Unlocking the superpower of AI for agents: making a real difference while we wait for the Metaverse. - guest blog by Jonas Berggren.

Why do some products just work? - the rise and public acceptance of AI

Why do certain products effortlessly capture our attention and become indispensable, leaving us unable to fathom life without them? Apple has accomplished this feat multiple times, although less frequently in recent years. Take the iPod, for instance, which revolutionized the way we enjoy music through its sleek design and seamless integration with iTunes, resulting in the creation of a groundbreaking music ecosystem. And let's remember the transformative impact of the iPhone, which turned the concept of a phone on its head, evolving it into a versatile pocket computer capable of performing a multitude of tasks. Similarly, the iPad's introduction led to a whole new category of tablets, defying our expectations of keyboard-centric devices. In recent months, the ChatGPT chatbot from OpenAI has experienced the kind of attention that product managers dream of. It took five days for ChatGPT to reach one million users. It now has over 100 million users. It took Netflix 3.5 years to reach 1 million users. It took Instagram 2.5 months. It is likely that Mark Zuckerberg is still wondering why the Metaverse never exploded in popularity the way that ChatGPT has. Enough has been written about the Metaverse to leave that discussion elsewhere, but let's say that ChatGPT is extremely simple to use, it can be accessed free, and it has a distinct purpose - it is immediately useful. Complete immersion in the Metaverse requires investment in an expensive Virtual Reality (VR) headset. This is a barrier, but the need to purchase a device never stopped the iPhone. The problem is that the Metaverse has yet to define its purpose. Nobody is missing out by not being there. Some customer service specialists predicted that brands would soon need an extensive presence in the Metaverse because this would be the next wave of technological development. In the same way, companies had to embrace social media when it arrived; they would now need to embrace the Metaverse or risk ignoring customers attempting to buy products in a virtual world. Major retailers, such as Amazon and Walmart, are exploring how the Metaverse will change the way people shop, but the most immediate changes are happening in the physical world. Retailers are blending e-commerce with in-store experiences, such as allowing customers to try on clothes virtually or view 3D models of products. The Metaverse is still in its early stages, and it is unlikely to go mainstream until there is a compelling reason for people to visit.

The public acceptance of AI is a different story

This is important for executives planning their brand's customer experience for several reasons. First, it has demonstrated to millions that bots can work well. They can listen to natural language and respond the same way. Second, there is the opportunity to use these tools in the existing CX environment to create greater efficiency and productivity and to leverage the existing skilled workforce. It can augment and elevate them. Let's explore the first point. Bots can actually work. This is the complete opposite of our popular experience. People love to hate chatbots. Comedians have entire routines talking about customer service nightmares. Now, the comedians are finding that the bots can write jokes. Some business journalists have assumed that chatbots capable of understanding natural language mean the end of the customer service agent. Why would humans still be needed if the bots can answer any question? The first thing to remember is that this is not going to work outside of the box. It is possible to use an existing large language model (LLM) created by Google or OpenAI. However, it's important to note that an LLM doesn't possess deep understanding; rather, it predicts the next response based on vast amounts of data it has been trained on. You then need to apply a natural language understanding (NLU) system so it can communicate with the outside world.  On top of all this, you need to apply the specific data that your bot is likely to be asked about: your products/services and their common issues, their specification, and information about your company. If you are a retail brand with a mix of stores and online sales, then this dataset needs information on store locations, opening hours, and possibly even available stock.  If you are going to support customers in multiple languages, then this entire process needs to be repeated in all those languages. Think of a retailer like Decathlon with a wide range of products that are available online and in over 2,000 stores in 56 countries. That's a lot of work. All this data needs to be assembled into the correct framework so your bot can answer questions on anything from opening times in Paris to the price of a kayak. It then needs to be configured and fine-tuned-tested to function as planned. It's clear that the opportunity for very powerful bot interactions is possible, but as I mentioned, it's not out of the box. OpenAI could scoop up all the knowledge in Wikipedia, but that will not help when a customer asks specific questions about your business. On the positive side, though, my experience is that if you can train the bot to answer around 1,000 - 1,200 different intents about your business, it is doubtful to fail. Most customer questions fall into a common group, and once you plan for 1,000+, that includes the long tail of less common queries. Using a bot with end-to-end automation that answers both common and less common queries reduces staff because the bot will handle most volumes. The agents will be able to focus on more complex tasks or walk the extra mile for the customers they handle, but the agents would still be fewer with a powerful bot in place. There's, however, a shift where agents' roles can be turned into "AI Trainers" or explore other positions that support business growth, similar to the example of Ikea, which is training call center workers to become interior design advisers.

What does this look like in practice? Consider these areas of the customer service process

• Quality: every interaction with customers on calls or messages can be scanned, and unusual engagement can be flagged for team leaders to check - ensuring all customer interactions are checked for quality.
• Training: onboarding and training can be managed by AI, rather than requiring classroom sessions, making remote training more straightforward and more flexible - agents can take lessons in their own time.
• Coaching: ongoing coaching can be personalized based on areas where the agent needs support.
• Immediate AI response: text messages and emails can receive immediate responses, managed by AI and personalized rather than just being a standard auto-response message.
• WFM: AI can analyze contact volumes to plan workforce management requirements down to each minute, allowing more flexibility for agents to select when they want to work - matching this to the client's needs.
• NBA: advising agents on the next-best action is an extremely powerful tool where the AI listens to a call and suggests information or actions to the agent. They can ignore the advice or use it as they prefer, but often it will prevent putting the customer on hold while they search for a document.
• Translation: applicable to all forms of engagement, such as voice, texting, chat, and email. For all these channels, it is possible to have agents working in their home language but supporting customers across many different markets. The customer calls or sends messages in their language, and the agent uses their language with the system translating in real-time.