Napoleon is credited for saying something along the lines of “if you want a thing done well then do it yourself.” His rule doesn’t apply however when it comes to creating scalable, commercial-strength virtual assistants for automated customer services. To be fair, he wasn’t to know.
If you’re older than Billie Eilish but younger than Mr Bonaparte, chances are that you’ll have had experience of automated customer services either on the phone or via text in some form. While it’s true to say that many people consider these experiences basic and many others eye-bleedingly frustrating, these services often offer a quick commercial ROI without significant investment in technology and specialist staff. The tools used to create these services often involve low-code or no-code solutions, and the result helps deflect a percentage of customers away from expensive call centres. These applications make our research scientists’ toes curl, but the tools are inexpensive and save money.
“When it comes to creating sophisticated automated customer services for voice and chat channels, the lure of an in-house tech team is strong. Don’t succumb.”
Here at action.ai, we’ve raised the bar. We’ve built a product that delivers large scale customer service automation rather than deflecting a minority of customer calls. While the cost is higher, the return on investment is extraordinary. And we’ve made it easy for businesses to use the technology.
When it comes to creating sophisticated automated customer services for voice and chat channels, the lure of an in-house tech team is strong. Don’t succumb. It is extraordinarily hard to create automated services that support the majority of customer queries, that delight people with both accurate understanding and response – and all whilst being commercially scaleable. It’s a Herculean mission that takes years of research, understanding, trial, error, and inevitable setbacks – and that’s after recruiting the right qualified experts in linguistics, machine learning, and automatic speech recognition.
The fact is, when you’re dealing with extremely complex technology and you don’t have the expertise, ‘doing it yourself’ is a waste of time at best. Sorry Mr Bonaparte. When it comes to creating rich automated experiences that understand users first time and every time, you need to bring in the big guns, and here’s why:
In order to create a virtual assistant that automates customer services scalably across a significant range of queries, you’ll need a team of deep learning experts with highly specialist skills in computational linguistics. You will also need senior software engineers with expertise in building virtual assistants and specialist UX designers who understand how to shape and support a conversational model. And if you want to support speech by offering automated services over the telephone, you will need additional skills in automatic speech recognition, speech synthesis, and telecommunications infrastructure. Creating bots with Luis.ai, Dialogflow, IBM Watson, IPSoft, or no-code or low-code models simply won’t make the grade. Finding someone with the skills you need is hard because experts in this field are extremely thin on the ground. If you do manage to find highly skilled individuals, prepare to invest in several years of research and development to apply those skills to creating your specialised automated assistant – one that is fast, extensible, accurate, and scalable in responding to your customers. Don’t be surprised if you spend a great deal of time trying to work out whether you actually have the right skills for the task in hand.
With our first-hand experience, we know it’s possible to create great bots – we have learned all of the painful lessons along the journey. We’ve dealt with the challenges, and we’ve productised the result. The solution is far more cost effective because we reuse our models and expertise and tools to supply clients with services at a fraction of the cost and time.
Our Technology over a billion actions in less than half a second.
If you want to build extraordinary conversational technology, you need extraordinary component parts. We’re proud that ours uses the world’s most advanced natural-language understanding technology and automatic speech recognition.
Our speech technology has ground-breaking word-error rates. It actually predicts when users will start and stop speaking by making use of signals such as prosody and semantics – and knowing when to ignore coughs and ambient noises as well as respond to pauses in user discourse. This not only avoids awkward silences during interactions with users, but in turn reduces interruptions from either side of the conversation.
Somewhat romantically, our tech even recognises that rhythm has meaning, making use of prosody both when it listens and when it speaks back.
Our voice bots are always listening – even when they’re busy speaking. When they’re interrupted mid-flow, they make a choice as to whether they should address the interruption there and then – or whether they should continue to speak and address it afterwards – just like humans do. Somewhat romantically, our tech even recognises that rhythm has meaning, making use of prosody both when it listens and when it speaks back. It also places appropriate and natural emphasis on responses and actions, and produces the effect of endearing human-like qualities such as empathy and kindness. These are crucial in a customer service environment where customers have a higher propensity to arrive on the scene with a stressful problem that needs resolving and emotions can run high. The type of bot voice and how real it sounds can also affect the customer experience, so we can tailor that voice according to the needs of our clients.
Our technology will deal with out-of-vocabulary words – these are words that have never been seen or heard in the training set. When it seeks to understand voice, it not only understands a user’s spellings, but knows when to ask for words to be spelled out – such as tricky words in names and addresses; this is a much easier problem to manage through text channels. Our tech also uses knowledge graphs in order to understand complex conceptual-relationships specific to your business domain, automatically helping you understand customer needs through relevant representations of customer scenarios; and as part of our training data exercise with you, we can often elicit such structures algorithmically from unstructured company textual information.
The technology understands users when they say multiple things in one turn of dialogue. This is an important factor in supporting the way we naturally speak – which of course is not always with single intents. Indeed, one clause can influence another and our machine learning deals with coreference resolution. Our technology will also use conversational state to inform meaning. It recognises implicature: for example, when a banking customer tells our bot “I’m worried about whether I can make that payment next month”, it knows they’re interested in knowing about how their account balance is offset by an anticipated expense. It knows not to start giving advice on anxiety or how to be more frugal.
It also understands categorical entailments of open-ended, natural-language inputs. Catchy, right? To give an example, we’ve created a sophisticated virtual assistant for a client that helps their customers manage budgets and knows the difference between (i) a customer talking about “money I spend on food to feed my family” which means an element of basic grocery expenses, and (ii) a customer who talks about “money I spend on food on nights out” which means a component of leisure expenses. The food categories for both examples are implicit.
You don’t need your own specialist in-house expertise in order to make our tech work ongoingly
From early stage development of our conversational AI, we have been committed to creating a solution that is easy to both implement and maintain by clients, whilst keeping client data securely in their own hands.
Today that’s what we offer. We can tailor our product to your business quickly, and because it’s fully kitted out with the full spectrum of our expertise, you will never need your own in-house experts to be able to use it and maintain it. You simply plug it in and it’s ready to go. Our modules are specifically geared towards banking, healthcare, and support for complex form completion.
From the first day that your conversational AI goes live and your first customer dials into customer services over the phone or types a message – our technology kicks into gear. The actions it performs automatically run within a sophisticated pipeline of parallel dynamically-interacting models. This speeds up responses and actions, makes the results extremely accurate, and makes sure every customer receives top quality service at once.
We provide our clients with the dialogue management source code they need. They are absolutely in charge. Our clients hook up to our Product API, on-premise or in the cloud, and they’re then empowered to connect their own in-house APIs and systems.
Clients can use our product and apply their own standards, control the tone of what their service says to customers, and of course change what their service does. In other words, our tech understands customers in their words and on your terms.
