From Taco Bell to DeepSeek: 4 ways AI is disrupting CX

We’ve all heard of the latest chatbot innovations and decision engines that can push products and services according to a user’s needs. But with generative and agentic AI now in the mix, there are many more ways artificial intelligence (AI) is changing CX.

“The big thing we’re seeing right now is the rise of agentic AI that can solve issues that require some form of action, some form of business logic and some ability to follow up and reason,” says Deon Nicholas, founder, president and executive chairman of Forethought.

“Then what I think will happen, is because AI is getting so powerful, you will start to see knock-on effects, which will facilitate other use cases, like agent assist. Human agents are going to become more critical to the more complicated issues and they will become your product experts,” Nicholas adds.

As a result of these developments, Nicholas says it is likely the role of human agents will be elevated to drive revenue in their new roles as product experts.

Four brands with ground-breaking AI use cases

Before that happens, generative AI, automation and AI-powered search and personalization are already making waves. Here are four major disruptions that have made headlines in recent months.

Swiss International Airlines: Wanderlust bots

Swiss International Airlines now offers a generative AI-powered chatbot that can help customers discover new travel opportunities by asking about their specific holiday preferences and bringing them fresh ideas.

“We want to inspire people when they visit our website to look for trips,” says Edward Pauls, the airline’s digital adoption and innovation expert.

“They know they want to travel but perhaps don’t know where to. With Heidi, they can tell us that they are a family of three with a dog, and they’re looking for destinations with a beach,” Pauls adds.

Powered by Google's LLM and other services, Heidi invites travelers to explain what they are looking for in a vacation, “the same way you would tell a friend, and it will suggest destinations, available flights and even a complete itinerary”.

Descriptions can be as simple or as complete as the prospective passenger wants and can then be modified as needed.

Find out more about Heidi during All Access: The AI revolution in CX.

Taco Bell: The AI drive-thru

As demonstrated by Wendy’s and Taco Bell, artificial intelligence (AI) is changing how the quick service restaurant (QSR) sector delivers experience, by automating tasks that don’t always require human oversight and allowing humans to deliver a more personal touch elsewhere in the customer journey.

“At many Taco Bell drive-thrus customers now tell their order to the AI and it populates the order that goes to the kitchen. The capabilities are very advanced now,” says Steve Hsu university professor and founder of AI startup, SuperFocus.

The AI developments follow the introduction of the “Defy” concept in 2022. For anybody who has watched The Founder, the concept of refining processes to drive excellence in QSR – and reduce order times – will be familiar. AI is simply the modern-day answer to the automated burger making machines featured in the film.

Hsu has invested in multiple AI start-ups and through SuperFocus and is now working closely with the restaurant industry to automate the first step in a diner’s journey.

“It could be a fast-food restaurant, a fancier restaurant or one that is available for delivery. On our platform we can take the menu from the restaurant and upload it into AI, then we give it some basic rules of what substitutions are allowed, and so on. At that point, it can take orders and it will populate the order data fields in whatever format you want,” Hsu explains.

TotalJobs: Finding a new role with generative AI search

UK based jobs search engine Totaljobs doesn’t just list thousands of open roles; it also offers job hunters a Job Search Companion.

The generative AI-powered search engine is designed to support job hunters through one of the most emotional user journeys that exists, using conversational prompts.

“A job search is a life-changing event,” says head of product Somnath Biswas, who pioneered the bot. “There is a lot of stress, you will apply for a hundred jobs and 99 times you will be rejected. The reason why we wanted to go down the conversational route is because we wanted to recognize and account for that emotion.

“From a CX perspective, if you can bring in this element of emotion, that makes the experience a lot richer,” Biswas adds.

It also helps to extract relevant information from job hunters that can make their journey easier. Instead of being presented with a long form to complete in order for the site to match a user to a job, the user instead has a conversation about their needs and through the power of generative AI search, can be matched with a role that suits their preferences.

“In 2016, if you had a lead generation form, the conversion ratio was around 1.5 to 1.6 percent in terms of people filling up a form. Whereas, if you have a chatbot on the other side and you are having a conversation, that conversation ratio increased to as much as 13 to 15 percent, because it's an incremental experience,” Biswas explains.

Moonpig: AI-generated personalized fonts

In late 2024, the UK’s largest online greeting card retailer Moonpig, unveiled a tool that allows users to create a “unique” text font, based on their own handwriting. The font is then stored to the user’s account so they can use it to write the message in every card they create on the platform.

Personalization is Moonpig’s USP, but allowing users to “hand” write cards remained the last frontier to a truly personalized end product. A year in the making, the proprietary AI can also be used in conjunction with Moonpig’s ChatGPT integration, which can refine or draft the message from scratch.

Overcoming AI’s top challenges

As these case studies outline, there are many ways in which AI is improving the processes that create a customer’s experience, but AI still poses many challenges to the modern organization.

Cutting through the noise

According to Nicholas, the first barrier to AI investment is the noise around AI. “There is a lot of AI out there. There always has been. And most of it is more artificial than intelligent,” he says, referencing how hard coded decision tree chatbots were once referred to as “AI” in the same way that some of today’s most sophisticated tools are also called “AI”.

“Although we were all using the same word, not all AI is created equal. That means getting informed on what's out there is the first barrier,” he adds.

Hsu agrees: “If you run a call center, but you're not an AI expert, you're being inundated by people trying to sell you stuff and you have no idea what's really going to work or what's not going to work.”

To ensure AI investments generate a return, Hsu and Nicholas say the first step is to conduct rigorous testing – preferably before any money has changed hands and also in a real-world scenario – to ensure the tool operates as advertised.

Hsu says: “If a vendor says they are working with Taco Bell, ask to go and see how well their technology works at Taco Bell. You should do very thorough due diligence and not simply assume what you are being told is true. You need to be very careful to make sure the technology has capabilities that are claimed.”

Achieving returns on AI investments

Such due diligence is the first step to ensuring an AI investment will generate a return. However, Nicholas highlights that investment is often needed before it’s possible to test AI in real-world situations.

“A lot of support and customer experience leaders are forced to invest up front to even be able to get to the point where they can test a technology out. This is why I think a lot of AI spend today is experimental. It's people literally trying to figure out what's going to work,” Nicholas says.

Some vendors now offer more flexible pricing models, offering AI-as-a-service or, in the case of Superfocus, as a share of the overall cost saving. “We have a very low upfront cost and no integration fee,” Hsu says of the low-risk approach. “In other words, for every hour of human labor that we save you, we ask for a fraction of the savings. This means we can save a company a huge amount of money because they’re only paying us if they’re saving money.”

Other companies take the subscription approach, allowing buyers to de-risk their investment by paying monthly, instead of upfront. However, Hsu says “a lot of people fail to calculate the extra considerations, like integration and training costs”.

Keeping security and access top of mind

LLMs trained on public data have driven many major innovations, but to understand a specific business they need to be adapted – and the only way to do that is to give the model access to sensitive, or previously secure, data.

“You need the AI to be hooked into the system so it can take the agentic actions, but that then creates a trade-off and in order to realize the full power of the technology, you need to give it access,” Nicholas says.

The upshot is that, in this process, security is critical. To combat the obvious issues, Forethought’s technology can automatically redact personally identifiable information (PII) before it’s entered into an LLM.

“You have to treat your AI as if it was a very, very new employee. You wouldn't give them access to all the systems,” Nicholas says, advising organizations should pursue “zero-knowledge-based cryptography for AI”. He explains: “You want systems where you literally have to put in the correct answer before you can even access the system, even as an agent.”

A system with access to all a customer’s PII could technically then tell that customer their secret phrase or even pin number, because without the PII safeguards offered by Forethought, the system that is verifying the caller’s identity is the same system that can access all that caller’s information.

“You never want to be training on proprietary data in a way that the model could, in theory, hallucinate the right answer,” Nicholas explains. “The beauty is that if you encode that into AI, you actually get an even more secure system than if a human were in charge because the AI literally, mathematically, cannot access the information without following the right protocols,” Nicholas adds.

How DeepSeek could change the status quo

One of the major benefits of AI is that in using it to streamline organizational processes and human contact, significant cost reductions can be achieved. The irony is that with DeepSeek now available it is possible to replace existing AI with even leaner models, which save money by using less computing power and therefore less electricity.

“Because DeepSeek is open source, you can run it for free and just pay for the compute,” Hsu explains. “That means if you can swap out the existing AI models for DeepSeek open source technology the savings could be in the region of 10x.” Furthermore, because the compute demand is lower, it is even possible to run these systems in an en premise data center.

“A lot of the times the drive to use AI comes from the CFO in a company, who then instructs the contact center leader to reduce costs by using AI-powered technology,” Hsu adds.

“The contact center leader isn’t an AI specialist, so they find themselves talking to vendors who promise their systems can do anything. However, the reality is that’s a big investment and often they also have legacy systems, which make things more complex,” he continues.

As CX Network has previously covered, the arrival of DeepSeek could drastically change the AI vendor landscape.

On the specifics, Hsu says: “I think every company deploying AI, whether for customer support or another application, should be testing the open source models. They are much more efficient, and the security situation is actually better with open source models, because the code is running in your own data center. There is nobody else in the loop.”

Nicholas says that at the application layer, DeepSeek is unlikely to change much. However, in competing with OpenAI, Anthropic, Gemini and others, “what we're seeing, I think, is actually a commoditization of the model layer”.

“All these LLMs are roughly going to start to perform on par with each other. There are some differences, for example, in the bias that has been reported in DeepSeek. But overall, it's probably going to democratize access and there are going to be many more models and more opportunities for smaller companies to get access to this stuff,” he explains.

OpenAI has already started to respond to the new landscape. In early March it confirmed the release of a suite of new AI agents: a “high-income knowledge worker” agent reportedly priced at US$2,000 a month; a software developer agent reported to cost $10,000 a month; and the most expensive agent, priced at top-level $20,000-per-month, aimed at supporting “PhD-level research”.

The technology is aimed at everybody from high income professionals to enterprise clients alike that can afford to pay a premium. But as with everything in AI, this could all have changed by next week.

To help you keep ahead on the latest AI capabilities, CX Network is hosting the free to watch All Access: AI Revolution in CX series, March 25-26.

Quick links

[inlinead-1]