The Rise of AI-Human Collaboration in Customer Experience: David Mulholland

The rapid emergence of AI is transforming customer experiences (CX). By 2025, Gartner predicts 90% of customer interactions will be handled by AI. While this signals immense potential, simply implementing AI tools is insufficient to elevate CX. The path forward relies on integrating human skills and AI capabilities into collaborative “dream teams.”

My perspective stems from over a decade of experience driving CX innovation for leading brands. I’ve seen firsthand how a human-centric approach collaboration can unlock superior experiences. But this integration is not without its challenges.

The Jagged Frontier of AI

Leveraging AI for customer experience is complex. A common misstep is presuming AI’s flawless proficiency across tasks. Yet, AI’s capabilities have a “jagged frontier,” a concept from Boston Consulting Group’s research, indicating uneven and unpredictable abilities.

The jagged frontier is unseen, without a guide to forecast its highs and lows. Machines “reinvent themselves rapidly and unpredictably,” and the only method to reveal the rough edges is wide-ranging real-world testing. Don’t guess where AI will excel or fail.

In practice, this frontier resembles a trek up a mountain—smooth paths occasionally, but abrupt cliffs and limitations at other times. An AI chatbot, for instance, may efficiently manage common customer service FAQs, yet stumble with nuanced complaints requiring human discretion.

This uneven landscape acts as a skill equaliser due to its variability in where it will assist agents.

Productivity can jump 43% for less proficient agents versus 25% for top performers. Like other tech, AI enhances overall capability while diminishing elite skill advantages. This could mean a significant change in how we do business in service centres.

Research has shown employees using ChatGPT outperformed those who did not by a lot.  On every dimension.

Employees using AI finished 12% more tasks on average.  Completed tasks 25% more quickly.  Produced 40% higher quality results.

Scale this over a business with 300 employees, for the cost of ChatGPT Enterprise and Office 365 Copilot, you are effectively supercharging your business through the investment of the equivalent of one new employee, creating another 30 new employee’s worth of productivity.

Unpredictability makes AI incorporation into customer experience strategies challenging. Teams struggle to predict AI’s success or failure points accurately. In my personal use of ChatGPT to create a strategy, it was hopeless at creating the strategy no matter how good my prompting was. At times it had hallucinations where it sounds convincing but was incorrect.

Yet when I prompted it to provide the steps, I needed to create the strategy, which I implemented in part with assistance of the AI for the tasks it might be quicker to do, and with careful curation, that strategy was created in record speed at a quality which I needed it to be at.

The takeaway? You have to build the ability and the confidence to effectively guide your way through this frontier through sustained, real-world usage.

You must inherently have the skill to create a CX strategy or a design or a focus group plan before you can work with AI to make it more efficient for you to perform that role.

Collaboration and governance can smooth the jagged edges over time.

Strategies for Human-AI Collaboration

As we journey across the challenging and uneven frontier of AI capabilities, our path leads us to the exploration of strategies for fostering productive and ethical collaboration.

In product design, AI offers an opportunity to take on repetitive design tasks, ‘freeing designers to focus on higher level vision and problem finding.’ By delegating rote work like creating plans, desktop research and writing copy, designers can spend more time on imagination and innovation.

Navigating this jagged landscape of collaboration, two intriguing models have emerged – the Centaurs and the Cyborgs.

The Centaur model advocates a clear division of labour. Humans are entrusted with tasks that call for judgement and decision-making, while AI is assigned data-intensive work where it outperforms humans. This model is evident in customer support systems where AI chatbots efficiently handle high-volume queries, seamlessly escalating complex issues to empathetic human agents for resolution.

In contrast, the Cyborg model entails a fluid back-and-forth between human and machine on each task. Humans kick-start the process, AI generates options, and humans then refine the output. This approach, which I’ve personally used while writing with ChatGPT, pre-empts the over-reliance on AI by ensuring human oversight and direction.

However, the AI journey isn’t without its pitfalls. Solely relying on AI can lull humans into complacency and passive over trust.

To counter this, vigilance against overreliance is essential. The temptation of automation’s convenience must be balanced with continual human reinforcement. AI should be seen as a tool to elevate human potential, not subsume or replace it.

The key lies in proactive integration versus passive reaction. We must consciously shape AI as a collaborator that augments human strengths, not merely as an automated replacement.

Some of the promising developments (or so I thought) I’ve seen which create this AI & Human collaboration is using ChatGPT Vision where you can provide an image and it will interpret it.  Using this you can upload a wireframe and ask it to do a number of tasks, from providing accessibility advice, to asking for UX feedback and even developing the code in the framework of your choice.  Now the generation of code – that is a wild prospect.  And this is jagged frontier.  Sure, it can do it, but should you?

Well as research has shown it’s not quite there yet.

GPT-4 Discovers 26% of UX Issues at an 80% error rate: 1/8 Is Harmful, and 7/8 Is a Waste of Time

In Summary: ChatGPT-4 Is Not (Yet) Useful for UX Auditing

It’s up to you as expert navigators to craft that path and act as a Cyborg or whether it’s worth implementing AI tools or AI automation so you can divide the labour as a Centaur.

The path forward is not one where AI stands alone at the heights, with humans relegated to the sidelines. Instead, it’s about finding common ground where both humans and AI can be integrated into an elite team.

Responsible Implementation

AI system as a novice employee

Think of an AI system as a novice employee – it requires extensive coaching and management!

You might find it helpful to imagine it as your sidekick, whatever works!

Like any new hire, AI will need extensive training and feedback to learn the nuances of its job responsibilities appropriately.

Left to its own devices, an AI novice may fall back on simplistic or repetitive solutions without considering diversity of approach. It lacks the breadth of experience that human colleagues bring. For this reason, direct management is so critical as models advance in capability.

Just as we would develop a training plan for an entry-level employee, proactively establishing governance over increasingly sophisticated AI is key. It prevents novice mistakes from spiralling out of control.

By providing robust oversight and opportunities for learning, AI and its human coworkers can thrive in resilient partnership. Regular performance reviews identify errors for correction, while outcomes-based audits reinforce the desired standards of service.

With the rate of AI’s learning curve, reactive approaches will not cut it. Training AI “new hires” demands aggressive leadership from the start. By viewing evolving systems as perpetual novices, we can successfully coach them into collaborators that enhance our expertise.

Break down your tasks

“How can we efficiently incorporate AI into our workflow?” The answer lies within a three-fold strategy coined as the ‘Me, Us, and AI’ approach. Take your tasks. Break them down into smaller chunks that it takes for you to fulfil that task.

Let’s illustrate this with a task central to product design – conceptualising a new product. Instead of offloading the entire task to AI, we begin by breaking it down into manageable portions. The first step might involve the AI conducting extensive market research, identifying consumer needs, and pinpointing gaps in the current market.

The designer could then generate initial concept sketches of a product that addresses the identified needs. To ensure we have a wide array of choices, we would task the AI to create three distinct design variations. Perhaps we’d like the design to be tailored to appeal to our target audiences, so we could use AI to create those options.

By dividing larger tasks into smaller, precise ones, we make it easier for the AI to work effectively. This approach enhances the quality of results, allowing us to focus on specific areas rather than spreading our efforts too thinly and settling for average outcomes. These detailed, high-quality outputs serve as a robust foundation for us to refine and build upon.

The ‘Me, Us, and AI’ process is a continuous strategy, applicable to various tasks within product design. Over time, we’ll discover more tasks that can transition from the ‘Me’ column (tasks done by humans) to the ‘AI’ column (tasks AI can handle). Some of these tasks may even fall into the ‘Us’ column, where humans and AI collaborate, harnessing the strengths of both.

Integrating AI into your workflow in this manner allows for increased focus on the creative and strategic aspects of your job – the aspects that truly drive value. AI becomes a valuable team member, enabling us to work smarter and innovate faster.

Develop in-house expertise

This leads to the next point: develop in-house AI expertise alongside adoption.

This function is called AI Operations – AI Operations is the use of AI to augment your work or business operations.

AI Ops involves developing and deploying AI-powered systems, copilots, and workflows to help us with our work. The goal of AI Ops is to augment people and teams; to elevate humans into defining and delegating the work. People will manage the execution of tasks instead of doing everything themselves, and work alongside AI systems where appropriate.

What does this team look like? Someone that knows about AI, understands automation, can use tools like Zapier and has a service and process design capability.  This enables this team to move through internal and customer facing teams, helping them handoff their mundane tasks and help them become collaborate with AI as ‘cyborgs’ on all their complex tasks.

Together, these cross-functional experts can work to integrate AI systems so that they create resilient partnerships. By investing in knowledgeable oversight from the earliest stages, we can help ensure the responsible development and deployment of the basics and be prepared for more advanced AI models.

Start small, scale carefully

It would be a good idea to start small with mundane tasks versus higher risk creative ones when implementing AI. Running controlled pilots along the way. Scale thoughtfully based on developing expertise. As capabilities improve, tackle more complex challenges while maintaining human oversight.

Build out an AI roadmap so you are building out your AI strategy in a planned way.

ChatGPT makes it possible for non-experts to easily build AI applications like chatbots by generating the required components (dialogue, etc.)

Provide lots of context in prompts, ask for multiple options, and iterate. Build a library of effective prompts. Watch out for AI overpromising – it can’t yet fully replace user research with real customers. It can’t replace developers.  It can’t replace designers.  But it can enhance their expertise.

CX Strategy

AI enables more personalised and frictionless customer experiences (CX) in several key areas.

First, customer journey analytics powered by AI can pinpoint specific pain points and improvement opportunities. This provides actionable insights versus just overall satisfaction scores.

Strategically applying AI to high-friction touchpoints first is recommended. Chatbots, for example, can automate simple inquiries and deflect these from agents. Tools like DialogFlow CX create natural conversational experiences for customers.

Knowledge sharing platforms leveraging AI enhance expertise among service center employees. These systems can provide relevant information to agents in real-time during customer interactions.

Backend workflow automation is another high-impact area. Intelligent process automation can streamline tedious paperwork, appointment booking, and other workflows. This saves agents time and gives them capacity to focus on higher value interactions.

AI also enables more personalized CX by powering detailed customer segmentation and tailored messaging. Training programs utilising AI can efficiently tailor learning to each individual employee’s strengths and development needs.

Finally, AI-powered CX analytics tools are invaluable for generating actionable insights. Rather than just providing data, platforms like Dovetail analyse vast amounts of CX data and highlight key improvement opportunities.

Thoughtfully integrating AI across these areas can make customer and employee experiences markedly more personalised and frictionless.

Adopting AI isn’t about achieving perfection. It’s about “improving outcomes” and empowering your teams to provide ever-better customer experiences.