Insights and Inspiration

THOUGHT BOX

BACK
digital customer experience

Exploring Customer Experience in the AI Era | Podcast Summary

Paul Sweeney | Chief Strategy Officer

Webio’s Chief Strategy Office, Paul Sweeney, spoke to Andrew Moorhouse, Founder and Director of Consulting at Alitical, on a recent Credit Shift podcast. This blog is a summary of what the two conversational AI experts discussed in an hour of compelling idea exchange.

You can listen to the full podcast here: Exploring Customer Needs in the AI Era.

Is ‘Digital Transformation’ a Term that is Losing its True Meaning? 

Enterprise organisations haven't fully grasped how to align everything for a proper and seamless digital transformation. The 'transformation' part is often forgotten; it's not just about using a digital interface for the same old processes. It's not just about putting technology where old processes were - the image of paving the cow path comes to mind -it's about envisioning what opportunities open up in a fully digital world. In essence, the transformative power of going digital extends beyond just replicating existing processes. It's about reaching that next level and exploring new possibilities in a fully digital landscape.  

For instance, consider the 100 million kids talking to the AI bot in Snapchat. It serves as a friend, providing an avenue to share worries or complaints. Now, imagine similar technology applied to mental health services. Digital platforms could offer therapy and counselling, making mental health services accessible to millions who couldn't afford it before. This is digital transformation in action where it has a concrete real-world application. 

Successful Triage is What Sets a Useful AI Chatbot Apart from a Failed One 


Can Chatbots Really Understand the Emotional State of a Customer?
 

In order to triage a customer correctly, you need to understand what their problem is and their emotional state. In a conversation between an AI chatbot and a customer, it is a tricky task for a bot to pick up a person’s state as implicit dissatisfaction goes beyond sentiment analysis.  

For example, the customer might say, "I've waited at home all day for my delivery," but the chatbot could well miss the frustration in their message and merely read the words as they stand at face value. The goal here is to dive deeper to figure out the root cause through analysing the words in context.  

Companies want a way to accurately triage a customer to the right place to help them properly. It’s interesting that when you talk to operational leaders and ask them if they'd prefer a bot or a human in their contact centre in 10 year’s times, the answer is often a bot. But the challenge is the roadmap to transition from old tech, which might have been untouched for 20 years, to a digital AI experience.


Listen to an extact of Andrew Moorhouse's recommendations

 

Is Live Chat Dead? The Struggles Facing Chat Solutions 

Live chat solutions are popping up everywhere. However, when we examine the course of live chat a little deeper, a puzzling trend emerges. Despite the widespread adoption of self-service automation, call volumes in enterprise firms - that invested millions in live chat - at best, plateaued. In 95% of these firms, call volumes for certain issues have actually increased, sometimes by as much as 8% over the last three years.   

Consider Bank of America, which spent $125 million on its Erica system. Yet, call volumes haven't seen a significant drop, and this suggests a fundamental issue that needs uncovering.  

Contact centres often advocate for meeting customers in their channel of choice, leading to what some call "Omnichannel madness" or "OmniShambles." A significant percentage of detractors, around 35%, find their issues unresolved on a chat channel. These include problems like ATM disputes, direct debit issues, and fraud concerns. Take banking fraud, for instance. When addressing it through a chat app, the chatbot often simply escalates it to a human, and so doesn't deal with fraud issues themselves at all.

The problem seems to stem from a lack of foresight. Many companies rush to implement conversational AI without a comprehensive plan. For instance, a tier one UK bank has 110 intents within its chatbot, but there are 410 different self-service terms on its website and app.  

The issue becomes more apparent when teams operate in silos. Digital chatbots, developed by one team, operate separately from the team responsible for the self-serve journey. This disjointed approach results in a broken system.  

Some companies, despite excelling in self-serve deflection, face overwhelming call volumes. In a bank with millions of calls, the transactional calls—those involving moving money or making payments—still dominate, with promised reductions in all volumes by using live chat not materialising. The challenge is how to orchestrate customer contact strategies. 

How do we Remove the Hurdles that Trip Up Digital Engagement and Self-Service Efforts? 

What's needed is good orchestration— an approach that integrates different teams to find the single best place for each customer issue resolution. Without this, the touted benefits of an omnichannel approach only add complexity and expense.  

Other issues that need to be addressed for a successful chatbot should include premature chat ending, security fails, escalation fails, ownership shift/channel shift. Designing a chatbot is about joining up all the parts of the entire customer engagement process - you should plan out the logical progression of a conversation. For example, altering session length based on conversational context should be a standard feature.  

Chatbot AI models need to be properly trained in the language that the customers will use, for instance, when someone says they've been ‘scammed’, models should be able to understand variations like "I've been robbed" or "I've been fleeced" and classify that intent as fraud.  

If this natural language understanding component is lacking, it's because the coders and the data science team don't really understand the business side and the business isn't very good at articulating what they really need to the coders. The linguists and the coders often sit in different, defined boxes and they don't gel properly to build brilliant AI conversations as a cross-functional team. 

UPS: An Example of Intents Dictating Resolution Route 

Let's take the delivery company, UPS, as an example. "Where's my parcel?" is probably the first thing you'd think a customer would ask, in fact, about 35% of UPS enquiries are just that. But there are 50-plus versions of "Where's my parcel?".  

Remember, the goal is intent-level orchestration, so, "Hey, my parcel is supposed to be delivered today. Is it still en route?" will be read as a parcel location query and if you message, say, UPS Canada through Facebook Messenger, Twitter, their apps or their website, once you're authenticated, you get a link to an app which shows a picture of a little vehicle driving on a map. That's one intent example and how it’s resolved. 

 Another intent could be if the customer said, "Did you leave my parcel with a neighbour? My neighbour Bob doesn't know where you've left it." There's a different resolution to the first instance because here the agent needs to check with UPS whether the parcel was delivered and under what circumstances. Did the driver leave the parcel with Bob, the neighbour, or did something else happen?  

So, you can see from this example that there are different resolutions for each intent, and you can't just go 'Lost parcel’ as a blanket intent label. Here's the triage working again, using multiple intents to decide on the best route. 

digital debt collection

Rethinking Conversational AI  

There's a rising voice among people in AI conversation design who are saying, "Let's rethink conversational AI".   

First of all, you've got to unify each intent. What this means is that for each use case, you have to get people from all areas in the business - marketing, operations, digital team, etc. - to come together and go, "What is this?". You then need to define the single best place for that resolution. And in a large enterprise, there could be hundreds or thousands of intents and resolutions to consider. The next step is to put in the right tech where you can triage these intents.  

For example, in the UK the Department of Work and Pensions (DWP) have 60 million public phone calls per year. It's staggering. These are the companies who do disability payments, living allowance payments, child maintenance payments, job seekers' allowance payments, and Universal Credit rates in 19 different service lines of payment. Over a third of their calls, 20 million phone calls, are people going, "Where's my money?", "Am I going to get paid today?"   

And the astonishing fact is that the DWP have no proper tech in place to deal with these calls. What they have is an antiquated system called "next available agent." So, everybody is racked up in a queue that can take an excessively long time to go through.  

Take the DWP again. If they had a self-service portal, customers could access their information themselves, and the data should tell them the date and time their payment is due, just as an agent would have done. 

Triage and Empathy are Key to Helping Vulnerable Customers  

A serious problem arises with old contact centre tech if there's a vulnerable customer on the line waiting to be answered. There are cases of customers standing on the edge of the bridge saying, "I've got no money for food. I can't feed my children, if you don't pay me, I'm jumping." But the tragedy is that they're in an hour-long call-waiting queue with no triage. Cases like this shine the spotlight on the real need for vulnerable customer triage.   

You need to triage on the front end and get the customer identified at the intent level. Getting intents right all comes down to properly training your AI on real-life situations and conversations. This way you build a custom language model for your own industry that can handle queries related to your business, and this makes the process work while leveling the roadblocks to actually helping a customer.  

A conversation showing vulnerabilities should be captured quickly and escalated promptly and dealt with efficiently. The process has to follow the path the whole way through for each specific type of problem, from routing to queuing to agent. And that isn't a magic button; that is training, that is looking at the data, that is analysing conversations, that is retraining intents and relaunching models.  

If you can gather that important information upfront with an AI chatbot that can identify intents, without a person having to wait in a queue, then they can be seen to straight away and can be routed to an agent if needed. Also, AI chatbots are taught to speak empathetically to reassure a person that they have been heard and that someone will attend to them. Conversation designers have to look at both sides when building a conversation and get into the mind of the customer who is in a fraught situation.  

Also, the agents themselves need careful training to deal with people who don't understand the finance, who are afraid and who are not in a calm state of mind to articulate properly or receive what the agent is trying to say.   

Fundamentally, how you speak to a customer is of utmost importance. Contempt is something a customer picks up quickly from an agent, such as when they are dismissive and speak down to them as if they are stupid. Many of the FCA complaints come from the manner in which agents have spoken to customers. But it would be really technically difficult to write an algorithm that could measure contempt – difficult, biut not impossible. 

What Matters to Customers When Connecting with a Company?  

Although empathy is important, especially in fragile situations, sometimes a customer just wants to know when the problem will be sorted. In utility companies, the biggest cause of dissatisfaction was an uncertainty on time scales. Imagine a person standing in in their yard surrounded by sewage from a burst pipe:  

Customer: "When is your driver coming?" 

Agent: "Oh well, we're going to be there as soon as you can, Sir." 

Customer: "Brilliant! How soon is soon?" 

Agent: "I couldn't possibly say. But they'll be there shortly."  

When looking at building a digital customer engagement solution, it is important to measure the right metrics. What is it that matters to customers? What frustrates them when they call in to a contact centre? These concerns need to be included in the foundation of conversation design. 

Conclusion  

A mindset shift is needed to get conversational AI working right. Simply buying a piece of technology but not addressing the real problems of agent-customer conversations is not going to solve the problem. The key here is to meet the customer in their channel of choice but triage, triage, triage to make sure they get to the place where their query can be resolved, be it with an AI chatbot or an agent. 

Everything you need to deliver great customer experiences and business outcomes

Experience the wonder of Conversational AI for Customer Engagement

If you need to improve your customer engagement, talk to us and we'll show you how AI automation via digital messaging apps works.

You will love the Webio experience.

We promise.







 




CONTACT US

DEMO REQUEST

Register for a personalized demo where our team will review how Webio's intelligent customer engagement solutions can positively impact inbound and outbound customer engagement in your business.

METRICS WE’VE ACHIEVED

52%
Uplift in Payment Arrangements
42%
Increase in Agent Productivity
57%
Decrease in Operational Costs
48%
Increase in Customer Engagement

YOUR DETAILS

WhatsApp Contact Button