ready for AI agents. However, this is easier said than done — relevant data can come from any number of sources depending on the kind of information the AI is looking for.

While the entire process is complicated, the first step in preparing data for an AI future is getting it clean and accessible, according to Laudato.

“I’m sorry, you've heard this 100 times, but you’ve got to get your data right,” Laudato said. “The data has to be rich, it's got to be clean, it's got to be in a place where it can be consumed.” 

The information needs to cover every potential element of a transaction as well, according to Laudato. AI agents don’t just need to know what products are in stock, they also need to know about special offers that may apply to orders and what payment systems are available.

The process of identifying data is just the starting point, according to Laudato. Companies will need to consider the formats in which they present the data and which bots are allowed or blocked from their site.

There is no single source of truth

Retailers have a lot of data-based work ahead, whether they are preparing to design their own AI platforms or preparing to work with third parties, according to Nikki Baird, VP of strategy and product at Aptos.

Details from historic prices to physical location are key to any AI application, whether it’s helping a customer place an online order or ensuring store operations run smoothly.

“You need to bring in inventory, you need to bring in the awareness of where that inventory is,” Baird told CX Dive during an interview at the NRF show. “You need to understand promotion price changes and when that promotion expires and the regular price goes back into effect.”

Most companies are working with multiple relevant data sources, and each is best suited for specific queries, according to Baird. Point of sale data is great if an AI needs current pricing information, while sales audit information is better for historical data. 

When a customer’s own data is relevant to their query, the CRM system likely holds the answers.

“You have to have the right context attached to what data source you use at any given point in time,” Baird said. “It's not realistic to say I'm only going to have one version of the truth and that all of the data is going to be here.”

Retailers need to understand the nuances of how their data points fit together, according to Baird. Whether the AI is shopping on behalf of a customer or helping an associate answer a question, the technology needs access to information relevant to the current task. 

‘No one has missed the AI train’

The amount of work ahead may seem herculean, but even the biggest players are just at the beginning of their AI rollouts.

Laudato cited a November Harris Poll survey, which found 40% of shoppers were using AI tools in their holiday gift-buying journeys. However, while the technology is popular for product discovery and information gathering, very few people are letting AI agents complete purchases on their behalf.

This is a sign that consumers are just starting to get comfortable with the technology, according to Jack Hilger, senior director of North America Product at Visa. 

“Nobody has missed the AI train,” Hilger said during the session. “I think we still have a long run ahead of us in terms of getting consumers comfortable with agents, and getting an ecosystem ready to support agents.”