Feature
The Gist
- Human work in an AI world. In a world where AI is becoming more and more capable, how do human and AI work compare?
- Customer interaction. AI chatbots enhance customer service by handling multilingual inquiries, providing personalized recommendations and automating processes.
- Human collaboration. Augmented intelligence emphasizes collaboration between AI tools and human users, rather than AI operating independently.
- Marketing shift. By 2025, Gartner predicts that 30% of outbound marketing messages from large organizations will be synthetically generated using AI.
Editor's note: This article is Part 2 in a four-part series where we explore the reporting and article-writing differences between artificial intelligence and humans on the same topic (AI-augmented customer experience). The first part features an article where CMSWire reporter Michelle Hawley asked ChatGPT-4.0 to write an article based on outline prompts. Today, our series continues with our very own human writer, CMSWire staff reporter Jennifer Torres, who did not use artificial intelligence. More about this series at the conclusion of the article.
From customer service chatbots that respond to customer questions and handle multilingual inquiries, to tech that provides hyper-personalized recommendations, automates processes and reduces fraud, AI-augmented tools save human workers precious time by streamlining the shopping experience and handling the “grunt work.” They can also enhance it with instantaneous granular insights into the wants and needs of individual consumers.
While artificial intelligence can operate without humans, augmented intelligence is the collaboration between AI tools and the people who use them.
Simon James, managing director of data science and AI at Publicis Sapient, a global digital transformation consulting company, has seen AI broadly adopted in customer service, both as customer-facing chat style applications and behind the scenes-based assistants for contact center agents.
“Traditionally, the primary role has been front line support, reducing the volume of inbound requests to human operators,” James said. “These requests tend towards FAQ style queries; common questions that have clear answers. Where discretion or judgement is required, those interactions are generally dealt with by humans.”
According to James, earlier generations of AI solutions were often more robotic in nature, following strict logic that created a certain level of frustration among customers who struggled to ask a question in a way the bot could answer. But with the explosion of generative AI, conversational UIs can now better understand the intent of the user’s question and interpret them to formulate better responses.”
Gartner predicts that 30% of outbound marketing messages from large organizations will be synthetically generated by 2025.
Understanding AI-Augmented CX & How It Works in Customer Service
Peter Dorrington, chief strategy officer and founder of XMplify Consulting Ltd., said the use of AI in Robotic Desktop Automation (RDA) is already widespread and appears in a number of guises.
“It is particularly useful in guiding an employee to relevant content and tools to help them do their work, finding a form, partially completing a task, and so on. It has the virtue that it can ‘institutionalize’ and propagate learning within an organization,” Dorrington said. “For example, it can help an employee find an answer to a new question and once a correct answer has been found, the AI will remember this and offer it to the next employee who has the same question. The whole organization ‘learns’ the answer without it having to be broadcast. Similarly, it can guide an employee through complex business processes, offering decision support at critical points, along with pros-and-cons.”
Further, he said, AI can also use deep learning to find hidden relationships, patterns and trends in very large data sets — far more than an unaided human could ever hope to process. And it can do this not only at speed, but at scale and across time, reviewing data from multiple channels and sources.
AI tools can then extrapolate these analyses into predictions about the future and uncover what is likely to happen next from a given set of assumptions.
Related Article: Getting Full Benefits of AI-Human Collaboration in Customer Service
Learning Faster With AI: How AI Can Help Customer Service Teams Adapt More Quickly
Because the vast majority (some say 60-80%) of data is unstructured data, James says companies generally analyzed very little of it due to cost and time — but that’s all changed with AI.
“Analyzing and summarizing this data is a core use case application for generative AI, and through the right adoption, allows businesses to more easily identify patterns in customer service data in order to inform product strategy or customer experience improvements,” James said. “For example, understanding what the most common questions are in customer service data and where the user was in their journey or experience when they began a customer service interaction, enables businesses to automatically extract FAQs and publish that as content in the most appropriate steps of that customer journey to reduce future demand.”
In talking to a wide range of organizations recently, Dorrington discovered that between 60-80% of them are experimenting with generative AI such as ChatGPT, Bard, etc.
“For example, using AI to find specific documents from large repositories, provide links to them in their responses, and even summarize key points from the portfolio,” Dorrington said. “This is invaluable in organizations that have large amounts of unstructured data and need something more sophisticated than a search engine to exploit them. We have already seen public versions of this with AI-enhanced internet search.”
Implementing AI-Augmented CX: Best Practices
For James, the first focus should be mitigating the risks of incorrect AI-generated answers to customers’ questions.
“Generative AI is well suited for customer service, but the responses need to be factual and if there is uncertainty, that should be communicated to the customer,” James said. “A good data scientist can tune a generative AI model to be more factual and allow for certainty.”
Next, he said, ensure colleagues are involved in every stage of development including in live deployment.
“Do not underestimate the amount of time required for testing and avoid the temptation to take the human out of the loop,” James said. “AI works best when enhancing domain experts not replacing them.”
Finally, you need to determine whether to build or buy, and either way, you need to be comfortable with the roadmap for the solution.
“In such a fast-moving field, you do not want to be locked into a vendor who might not emerge as a leader down the line,” James said. “Equally, if you are building your own in-house solution, you need to ensure you have a clear vision for how the AI customer service application will evolve over time. How will it deal with new data? How will you optimize performance? This is not a one-and-done implementation but an ongoing service that is being built.”
Dorrington’s advice is to remember that AI is harder than it looks to manage, especially in the early stages of "training the AI." It needs a lot of good data to learn from, and the results have to be carefully evaluated.
“Some AIs have worked very well in controlled conditions only to fail miserably in the real world,” Dorrington said. “For example, it can detect ‘spurious correlations’ (where two unrelated things are incorrectly perceived to be linked) and these spurious correlations are treated like actual correlations ... The spurious correlation works until it doesn’t, and the result can be ugly.”
Further, he said, implementing AI does not necessarily mean lower costs.
Webinar
Sep
25
CMS Briefing: A Live Look at What’s Next in AI-Driven Platforms
Learn how leading organizations are using AI‑driven tools to publish faster, personalize smarter and stay secure.
Register
Webinar
Oct
15
Agentic AI Playbook Real-World Customer Service Use Cases You Can Deploy Now
Boost self-service by 30% and slash call volume by 63% with agentic AI.
Register
Webinar
On demand
Ready or Not: How Data-First Organizations Are Unlocking Agentforce Potential
Learn how to cut through the noise, activate Agentforce and build a Salesforce AI strategy that actually delivers.
Watch Now
Webinar
On demand
AI in Customer Service: Faster Resolutions, Happier Customers
Don’t let rising demand burn out your team. See how to build a smarter, more resilient support org.
Watch Now
Webinar
On demand
From Hype to High-Impact CX Strategies That Actually Scale
Turn buzzworthy AI and outsourcing trends into measurable CX wins with fresh 2025 data.
Watch Now
Webinar
On demand
Insights to Action Rethinking the Contact Center for Real Business Impact
Join our exclusive webinar to hear CX executives share their innovative strategies for transforming service delivery.
Watch Now
“Some organizations have found that the headcount has remained the same because workers have adapted to spend the time previously devoted to completing low-value tasks to ones that are more suited to humans and also more valuable,” Dorrington said. “In a few cases, the headcount has reduced but the wage bill has gone up because the remaining workers are much more valuable, including to the competition, and pay and benefits had to increase to reflect this.”
Finally, he adds, AI only knows about what it’s ‘read’ and the feedback it’s been given.
“One way to think about generative AI is like predictive text on steroids, it can give very plausible answers, but they can be wrong, incomplete, or even contradictory when asked the same question twice,” Dorrington said. “An over-reliance on AI, without a good understanding of its strengths and weaknesses is likely to end in very negative (and public) failings.”
Related Article: 3 Basic Human Psychological Needs That Matter in Customer Experience
The Future of AI-Augmented CX
There is a tremendous amount of attention on using AI to supplement a shrinking workforce, according to Dorrington, and AI is very good at doing low-value-adding, repetitive tasks, freeing up human workers to deal with the unusual, creative, or human-centric tasks. However, he added, it is important not to overstate the capabilities of AI in this regard.
“It is most effective when used within a tightly defined scope — we are still many years from a ‘generalized AI’ that can perform like a human," he said. “For example, AI can be used to ‘read’ a hand-written form, match the content to expected norms and even correct some mistakes, like mismatched invoice numbers.”
But as for augmenting customer-focused interactions with chatbots, Dorrington said it can be risky.
“We also need to remember that AI has no intrinsic sense of morality or ethics, other than what was hard coded in,” Dorrington said. “If we give an auto-trading bot the goal to maximize profit, it will inevitably tend to lie and cheat. This is an effective way to achieve the goal if you don’t need to consider whether it is right or wrong. We also need to have a pretty good understanding of our own ethics before we can hope to code them into AI.”
About This Article and Series
Now, a little more about this series:
Here is the structure and details that were given to each author in the series:
- Word count guidelines: 750
- Article subject: AI-Augmented CX: How Customer Service Teams Can Learn and Adapt Faster Than Ever Before
- Structure guidelines for the articles:
- Intro: What is AI-augmented CX and its importance in customer service
- Understanding AI-Augmented CX: how it works in customer service, benefits
- Learning faster With AI: Examples of how AI can help customer service teams adapt more quickly, how AI can analyze data for patterns and insights
- Implementing AI-augmented CX: best practices for implementing AI in customer service, challenges involved and how to overcome them
- Conclusion/future of AI-Augmented CX
Post-Writing Thoughts: CMSWire's Jennifer Torres on Reporting
Editor's note: Below are some thoughts from the author herself, Jennifer Torres:
As a young reporter my newsroom was pre-Internet. But it was a step above the typewriter. We had an Atex system — a word processor that connected everyone in the newsroom. I loved it and soon after, when they were all replaced with brand new Internet connected computers, I really hated it. Mainly because I didn’t understand this new tech, and learning seemed a complicated, time-consuming process. Now, with more than 20 years of experience under my belt, I can’t really recall how we did it back in the day. Today, I greatly depend on the Internet to find credible sources, research and information.
With the introduction of OpenAI’s ChatGPT last November, I was curious and used it mainly for fun, silly queries, until the day I found myself blocked when trying to come up with a headline. So, I gave ChatGPT a try and voila, I had a headline. And it was good. Since that time, I frequently use it to explain difficult concepts in simple terms or to provide several options that communicate a thought. It definitely makes mistakes, so I never take what it gives me as gospel, but in many instances, it’s been a lifesaver in breaking up a severe case of writer’s block.
However, for this article I was presented with the challenge of not using ChatGPT. And although it hasn’t been around for long, it has become engrained in my writing routine.
So here we go.
- Phase 1: Research
- Time spent: 1 hour
- My first thought was to plug the title of my article into ChatGPT and ask it for several possible angles/outlines. But, being barred from that, I plugged the phrase into good old Google to see what else might have been written on the topic already. I found an article from Harvard Business Review called “Customer Experience in the Age of AI” and took about 10 minutes to read it for reference, ideas, etc.
- I then read another article (another 10 minutes) from ZenDesk called, “Here’s How Customer Service Teams are Using AI.” I made some notes from each of these to help angle my approach.
- I then took another 10 minutes and headed over to CMSWire to see what we have written about the topic and found an article that still had some good info.
- Finally, I spent about 30 minutes poking around to find any recent research on the topic – I read through several and found one that I put aside for possible reference in my article.
- Phase 2: Outreach
- Time Spent: 30 minutes
- I created a blanket correspondence that I emailed out to several possible sources, asking for their insight on the topic. Then I went on LinkedIn and searched for people who were posting about the topic for possible sources.
- Phase 3: Organizing research
- Time spent: 2 hours
- As the replies to my query filter in, I create a new word document for each and review their responses, highlighting the areas I think would fit best in my story. I reach back to anyone for whom I would have further questions.
- Phase 4: Writing the article
- Time spent: 2 hours
- With five areas to address in this article, I comb through my interviews for material that fits and start writing.
- Phase 5: Uploading the article our content management system
- Time spent: 1 hour
- This use to take me a lot longer, but the process of uploading the article to our CMS (which includes finding an image, editing that image, and adding keywords and links) now takes about 30 minutes. In addition to this, it takes another 30 minutes to arrive at appropriate headlines and subheads.
Total time spent: 6 hours, 30 minutes
Analysis: How Did This Human Reporter Do?
Editor's note:
At first glance, the first thing we notice here is that all three authors in this series (third one to come) were assigned a 750-word article, and this one is closer to 1,500! Jenn is a fantastic reporter and went above and beyond while immersing herself in the story. The fact is, neither she nor ChatGPT listened to instructions for word count, with ChatGPT's coming in well shy of 750 and Jenn doubling up. Do editor's even matter any more? :)
Another difference from the previous article — written 100% by ChatGPT — is the formatting. Part 1 was done manually after the ChatGPT generation; Jennifer was able to write and format her Part 2 piece here all at once, even inserting relevant hyperlinks. She also included citations — accurate ones. ChatGPT might include a citation if requested, but these should always be double-checked for accuracy. Jennifer found relevant quotes and fact-checked them as she went, sticking to the structure of the article that was requested. And, in a total separator from ChatGPT, Jennifer actually talked to human sources. This adds a huge element of journalistic integrity, whereas ChatGPT didn't, and can't, interview live sources.
Next up in this series: CMSWire reporter Raleigh Butler's article where she writes her own draft and feeds it to ChatGPT for a final product.
