How to Advance Self-Service with the Latest Generative AI Bots

Air Canada’s chatbot made headlines in 2023 for giving a customer incorrect refund information—and the airline was held liable.

A Canadian court found Air Canada liable for negligent misrepresentation after its chatbot gave a customer incorrect information about bereavement fares in 2022. The customer, Jake Moffatt, was mistakenly told by the chatbot that he could book a ticket immediately and apply for a refund for a bereavement fare after his trip had been completed. When Moffatt applied for the retroactive refund, he was told that Air Canada’s bereavement policy does not apply to travel that had already occurred.

He filed a lawsuit against the airline, and the tribunal found that the airline was responsible for all information on its website, regardless of whether it came from a static page or a chatbot.

This landmark decision held significant implications for other companies using AI chatbots for customer service. It set a precedent that businesses can be held liable for inaccurate information provided by their automated systems.

And then there was the British Airways chatbot that gained notoriety when it was unable to recognize London or Heathrow Airport for customers looking to make bookings involving either of those locations.

Beyond the fact that both of these instances involved airlines, they shared another similarity: They both damaged the reputations of chatbots and set back efforts to move customers to self-service.

But today’s chatbots are reversing that, with artificial intelligence-powered chatbots starting to restore consumer confidence and make self-service attractive and effective.

Supported by generative and agentic AI, modern self-service chatbots can provide many more benefits than the self-service technologies of only a few years ago, says Rebecca Miller, a senior CRM product strategy manager at Pegasystems, a provider of AI-powered platforms for enterprise transformation. “A genAI chatbot can do many of the same things that a human agent would do. Examples of that could be answering detailed questions while dynamically outlining multi-step processes and then actually being able to resolve the inquiry. It could be things like managing a first notice of loss or complex billing inquiries.”

Such genAI tools can dynamically guide customers through complex resolutions and adapt conversations based on the information customers provide, Miller explains.

Matt Henderson, vice president of research at conversational AI voice assistant provider PolyAI, agrees that modern genAI tools are better at reasoning and understanding than older technologies.

The Design Dilemma

Modern chatbots offer a lot of benefits (see sidebar on page 21), but to achieve those benefits, Miller says, they need to be designed with the following capabilities:

• Workflow integration.The most powerful self-service systems don’t just answer frequently asked questions; they execute the same processes that human customer service reps would. This requires AI agents that plug into the enterprise core, not just surface-level scripts.
• Semantic understanding across channels.Customers should be able to start on web chat, switch to SMS, or call into an interactive voice response system without ever losing context. AI tools must understand intent semantically and consistently orchestrate workflows across all channels. And when an AI agent can’t solve the issue, it should be able to seamlessly transition to a human CSR without losing context.

And though genAI tools might be able to handle various aspects of self-service, companies need to ensure that those tools are flexible enough to handle new situations, such as the introduction of new payment options, social media channels, or large language models. They need to be quickly adapted to any new environment, says Jacob Murray-White, director of go-to-market strategy for conversational AI at Verint.

Murray-White adds that many companies are rushing to deploy genAI for self-service but neglect to consider that the tools need to be easy to manage and provide a powerful ROI and excellent customer experience while continuing to improve on what they deliver rather than just existing for the sake of having technology in place.

The latest genAI tools improve themselves by evaluating whether the customer received the desired self-service support and then taking steps to make that support stronger, according to Henderson.

Keep Humans Around

The best genAI tools today are able to handle more complex interactions. However, experts agree that today, and even several years into the future, genAI won’t be able to handle every customer need all of the time, so there will be times when human agents will need to intervene, even if only to guide customers or AI agents through the rest of the self-service process.

Miller adds that genAI tools also need to be designed to recognize when the interaction needs a human and route the call or text to a human agent.

“Self-service success isn’t just about deflecting calls or reducing costs,” she states. “Self-service should provide customers with the same level of service that’s typically provided by [human] customer service reps while still offering a clear and trusted path to a human [customer service rep] when needed. Success is measured by resolution and consistency, not just deflection, and if a customer can complete a task, whether that’s updating account information, disputing a charge, or requesting a service, without having to repeat themselves or switch channels.”

Building Self-Service Tools

Getting there even today is not easy. The first step in developing genAI self-service tools is to define the issue that you want the tool to solve, says Cláudio Rodrigues, chief product officer at Omilia, a conversational AI systems provider. “There are many people with generative AI that are just trying to stick technology to the problem. It’s the for-someone-with-a hammer-everything-is-a-nail approach.”

If companies don’t start by defining the issue, they will not just get a bad result, they will get a very, very bad result, Rodrigues says. “Garbage in, garbage out is still a reality.”

For that reason, it’s also important to have “a clear data strategy and a clear organization of the data needed for the problem that you’re trying to solve,” he adds.

And then organizations need tooling that provides clear observability into what’s been done or is attempting to be done, Rodrigues adds. “You need to be permanently tracing and evaluating the models and evaluating the tools that you have chosen to solve the problem” to make sure that they are meeting required key performance indicators or service-level agreements.

Miller also recommends that organizations start looking at their desired self-service processes and then build out to the specific channel.

Switching to a cloud environment is also essential, according to Miller, who notes that if genAI tools are not in the cloud already, they will need to be reviewed to ensure they meet privacy standards and corporate policy guidelines.

“That’s super-important from a risk perspective as well, because if you’re not tying it to your policies and the governance that you have in place, you’re going to have these chatbots not only give kind of unsatisfactory answers, they may actually give detrimental answers,” Miller says. “That can cause havoc for a brand.”

Preventing Hallucinations

Called hallucinations, these detrimental answers can do real damage, as both Air Canada and British Airways learned. But there are steps companies can take to mitigate them.

GenAI bots need to be properly trained and have the right guardrails to provide appropriate self-service answers and to avoid hallucinations, experts say.

Sharon Lorenz, owner of Salesforce consulting firm Platinum Cubed, recommends adding verification steps and clearly defining where AI can provide answers and where it should defer to real experts.

“AI tool training should be rooted in company-specific knowledge, such as FAQs, support documentation, product catalogs, and real interaction transcripts. This process helps to narrow a deployed model’s scope, reducing the chance of hallucinations,” says Steve Zisk, senior product marketing manager at Redpoint Global, a customer data platforms provider. “Guardrails are equally important here: Retrieval-augmented generation (RAG), confidence thresholds, and automated escalation paths ensure that if the bot doesn’t ‘know’ something or which route to take, it either double-checks or seamlessly hands off to a human who can easily take over.”

To further avoid hallucinations, AI must be grounded in trusted enterprise knowledge, including product documentation, account histories, and approved policies, not the open web, adds Baker Johnson, chief business officer of UJET, a contact center platforms provider.

Experts agree that genAI bots will not provide the self-service benefits companies expect without transparency.

Every AI interaction should be auditable, with clear attribution to source content and rules governing which data the AI can access, Miller says. “This is especially important in regulated industries where compliance and trust are non-negotiables.”

“Customers don’t expect AI to be flawless, but they do expect honesty,” Johnson says. “A well-designed system knows when to seamlessly escalate to a human to preserve that trust. The companies that are succeeding with this today don’t view AI as a blunt cost-cutting tool but as a force multiplier that makes life easier and more enjoyable for both their customers and agents. And what that ultimately generates is loyalty, trust, and growth.”

Continuing Evolution

So much progress has already been made in a short amount of time, but Murray-White expects genAI tools for self-service to continue to become more powerful, faster, and more personalized as CRM technology vendors continue to innovate with them.

“We’re already seeing multimodal experiences where I can talk, interrupt myself, change my mind, or go off on tangents,” Murray-White says. “The ability to service those interactions in a near-human way is going to continue developing, as will the ability to interact with other channels. I will be able to talk and send messages and get answers. GenAI will allow us to do these things more naturally than we have been able to in the past.”

Murray-White also expects in the next few years to see superior integration between genAI tools for self-service and back-end systems, which will allow more complex customer interactions to be completed without human intervention.

“The other thing that we’re going to see is the growth of copilot bots, where the agent leverages genAI to go and do more and more things,” he states.

Rodrigues expects to see a much more aggressive rollout of genAI tools for self-service, particularly ones that incorporate voice, with smaller organizations increasingly relying on these tools to provide better, faster self-service capabilities to their customers.

“One pattern that we are starting to see is very large enterprise companies getting generative AI use cases into products that will set the example for smaller enterprises,” Rodrigues says.

“It’s a really exciting time in terms of just how quickly things are moving in this space,” Miller says. “Even a year from now, the contact center is going to look a lot different than it does today. I don’t think it’s going to be completely shifted on its head by next year, but contact centers are definitely going to be much smaller, more specialized operations.”

Miller further says that the percentage of interactions that can be completed entirely through genAI bots will double in the next year and could easily exceed 80 percent by the end of the decade.

But to get there, companies will need one or more large language models, an orchestration platform, channels to deploy the tools, document resources, easy integration into their CRM and enterprise resource planning systems, and the ability to extract data to make changes to benefit the customer. Making those changes will continue the evolution of self-service tools to make them more efficient and customer-friendly, and therefore more likely to be used by customers, Murray-White says. 

Phillip Britt is a freelance writer based in the Chicago area. He can be reached at spenterprises1@comcast.net.

The Benefits of Self-Service

eGain notes the following business benefits from self-service chatbots:

• Cost reduction.A single self-service interaction can cost as little as $0.10 compared to $8 for a live agent interaction.
• Increased efficiency.Customer service teams can focus on complex issues while routine queries are handled through self-service support channels, improving overall operational efficiency and resource allocation.
• Scalability.The right technology can handle unlimited concurrent users, allowing businesses to scale their customer service capabilities without proportional increases in costs.
• Data collection.Self-service platforms provide valuable insights into customer behavior, preferences, and common issues, enabling businesses to improve their products and services.