It is often difficult and frustrating to interact with chatbots via voice when they are not designed well. If you’ve ever tried doing something on your phone using Siri or Google Assistant, you’ve probably had situations where you’ve rephrased the same question in what feels like fifty different ways and still not been understood.
Sometimes the system simply doesn’t understand you at all, and other times the speech-to-text engine just does not recognise your particular pronunciation of a key word or phrase. Too often these issues compound upon one another until you have no choice other than to demand to speak with a human agent. While accuracy is improving (69% of users consider chatbots to be useful), around 33% of users still say that chatbots can’t help answer their questions – our experience is that the problem is bigger than this.
If you work in the banking world, you know that when customers become frustrated this has a negative impact on their trust, loyalty, and confidence in your service. More companies than ever before are offering conversational technology for the banking sector. If your organisation is looking to make an investment in such tech, you’ve most likely already been exposed to the dizzying amount of choice available. You’re likely also still a bit sceptical about some of the promises being made, especially if you’ve had negative experiences with chatbots in the past.
This blog will share with you some of the secrets to finding conversational AI technology that customers will actually enjoy using. Virtual assistants can deliver delightful customer experiences, and we want to make sure you know what features to look out for when evaluating a conversational AI supplier.
The keys to accurate speech recognition
Embrace the fact that language is imperfect
In a natural, back-and-forth conversation between two humans, the way in which we speak is always imperfect. We will find endless ways to avoid saying what we really mean: we hesitate, we pause, and we double-back and repeat ourselves. Sometimes we mumble or our words get cut off. It can become even more complicated when implicature is involved, and things like coreference resolution and repair are important aspects of language understanding.
One of the key mistakes developers of conversational tech often make is assuming that humans always speak in clear, concise sentences that convey an exact and rigid meaning. This is one aspect of what can make conversational tech a nightmare to work with. Too often, humans have to effectively speak like robots in order to be understood. If your customers have to change the way they naturally speak, this will cause frustration.
We recommend choosing a conversational AI supplier who places the imperfection of language at the centre of their approach to development. Banking conversations are often laden with financial terminology, statistics, and numbers – especially for business banking. There is always the possibility that a customer may use a term incorrectly. Virtual assistants provide customers with the opportunity to conversationally ask for help and clarification when they’re uncertain. The best conversational tech doesn’t assume the customer is always right, but works with them to get to the heart of their needs.
Support a variety of accents
There is nothing more frustrating than a conversational assistant that doesn’t understand the way you speak. Almost everyone has a bank account, and all of your customers deserve respect and understanding. Be wary of conversational providers who use a narrow range of training data that biases one particular accent or way of speaking – and the likes of Google’s pre-trained text-to-speech API is just too general for the language in your domain. Less advanced conversational tech will often struggle and consistently fail to recognise words pronounced in different ways.
When you ask for a demo from a supplier, make sure you test for multi-accent support before you make an investment. Your customers come from all walks of life, and it’s vital to make sure that training data is being sourced from a representative range of demographics.
Work with background noise
It is important for conversational AI tech to be made available over the phone, especially in banking. But building virtual assistants that support phone calls has its own unique challenges. Often people call in from busy environments when they have a spare moment, and sometimes call connection quality can be far from ideal. Chatbots that rely on keywords will often struggle if a particular word has been obscured by background noise or cut off when the 8KhZ signal wavers.
Avoid conversational technology that has only ever been tested in a vacuum and in ideal conditions. The best virtual assistants will anticipate what customers need and can maintain a high degree of accuracy even when there is interference in the call.
Allow for interruptions
In our conversations, we rarely rigidly exchange complete lines of dialogue like characters in a bad movie. We regularly talk over each other and interrupt with disagreement, protest, or clarification. And over the telephone, without a face for reference, it’s easy for two people to accidentally try and talk at the same time.
Good conversational tech will allow the customer to interrupt and incorporate the new request. A key cause of frustration creeps in when a bot keeps talking despite the customer’s protestations that it should stop. Conversational AI should never appear confrontational and should always try to work with the customer to establish shared meaning. The best virtual assistants are always listening, even when they’re speaking.
In banking, responses to customer requests can often be substantial and include a lot of information. The customer should not be forced to listen to a lecture about information they didn’t request or no longer need to hear. In an industry where confidence is essential, customers shouldn’t feel like they’re being talked over or not listened to.
Why support for context matters
Multi-turn dialogue
Another way in which bots frustrate customers is by not remembering previously asked questions or information already offered in a conversation. Sometimes when using a bot, you’ll ask a follow-up question only to find that the bot has seemingly already forgotten the context of what you are talking about.
Customers often call their banks when they have a complicated issue that cannot be resolved using a website or mobile banking app. To resolve these types of questions virtual assistants must be able to retain context as a human would.
When researching conversational AI suppliers, it’s vital to check whether multi-turn dialogue is supported. We recommend testing for this feature when you request a demo – and remember that customers often say more than one thing at once.
Personalisation across conversations
For a variety of reasons, sometimes phone calls can get cut off. When a customer calls back, their progress should be instantly recalled by the virtual assistant, and they should be able to immediately pick up from where they left off. If you have an issue one month, and then it recurs the next, the virtual assistant should retain the details of your first conversation to more expediently resolve the second.
There is another advantage to being able to recall information across conversations. It enables the bots to offer a personalised service to individual customers. And when a customer feels like they’ve been remembered and understood, this enhances their trust and loyalty to your bank. The best virtual assistants can remember conversations with customers even if they happened weeks or months ago.
Extraordinary conversational interfaces
At action.ai, our mission is to build conversational AI technology that will consistently delight your customers by exceeding their expectations. All of these speech recognition and language understanding principles are at the core of what we do. And our virtual assistants are trained to consider the context of the entire conversation to anticipate what customers mean.
By providing an enjoyable and intuitive service, we want to improve the loyalty of your customers to your bank. We firmly believe that the unique ways in which your customers speak and express themselves should be an asset for achieving their goals, not a hindrance.
