How do you design virtual agents that people want to talk to? What technologies make bots sound natural? If you want to increase the conversion rate of your call campaigns, expand your customer base and gain loyalty from as many customers as possible, here’s a guide from Tovie AI on developing bots that don’t annoy customers.
As virtual assistants gradually become ubiquitous in customer service, so does the backlash against voice bots. But with the right approach, conversational AI can dramatically increase your competitive advantages and improve business-to-customer interactions.
Crafting your bot script (aka bot scenario) is arguably the most important part, as it determines whether your virtual assistant sounds like a human or a robot. After all, the main purpose of any script is to offer users meaningful and naturally-sounding conversations.
However, when users pick up, they are often bombarded with offers or, on the contrary, have to endure an unnecessarily long intro. In both cases, potential customers will quickly hang up. The bot script needs to be concise and create value because the goal of this call is to offer specific services, and not talk about the business in general.
Sometimes, customers start asking questions without waiting for the bot to finish the phrase. In this case, the virtual agent may not hear everything the customer says. So, it’s best to ensure the bot “knows” when it needs clarifications or more details. Statistics say that 60% of conversations go off-script, if you don’t think through this possibility – the bot might become a lost cause. So, make sure to use different answers to the same question.
Besides, it annoys customers when the bot uses their first name – or their full name, which is much worse – 5-6 times per conversation. So, make sure you use conversational phrases, reactions to interruptions and, of course, A/B tests.
Give your virtual assistant a personality or an origin story. The idea is not only to teach the bot to recognise the contents of the message but also to adapt its communication style to it. When users are friendly and happy, the script should reflect that mood and vice versa.
Imagine that a customer is not happy with the bot and asks for a human support agent. If the company chooses to give the bot a casual tone of voice, it might say something like, “Give me another chance, please, or they’ll turn me off”. Twists like that encourage customers to empathise with the virtual assistant.
If the customer did not realise they were talking to a bot at the beginning of the call, but at some point asked: “Are you a robot?”, the bot must confess. Because if customers have doubts, they will try to break the bot. Statistics say that more than half of the people who identify the bot during the conversation, continue talking to it – just like they do to a human agent.
There are situations where customers go off script and ask the bot questions beyond the script. For example, a customer calls a computer repair shop to request a service, and at a particular stage, they ask if it’s possible to bring the device to the shop. The conversation goes off-script, the bot does not understand what it should say and begins to ask the same question over and over again.
To ensure the bot can handle complex scenarios and adapt to customers’ answers, it needs to recognise and interpret correctly as many customer intentions.
as possible. So opt for AI platforms with a large database of dialogues and NLU core, a natural language understanding technology that enables bots to recognise the natural language. With NLU algorithms, bots solve two basic tasks: determining speakers’ intents and recognising entities, the objects of speech.
Another important feature is flexible interruptions. Sometimes customers do not need to listen to the end of the bot’s phrase, they already know what services they need and want to move further. Setting up interruptions allows AI agents to react naturally and in real-time to interruptions or a new request from customers, and quickly move on to process it.
People often face a similar problem when receiving a call from a mobile operator with various offers: from changing a plan to subscribing to a mobile personal assistant to adding new services. The person wants a particular service but is forced to listen to everything the bot says – this is a waste of time.
There are also tools that allow to set up responses to interruptions depending on phrase lengths: i.e. the bot will not recognise words like “aha” or “uh-huh” as interruptions from customers.
As the name suggests, Text-to-Speech (TTS), or speech synthesis, is a technology for converting text into speech. Rapid developments in this area have made synthesised speech almost indistinguishable from natural speech. Today, synthesis is able to change words in an audio recording on the fly and enable bots to pronounce variables such as the customer name. Also, the choice of voices has expanded: there are marketplaces of voices for business where one can choose a voice from a catalogue.
Statistics say that among the top reasons users are reluctant to talk to bots are synthetic voices, unnatural reactions, and a lack of relevant or comprehensive answers. Today, for outbound communication businesses use a combination of pre-recorded voices and hybrid synthesis. Custom variables are any words, numbers, or letters not provided in the script – customer name, order number, passport details, etc. They depend on the context, so they cannot be recorded in advance. But thanks to hybrid synthesis technologies, the bot can say any variables in real-time and in the same voice.
Of course, one should consider business objectives when choosing a voice for the bot. For example, hybrid synthesis is perfect for lead generation, when it’s important to synthesise the customer’s name and middle name so that the virtual agent sounds as natural as possible. The same applies to banking information- credit volume, cashback and other personal details.