Chatbots fail for simple reasons. Read our five fixes.

In recent years, chatbots and conversational AI-powered virtual assistants have become a staple of service in almost every industry and sector. But some are much more beloved—by the brands that have created them, and the customers that use them—than others.

Whether you’re designing a new chatbot and you want to hit the ground running, or you’ve an existing solution that isn’t delivering the results you hoped for, our new guide, 11 Chatbot Fails and How to Fix Them  provides essential advice from our expert team. (The same experts that are behind our own conversational AI solutions, and who helped Nuance become the highest-rated vendor in Opus Research’s 2021 Enterprise Virtual Assistant Evaluation for the fourth consecutive year.)

For this blog post, I’ve whittled those eleven reasons why chatbots fail down to a top five…

Reason #1: The AI isn’t up to the task (and the NLU model isn’t informed by voice data)

To serve a customer successfully, a chatbot has to correctly identify what they want. But most chatbots aren’t sophisticated enough to understand natural human language and conversation—and that limits the kind of experiences they’re able to provide.

A basic chatbot will be able to recognize predetermined keywords in a customer’s request, correctly guess their “intent” (most of the time), and serve up a scripted answer. And that’s fine if your ambitions don’t extend beyond automating your FAQ page.

But once you start looking to create conversational, transactional self-service experiences, the quality of the AI underpinning your chatbot becomes a key consideration. Ideally, you’ll want a platform that provides an advanced Natural Language Understanding (NLU) engine, with deep neural networks pre-trained on large-scale verticalized datasets.

To understand why, just consider how, as native speakers, we effortlessly infer meaning from context. If a friend messages you to say “Let’s grab a coffee at 11?”, and then, a few minutes later, “Actually, can we make that half past?”, you know they’re still talking about the coffee. But a basic chatbot, scanning that message for keywords? It’s going to find virtually none of the information it needs to understand what’s being rearranged, or the new time.

Get the AI right, and the ROI can be immediate. When a leading US home improvement retailer wanted to deploy a new virtual assistant, Nuance developed an NLU model that:

  • Understood 82% of its customers’ intents on day one
  • Understood 87% of their intents by week two
  • Delivered 100% routing accuracy for cases that needed a live agent

When you’re building the NLU model for your own virtual assistant, it’s smart to make use of the data you already have. Your voice channels can provide a rich set of information on why your customers are calling you. Dive into it to better understand what intents your VA should support, and to make sure you’re creating a consistent UX across voice and digital.

Reason #2: Chatbots get implemented without a clear objective

Why are we creating this chatbot? What do we want it to achieve for our business? It’s all too easy to assume you know the answers to these questions. Especially when you’re racing to outshine the self-service experiences being developed by your competitors.

But when a brand fails to explore such questions, their chatbots rarely succeed—not least because there’s no consensus on what success looks like. Even worse, they risk wasting resources on an experience that, far from delivering measurable business benefits, confuses and frustrates their formerly happy customers.

So, decide what you want to achieve upfront. Or, if you’ve already fallen into this trap, decide right now. It’s never too late to course-correct.

Your primary objective might be to improve customer experience. You might be more interested in reducing customer service costs or streamlining your live agent operations. You might want to drive sales or deepen your business intelligence. Allow your objectives to shape your chatbot’s design—informing everything from how advanced its conversational AI needs to be, to what information it needs to access.

Let’s say your primary objective is to save your contact center agents the time—and drudgery—of handling the most repetitive customer requests. It might lead you to spin up a basic chatbot that can answer the vast majority of your common FAQs. Alternatively, let’s say you want to maximize sales revenue. This objective might lead you to design a more sophisticated virtual assistant driven by powerful AI, capable of popping up with timely product recommendations for your customers and cross-sell suggestions for your agents.

One major global telco deployed Nuance Virtual Assistant to help with another popular objective for chatbot implementations: meeting customer demand for service on digital channels. The virtual assistant was soon handling 50% of cases without the help of a live agent, and delivering $14M in annual savings through contact deflection. With such a well-defined mission, its success was easy to see.

Reason #3: Brands (and bots) don’t learn from customer behavior

Have you ever been out walking, and seen a muddy track breaking away from the defined path? If so—and if your footwear was up to it—maybe you took that track yourself, realizing that it would cut off a corner and shave a minute off your route.

Now, imagine it’s your job to define the main path. Do you keep it where it is? Or do you learn from that well-trodden track, branching off, showing you the way people really want to go? Do you change your path so they don’t have to traipse through the mud?

If you’re motivated by serving those people effectively, you make that change. And whether you’re designing or troubleshooting a chatbot, you should be constantly looking for those muddy tracks. Analyze your customers’ preferences and behaviors before you deploy, to learn how they want to interact with your brand and where you can smooth their journey. Analyze after you deploy, to discover if they’re following the paths you’ve created.

You might find that your customers are constantly coming to your chatbot with needs it can’t support. If so, you might want to rethink its capabilities. Or signpost them more clearly and improve your escalation strategy. Or all of the above. On the other hand, you might find customers are coming to your bot with appropriate needs, but expressing them in a way it doesn’t understand. Fixing this one issue could save many, many customers from defaulting to a live agent.

When you keep a constant eye on user behavior, you can build a chatbot that effectively supports a wide range of user needs. A leading bank deployed Nuance Virtual Assistant to handle over 350 customer questions and answers—it averaged over 30,000 conversations a month in its first three months, with a first contact resolution rate of 78%.

Reason #4: The chatbot hasn’t mastered the channel or platform

This is another trick that’s easily missed. An outstanding self-service conversation with a smart speaker looks very different to an outstanding self-service conversation with a web-based or messaging chatbot. Equally, an outstanding experience on WhatsApp could look very different to one on an OS messaging platform.

Every channel and platform has its strengths, weaknesses, and unique abilities. The brands that currently lead the world on customer service understand this. Rather than taking a copy/paste approach to deploying self-service experiences, they ask:

  • What makes a conversation on this channel or platform sing?
  • What expectations will customers bring?
  • What capabilities are unique to this channel or platform—and how can we use them?

Apple Business Chat, for example, offers integration with calendars and Apple Pay, unlocking opportunities for more transactional self-service experiences. Facebook Messenger, meanwhile, allows brands to share rich multimedia content—like “how to” videos, and set-up guides.

When you’re designing chatbots for customer service, understanding what’s possible with each channel and platform is the key to making sure your self-service experiences stand out from the crowd. And if you need inspiration, learn how we helped one global retailer create a completely fresh experience for its shoppers. Its customers can now upload a photo of an item during a chat, and its virtual assistant’s AI will use machine learning algorithms to surface similar products they might be interested in.

Reason #5: Your chatbot isn’t part of your fraud prevention strategy

Many brands and governments have seen fraud attempts soar in the wake of the global pandemic. Your chatbot or virtual assistant can, and should, play a key role in giving your customers confidence—and giving criminals the cold shoulder.

As knowledge-based authentication proves increasingly unsecure, you might already be looking to strengthen your customer identification and verification processes with biometrics technology. Capable of identifying customers—and fraudsters—based on inherent characteristics like the sound of their voice, AI-driven biometrics technologies are often a great fit for self-service customer interactions.

Imagine a customer calls you with a question about their account, and is greeted by your virtual assistant. As they explain their need, voice biometrics can listen to the customer’s voice to check that it’s really them. The virtual assistant can then offer the customer access to more sensitive self-service options, like exploring their recent payments.

In digital channels, you’ll want to investigate what’s possible with behavioral biometrics. By analyzing everything from the way someone swipes and types to the way they hold a device, it’s possible to deliver continuous authentication—and even spot when a criminal hijacks a customer’s self-service session.

Following a successful deployment of Nuance voice biometrics solutions in its IVR and contact center, one government agency has expanded its voice authentication program into its mobile app, giving citizens the option to use a single credential across its voice and digital channels.

Explore six more chatbot fails and fixes

I hope this post has helped you understand how to optimize your existing chatbot or virtual assistant, or design a new solution for greater success. To explore into another six reasons why chatbots fail—and make sure yours doesn’t—download the guide below.

Source link

Leave a Comment