30 Jun, 2023 / Xode Article

How to convert others with conversational AI_

Conversational AI is like the more socially adept sister of a chatbot. They’re better at responding to people and deciding what they need to know based on the person’s input. Read more.

For anyone not in the tech world, AI assistants may seem unnecessary and oddly impersonal. But, as much as you’d like to be able to speak to every customer on the phone personally, the truth is they have little patience for waiting around. And the more calls you get, the less actual work you can continue with. 

Conversational AIs, on the other hand, are great at completing simple tasks, handling enquiries and getting your customers ‘warmed up’. They don’t remove the need for people entirely — instead, they ensure that your customers are taken care of before you get a chance to check in with them. Because anyone talking to you on a live chat will typically expect a response within twenty seconds to two minutes!*

Conversational AI scripts also allow you to really test what’s working within your product or service category. You can find what works best for your customers by testing particular conversation patterns and seeing how they respond. You could even try purchasing or find online a pretested AI script that’s had guaranteed results.

However you do it, you’ll want to be thinking about conversational AI. Right now, it’s expected to reach 18.6 billion US by 2026!* That’s a lot of conversation!

Here, we want to key you in on everything you need to know about AIs and their conversational genius. Let’s begin.

But first, some ground rules

AIs and chatbots are not completely interchangeable. An AI (Artificial Intelligence) is a machine learning tool for modelling and predicting human behaviour. A chatbot, on the other hand, is a program designed solely to talk to you when you're on a website or app, usually in order to get you to undertake some activity. 

Conversational AIs, as we’ll get into further in a minute, utilise the learning ability of AI in order to talk to humans as if they themselves were human. 

In short, they’re both brain and mouth. But, for the purposes of keeping things simple, we’ve referred to conversational AI as plain old AI a few times throughout this article. But, to reiterate, they are quite different.

So why are these AIs so chatty?

A conversational AI is like the more socially adept brother of a chatbot. They’re better at responding to people and deciding what they need to know based on the person’s input. That’s because these smart, chatty programs recognise not just the individual words but sentence intent and can replicate that intent n a reply that sounds human. 

However, a traditional chatbot can respond to enquiries quicker than a person can, or even a conversational AI, because it works on simple responses. A basic AI may ask you to select from a preset number of answers, which reduces the amount of confusion inherent in understanding human speech. For example;

The basic example above illustrates how you can answer someone’s enquiry but still redirect the conversation down a predictable route by giving them limited answering options. However, this option may not predict all avenues, and it limits the number of responses someone can give, as you have to figure out the questions people will ask preemptively. 

But some solution is better than none! 

The history behind the AI

If you want to get straight to the ‘why’ of conversational AI, we suggest you skip this bit. 

The first ever program to ‘chat’ was ELIZA, named after Eliza Doolittle in George Bernard Shaw’s Pygmalion, a working-class character that is ‘improved’ by each interaction she has with others. This, however, would describe modern conversational AI and not 1994’s ELIZA, who was a chatbot through and through. ELIZA worked by word recognition, giving prewritten text when it recognised a certain word without understanding the context of that term. 

ELIZA’s job, if she could be said to have one, was as a therapist, with her ‘script’ based on common therapy techniques and answers. However, the amount of therapeutic help she delivered was pretty nonexistent, given she couldn’t really decide what words meant within a sentence.

AI itself is much older, depending on your concept of what a computer is. The basic theory of AI began in the 1950s, when two concepts were proposed, one known as symbolic and the other connectivist. This first meant allowing computers to create a symbolic version of the world in which they could reason in computer terms, while the second would force machines to acquire knowledge through learning. 

This second approach is closest to what we now think of as AI, and in particular deep learning, which is computer reasoning by mimicking the process of neural activity. It was, in fact, inspired by the discovery that our own brains work on electrical impulses or neurons. So, in a way, our own brains are the first model for machines.

But, just to be fair, the first real and recognisable AI was two separate programs for chess and checkers. They were both coded at the University of Manchester on the Ferranti Mark 1, the world’s first commercially available computer.

More about why modern conversational AI is great

While chatbots rely on preselect answers, a conversational AI uses machine learning, which means it analysis the text itself and tries to understand what’s going on. Even if that doesn’t sound too impressive for a person to do, it’s actually incredibly difficult for machines to do because they don’t have our inherent ability for abstract thinking. 

GPT-3 (Generative Pre-trained Transformer 3) is one particular AI which has broken the mould in as much as it learns based entirely on input. The more it's fed, the more it can create content that sounds passably human (and no, this writer is very human, thank you). The trouble is, GPT-3 isn’t so good at determining actual facts and interpreting abstract data, often stating its limited sources as hard facts.*

But here's the cool part: GPT-3 and its newer versions, like ChatGPT-4, are getting even smarter. They're getting better at understanding what you're talking about and saying things that make sense. They don't talk exactly like humans, but they're getting closer. This means they can help with things like answering your questions, writing articles, and more. So, as these AI systems keep getting better, they're going to change how we talk to machines and do a lot of cool stuff in the future.

As a learning algorithm for chatting with people, conversational AI isn’t rule-based, choosing how to respond to people based on the context of their message.

So, why do I need a script?

The same reason Shakespeare needed contemporaries. Everyone needs someone to emulate. 

Your marketing script will be the learning basis for a conversational AI, defining the boundaries and context of the conversation. It’s the launching pad for learning, and what prevents your AI from talking about glazed doughnuts when they should be selling car insurance.

How does it all work?

Without getting into the code of it all, conversational AI is a number of ‘structures’ which send an output based on what the input is. Think of it as the framework for categorising new knowledge. Based on each interaction, an AI can determine what set of rules its supposed to implement to get the answer. Here are some of the structures that enable a typical conversation to take place;

  • Analysing Received Input - This is a fancy way of saying your AI scans your text to find the intent and context of the message. All conversational AI will do this.
  • Dialogue Management - Heavily dependent on NPL, this is the actual response itself. So, if the question is ‘Do you sell peanut butter’, the dialogue analysis will check previous responses and the current question to give an answer. If the conversation goes badly, then that’s a learning curve.
  • Machine Learning - Machine learning (ML) is the studious part of AI. These are the algorithms upon which the AI is founded, which improve with time and previous messages. The most successful messages are obviously those that close sales, and you can tailor the success parameters of your AI to search for the result you want.
  • Natural Language Processing - Natural Language Processing (NLP) translates text from one language to another, summarising it and making it user-friendly for machines. NLP can even detect and respond to voices in real-time. This form is currently being succeeded by deep learning, which more accurately simulates the behaviour of the human brain and attempts to replicate it, specifically its learning ability. 
  • Reinforcement Learning - All responses, including the AIs, are stored and analysed later to check against when coming to terms with new information. 

So, how close are we to Terminator

Without getting overly dramatic, not at all.

While it may seem odd to hear about machines learning and mimicking human thought patterns, they still behave exactly as machines. This means they don’t really have any conscious ability to recognise what they’re reviewing or to think about it in any way that we would consider contemplative human thought. 

All AI, as it currently exists, can be summarised as machines learning to do their jobs better. The trouble is the language we have to describe these processes is far outstripped by the processes themselves, which is why we apply terms like ‘learning’ to these artificial processes.

Why might I use conversational AI?

Now that you can be sure your chatty electronic buddy isn’t plotting doomsday, you might wonder why you need him at all. But, if you’ve ever been stopped on a website by a popup message box asking if you need help, you’ll know roughly what a talkative AI can do. 

Typically AIs are embedded into websites to prevent customers from leaving, offering their services in the same way a human store owner would do a confused customer. The trouble is anyone can enter your website night or day without you personally being aware (at least at the time). And there’s no space limit either, which means you could be dealing with hundreds, possibly thousands, of customers at once.

Conversational AIs allow you to speak to these customers at the same time, taking them on individualised journeys and responding to concerns. They keep customers on the path and make them feel heard.

Good AIs will also be integrated across multiple user platforms, usually a separate CMS (Customer Software Management) program. This allows your AI, who had a conversation on your website yesterday, to respond via email or text with an update. Having this feature is incredibly useful because, unfortunately, no one is likely to flock back to your website unless there’s a good reason. So reaching them via common forms of communication allows you to finish the conversation or prompt them to return (maybe with an update on their enquiry, a product or a discount).

Who might use conversational AI?

Anyone in business. Customer service-based industries typically use AIs to handle enquiries before putting people on the phone, as a lot of problems are easy enough to solve with an AI, particularly if they involve referring people to documents or company policy. 

Sales and IT are also big, especially as AI can be counted on to close sales for a lot of products. Conversational AI can recommend which customers to follow up with and even help salespeople meet their quotas by continuing to sell in the background. However, you don’t need to use an AI for the entire sales journey — you can interject with a human responder when the sale becomes ‘hot’ or likely to proceed. 

Since AIs are terrific at recording information and tracking specific keywords, they’re also perfect for conducting surveys and guiding people through a list of questions. So if your jam is data collection, you might want to think about installing a conversational AI to help people answer those tricky-to-understand questions.

If you want to know more about chatbots in general, you can check out our other article, ‘AIs and chatbots are more or less expected now’. It’s got a lot of similar information which you can skip through, but is based on our own experience working within the financial services industry — so if that’s you, we highly recommend checking it out.

Personalising your AI

You can totally personalise the language and sound of your conversational AI to fit in with your company’s branding. That’s why we talk about writing or sourcing a ‘script’, the original material your conversational AI learns off. And no, your conversational AI won’t get drunk at the Christmas party and start going off on a tangent. Its operational parameters still heavily determine the answer it gives, meaning it can’t go far wrong without bumping into one of its preset rules.

AI scripts are also useful because they give proven sales tactics, meaning you can begin an AI on a winning streak.

Looking after your AI

While this sounds like a lot of maintenance just getting your AI to talk, this is all really happening in the backend once your website or app has been completed. The point of the AI is to act like a virtual assistant and to hand you the ready-to-go cases, people who’ve been vetted by your AI and are now ready to speak to a real person. Or, if you don’t need actually to speak to your customers, your AI can handle the whole operation, letting you know how it went as part of your sales data (and for that, you’ll need the CRM we were just talking about).

Ordering yourself a chatty robot

If you’re updating or designing your website anew, or creating some software feature, an AI (and a smart AI at that) might be the way to go. Even if you’re not directly selling something, an AI can help people understand your offering. Because no matter how many months or years you’ve spent designing your customer journey or writing clear and relatable content, someone somewhere will find a problem. This is where AI becomes your call centre, without the need to pay for an office, staff or a jumbo-sized tin of coffee beans.

At Xode, we like all our things to be made specially, including your chatbot or conversational AI features. Your support shouldn’t be merely an afterthought but embedded throughout the whole customer journey as something that’s truly useful. This means you can really tailor the functionality of your conversational AI outside of the typical user journey platform, catering for any number of possible responses. 

If you want to talk to us more about conversational AI or its very interesting history and applications, please reach out to us. We’d love to talk to you.