We already know two things about chatbots for sure:
They can increase ROI and customer satisfaction when used in customer support delivered by conversational messaging instead of voice calls. Customers type questions over chat and the bots add value by understanding natural language questions and answering the easy questions immediately.
They massively increase the convenience or ease of use in doing some tasks such as playing songs on Spotify or Youtube, or buying generic products. Alexa and Google Home fall into this category. Just say “Buy more toothpaste” to Alexa and toothpaste will arrive at your door in a few hours time.
If we want to understand how this technology will progress and advance we need to understand what chatbots are not good at right now.
There are also some challenges to interacting conversationally with software. The main challenges being:
Challenge #1. Many tasks are better done on a GUI than via text or voice commands. How do you go back and change something you said previously in error for example? On a GUI it’s simple to go back and change something. It’s also clearly faster to click buttons than type instructions or even say things in many cases.
Challenge #2. A related problem is, that since the interface is “invisible”, it’s hard to know what the bot understands and what it doesn’t understand. How do you discover all the things it can do that are relevant to you? You can try out a few things that you think it might understand (which is inefficient in itself) but what about all the things you didn’t think of?
The challenges are clear, but what do they mean for the future of bots? Are we going to have to wait for bots to be more intelligent before they can be really useful? Or does it mean something else?
In my view, it means something else. Soon bots are going to become more operational rather than more conversational.
To understand this we need a simple framework of bot evolution. The idea is that bots are going to progress through a number of stages which I have named Imitation, Operation and Integration.
Of course in reality, such stages or transitions are not as neat as I’m suggesting here, but the trend is the important point.
At first bots imitated humans, human customer service agents in particular. The best use cases for bots has been getting them to act as a filter for human customer service agents, answering the simple repetitive questions that customers ask and then escalating the conversations the bot doesn’t understand to the human agents.
The next step in the evolution of bots is the point where they outperform the human agents in an unexpected way. Bots are not limited to a conversational interface like humans are.
It’s going to be a while before bots can outperform human agents in understanding and resolving hard customer service queries, however bots can outperform human agents right now by doing operations through graphical widgets and proactively linking services together.
Of course extensive use of graphical and other interfaces by bots directly address the issues raised in Challenge #1 (GUI is sometimes preferable).
Even at a time when many people are not even aware of bots (although they certainly have used them in one way or another), we are making the transition from the Imitation to Operation.
Bots can use a conversational interface when it makes sense to do so, but they can also communicate instantly using any interface that best suits the situation.
For example, if a customer wants to know how to make a booking, the bot doesn’t need to describe to the customer how to do this or give the customer a link in the way a human agent might, the bot can simply allow the customer to book through a graphical widget that it presents to the customer in the chat. All the friction is removed for the customer. A human agent will have a hard time doing that.
And of course when the customer has finished their purchase the bot could suggest some other relevant actions, such as reserve parking, and offer to do it for them.
What the bot essentially becomes is an intermediary between the customer on the one side, and all the ways they have of accomplishing the task at hand on the other (including speaking to a human agent). The line between help, instructions and operations will become blurred.
Users will no longer need to know which software they are using as the bot will abstract away the underlying systems.
The third phase is where they become a part of every product, device and service at a granular level.
Not only can bots use graphical interfaces within the chat, the bots can be fully integrated with normal graphical user interfaces. Interactions with the bot change the GUI and interactions with the GUI change the bot in some way. We call this concept “CoChat”.
Bots inside an application will know when you have a problem and respond accordingly. Everyone know that every truly great idea was first tried by Microsoft 20 years too early in a way that no one ever wanted to revisit the idea again. In this case it was Clippy, the paperclip helper, which very quickly became associated with compounding any problem you had by appearing, with a big smile, after every problem you had unable to offer any assistance. You would need to minimize Clippy before you could deal with the problem.
This time however it will be different. The implementation will surely be more subtle and the technology more useful. If you are using a graphical interface or a product and are not sure how to do something, or perhaps don’t want to tediously click the screen to get something done, you might instruct a bot to do it.
Bots will be particularly helpful in product and app configuration. When you buy a product you will scan a QR code and from then on chat with a bot which will help you set it up.
If you see an advertisement you will be able to ask the associated bot questions and buy things right from inside the chat.
There will also be location based bots and the emergence of augmented reality will just accelerate this trend.
From a customer service point of view the human agent will also become more integrated with the product experience.
The user won’t have to separately contact the human agent by phone or otherwise and then start by explaining the problem. The human agent will automatically become part of the conversation / operation when it is appropriate for them to participate.
To a limited extent having a human as a fall back negates Challenge #2 (functionality is hard to discover) because humans are always a fall back in the case that the bot can’t help. Of course it will be impossible to have a human as a fallback for every use case, just as it is impossible to have a human as a fallback for every Google search. It however will be possible to have a human as a fallback for a far wider set of use cases than it is now.
I hope I have outlined a compelling short to medium term vision of our bot future. Of course, I have purposefully left out the point when bots become truly intelligent to the extent they are able to deal with context, memory of past interactions and ambiguity sufficient well that they largely negate the need for a human fallback. This is because it’s really impossible to determine when if ever bots will be this good.
It should be noted that bots will surely improve dramatically from their current levels of ability and that will reinforce their adoption and their progression through the stages of evolution outlined above.
It should also be noted that the perspective I have offered of bots here is a customer service centric one. This is because currently customer service is the most popular use case for bots given that the ROI is very clear. Bots answering simple repetitive questions scale much better than humans.
There is however a whole world of bots outside the customer service use case. These bots act in a similar way to applications of one sort or another in enabling customer to do tasks and there is no limit to how they can be programmed to behave.
In addition, customer service also tends to suggest a one to one interaction between the bot and the customer, but bots are also ideally suited to leverage group connections on existing social networks and chat platforms.
In my view, the Operation and Integration phases of the evolution of bots include many of these customer enablement bots.
These types of bots will be particularly prevalent on company social networks or messaging platforms where they can monitor and coach employees and generally help them answer questions and get things done.
Ultimately the bots may lead (hopefully) to the creation of a healthier type of social network where less emphasis is placed on how to mine human attention for advertisers to focussing on how to create real value in human lives by optimizing their attention and time.
Many things can be delegated to those million butlers on standby including filtering information and tasks to what is essential for the individual.
Bots that imitate humans have got a lot of attention in part because of their immediate relevance to the customer service use case. And, rightly so, because I don’t think there is a person on the planet who thinks that there is no room for improvement in customer service right now.
My message however, is that bots will progress from this narrow use case in ways that are unexpected. They will not progress quickly to human level performance in conversations, but they will become much more useful in many other ways. There is no limit to what software developers and bot builders will come up with.
It’s likely that there will be some winner take all effects in bots if history in software (and most competitive domains) is anything to go by, and perhaps shortly we will see a few billion dollar bot companies.
My message to companies and entrepreneurs is that every company is going to be affected by these developments. We are just at the beginning of this movement and already we are seeing multiple opportunities per company where bots can make a difference.
Companies, that don’t want to be left behind, need to anticipate how quickly bots are moving beyond the narrow customer service use case of the Imitation phase. Their digital strategies need to be future proof to these advancements. They need to take this into account when selecting the technology on which their bots will run.
Hopefully with the framework outlined in this article you are now better equipped to understand how bots will evolve and what to look out for.
Perhaps bots can ultimately provide some sort of antidote to the “attention mining” practices that the software industry has embraced so successfully and help us refocus on optimizing people's time and attention in a way that is more productive.
That’s our optimistic vision of a bot future.