Design & Technology July 18, 2018
The conversation of the future will take place between chatbots
If we’re to believe the news, chatbots are taking over the Internet. But why do we think this? It takes a lot of work to get them up and running and once they are working they still malfunction regularly. Therefore, the focus should not be on chatbots, but rather on conversational UI, of which chatbots are a manifestation. We are at the dawn of an era in which conversational UI will take on a central role in almost all conversations – first those between man and machine, and then as a type of personal assistant for people amongst themselves. So, the time has come to review the current state of affairs regarding chatbots and look ahead to the coming years where conversational interactions will change many things.
At the moment, chatbots still tend to act as customer service employees. They are a very simple first line of contact that are a good solution for answering questions that were previously answered by people by phone or chat. If the bot can’t help you, you are forwarded – often in a rather clumsy way – to a member of staff.
For some situations, such as announcing an official change of address, it is still faster to complete a form, but for most other cases the bot is ideal. When the bot works properly, that is. And many bots don’t, at least not yet, especially if they are not in English. They’re often just disguised Decision Trees that the user gradually makes (chats) his/her way through.
But there are a lot of English language bots on the market already that can understand human language quite well based on machine learning. The bots learn from past conversations and are continually ‘fed’ all the variables of conversations by specialised teams, so language can be interpreted by the bot in various contexts.
So many companies are still faking chatbot intelligence and their proper function relies on manual input. That’s OK for now, but it’s not future-proof. Companies should start experimenting as soon as possible with smarter methods for getting chatbots to learn, for example with the help of Google services. IBM and Microsoft are not sitting still either and have technology that can make chatbots a lot smarter. Currently, the Dutch language is not supported preventing the technology from reaching its full potential. As soon as the Dutch language is supported and the bots can learn from conversations from call centres, emails and other chat conversations, things will quickly pick up speed. The bot will then be able to answer more questions autonomously without the intervention of teams that have to train the bot, the bot will start recognising emotions and, based on this, will then be able to choose different ways of solving something.
Chatbots are currently mostly a matter of manual work, but they are the prelude to an organisation that communicates with customers on a human level .
Oxxio is doing it right
Oxxio’s chatbot answers more than 55% of customers’ questions in the app. Oxxio has two employees whose full-time job is to optimise the chatbot in the Oxxio app. The chatbot can do two things:
- Pro-actively inform customers about things such as energy consumption and the new year-end billing statement;
- Reactively answer questions.
The two employees enter all variables and conversations. This doesn’t necessarily make the Oxxio bot smarter, but it does enable it to do more and more things. This has contributed to a call reduction from 600,000 to 300,000 over the course of three years. Therefore, it is a nice and efficient way to handle simple customer questions.
The next step: having advanced conversations with a bot
Once chatbots are self-learning, conversations can become increasingly complex, thus becoming more ‘natural’. This will significantly increase the chatbot’s possibilities. One example would be receiving mortgage advice via a chatbot. This may seem very far off yet, but the idea isn’t so crazy, because such advice stems from standard calculations based on data.
It is crucial to understand that this needs enormous amounts of training and data in addition to manual input to improve the bot. Through learning and training based on the previously mentioned call centre conversations, emails and chats, the bots will get increasingly better at knowing how to approach various situations and conversations.
Good news for us as people and users! When you log in, you will talk to the same ‘person’ every time, who will immediately know your circumstances and history. The bot will access – possibly based on voice recognition – your information and review it in a fraction of a second. Paradoxically, this will make the experience of talking with a machine extremely personal and pleasant. Even more personal than when speaking with a person, because the person won’t know everything about you.
These conversations will not take place tomorrow, but for simple things, such as a question about a flight or changing a reservation, you will be able to talk to a bot pretty soon. On the company side, there will be no people involved in such situations. As is common for all digital developments, it’s only a question of time before more and more services will gain access to this technology. It’s just like with websites: previously, you could only buy books or shoes online, and now you can even buy a car.
Horizon 3: a personal chatbot for everyone
Google recently introduced Duplex, a chatbot personal assistant that can talk for us and or our behalf. Although Duplex makes a relatively simple hair salon appointment in the demo, it is expected that the personal chatbot will be able to arrange more and more for us. Making appointments, planning and booking a trip, ordering groceries – all without us needing to listen in. Language as an API between everything: from human to machine, from machine to human and from machine to machine. A personal chatbot that calls the chatbot of an airline or grocery store to book a flight ticket or schedule a delivery time.
Again, we’re not talking specifically about the chatbot, but about language and conversational as a universal API between everything. Can’t quite get your head around it just yet? Have some patience, it will be here soon enough.