How to build a smart conversation partner that your customers will actually enjoy chatting with?
Think about how many times you chatted with a bot on a website, to get help when shopping or to answer your customer support requests. Now think of all the times you got frustrated with those bots, because they wouldn’t understand what you typed, or because they couldn’t solve your problem.
This guide will give you 10 important improvements necessary for creating a smarter bot that will serve your customers right.
Automated customer interactions are actively growing, predictably 85% of customers will be affected by 2020 according to Gartner. Does it seem crazy to believe that any day bots will replace apps or even websites? This seems to become more real as more and more chatbots are created and given life by humans so that they can engage with their intended users. In many cases, chatbots become humanized at a remarkable speed!
The team of software development professionals at Maruti Techlabs proposes that we have to move beyond simply questioning the need for chatbots but rather develop a curiosity for the intelligence quotient of chatbots and their ability to bring the human touch to interactions as learned from their operators and users.
Different types of chatbots
There are four types into which bots can be divided: stateless, semi-stateful, stateful and loyal bots. If you zoom out, those four types can be divided into two: the first two are scripted bots and the last two are AI bots. Even though both types are able to fix an exact goal when engaging with their users, the way how they get there varies. Chatbots can be built either without or with machine learning. Hacker Noon software specialists clarify that in other terms the bots are set to either not learn or to learn from their users. Put quite simply, they are either dumb or smart bots.
The scripted bot
The first type, sequential or scripted chatbots are good at asking their users relevant pre-established questions related to a certain topic. These bots are efficient and good for entertainment as long as everyone sticks to the pre-settled script and the q’s match the a’s. Meanwhile, when an unexpected scenario occurs and a user feeds the bot with unexpected keywords, the conversation risks becoming challenged since this type of bot is not prepared to calculate this data – it is not smart enough to deal with a lot of user input.
For example, if the user asks the sequential bot to “please find cheap flights”, the bot will probably understand those keywords as its operator has beforehand prepared the bot for this scenario. Since familiar keywords are used, the bot will probably ask follow-up questions like “to which destination?”, “for which dates?” and “for how many people?”. When the user provides this required information, the chatbot service has the potential to work great.
Now, let’s imagine that instead, the same user inserts a lot of data like “dear bot, get me and my friend out of here for a long weekend to the city of love, max for 700/person”, the chances are high that the scripted bot is not going to understand all requests or needs the user is conveying since the user has triggered more information than the bot was prepared for.
Since the bot, in this case, is set to lead the conversation with its user(s), it will persist on getting answers to its specific pre-defined questions to fill in the desired parameters without leaving much space, neither, understanding for any alternative information. According to Hacker Noon’s guide to how chatbots benefit businesses, one could say that dumb bots are not created to learn anything from users that goes out of their scope. This situation most likely will frustrate its human conversational partner.
The conversational bot
We at BotXO are more interested in the second type of chatbots, namely conversational AI bots. Smart bots perform better because the technology and human efforts behind it are more complex. Being assisted by their operators, this gives them the ability to remember their users and previous conversations with those users.
Another advantage of intelligent bots according to Maruti Techlabs is that instead of only corresponding to exact sentences, they can perform more requests as they constantly learn and remember from their interactions with users. They are not built to define the exact path of the conversation but rather the opposite, they are left open for a lot of input from their users. Even more so, inserting a lot of data is a must when dealing with AI bots, since this allows the algorithms to calculate how close a user input is to an intent. This means that the user doesn’t necessarily have to input specific sentences in order to be understood. This kind of understanding in the conversation offers humans a more natural and personalized experience.
10 ways to build a smarter bot
By assuming that humans value having smooth conversations, one could ask: What does it take from a chatbot (operator) not to run a dry conversation or to avoid posing questions that are annoyingly repetitive? What helps to turn dumb bots into more understanding and satisfying conversational partners? How to reimagine their interactions with humans smarter?
Here is a list of 10 suggestions we find important to have in mind in the making of smart bots.
With the assistance from its human operator (it could be you!), your smart bot needs to:
- Become a good learner to know who its users are
- Become aware of its users’ needs
- Have the ability to sense the environment
- Be sharp to think
- Be quick to act
- Remain open to its users’ change of mind
- Formulate coherent and convincing responses
- Understand what mood you are in
- Casual talk since it helps to strengthen the chatbot’s “personality”
- Act as a helper instead of a collector
As explained in one of our earlier blog posts, the smart bot should keep some knowledge about its users in terms of what identifies them and “who” they are. For the chatbot to recognize patterns in data it receives and has earlier received, it needs to be “constantly learning”. Whether this happens is up to us humans.
Machine learning algorithms that are part of the technology behind the intelligent bot allow its operator to make sense of streams of data, which is afterward inserted in the same bot to help to improve its existing skills. When the bot operator is good at looking at and learning from a lot of data, the bot can, in turn, better perform in its following interactions with users.
Users might have very specific and at times complex needs so the technology behind chatbots needs to be complex as well to make sense of those needs. As exemplified earlier, human users might ask the bot to fly them somewhere specific at a certain time whereas simultaneously they have limitations on their travel budget or the number of passengers – and they might feed that information to their bot all in one go. How to include and make sense of all of those wishes?
It is argued that the smart bot has the human capability of gaining information efficiently or at least as efficiently as its human operator, as discussed above. To be efficient means that the bot is assisted to understand the intention of each request it receives and by making connections between them can respond most appropriately to all user needs at once.
Furthermore, considering how users interact, we propose that the technology behind smart bots is in some situations even better than humans at gaining information efficiently since it does not get stressed or forgetful the way humans do. People have often too many factors to care about, which gives bots an advantage. If they are smart enough, they will calculate users’ stress levels and respond carefully.
One could argue that users’ needs have a strong connection to their environment or the context they are situated in. Therefore, understanding this context is extremely important if the bot wants to give people a good experience.
Before the bot can perform a certain task, it needs to integrate with the physical and linguistic environment of the conversation in order to receive the information required. It is not only the capability of receiving data that matters but rather the importance of understanding what kind of requests and intentions certain environments trigger in users. BotXO’s chatbots are built as environment-sensitive, which means that they are operated to serve their users best depending on the specific time, channel and behavior.
A bot whose operator thinks sharp does not just understand the environment the user resides in, but it goes a step further. The smart bot makes a decision based on how it interprets the acquired knowledge. According to Maruti Techlabs, this decision is made by leveraging pre-existing knowledge of the user and new knowledge the same user is continuously conveying to the bot. The smart bot achieves a decision by the use of neural networks in machine learning.
Once the bot has taken the environment-sensitive decision(s), it has to properly react to keep the user engaged in the conversation. By that point, the smart or intelligent bot should know what to respond to the user. Progress towards the pre-defined goal of the interaction can be reached once the smart bot has gone through this sense-think-act cycle.
Smart bots allow the dialogue to jump between contexts that give people the ability to navigate without a defined path. Instead of remaining stubbornly stuck in scripted decisions, smart bots are open for new input or additional parameters from its users even if it means replacing or adding to some already gained information. In real life, people can change their minds at any time so the chatbot makes a smart move by respecting this change. This gives the bot users the feeling of more independence and freedom.
The smart bot’s ability to come with appropriate answers enables the conversation to flow more naturally like between two humans. To achieve this, the bot is assisted to learn the language nuances through NLU (Natural Language Understanding). NLU is a technique that powers conversations as it enables the bot to write in its users’ natural language when chatting with them. At BotXO, this function is tailored to individual languages to make things work in English, Danish, Swedish, German, Spanish and more.
Besides being multilingual, smart bots, like the ones customers can build in the BotXO platform, correct your spelling and recognize names. They use algorithms that help them to do checks for misspellings, look up names of persons, companies or places as well as to divide users’ sentences into different substitutes with the aim to find a link between them.
As a result of NLU, bots are able to understand the intent of their users’ sentences. Altogether, this knowledge helps the bot to reach its users in a more efficient way where interactions between the two become similar to interactions between two humans.
What makes smart bots extra cool and intelligent is their ability to decide what mood their users are in! They do that through sentiment analysis as part of Natural Language Understanding, while this feedback, which is provided by users, enables to sensitize the bot’s responses. The smarter the bot, the better it has become at deciding whether its user is dealing with a small issue or a very urgent problem that needs a solution. This relates back to understanding the language and the context of the conversation that allow the bot to decide the mood chatbot users are in.
Sentiment analysis also allows to measure how users feel about the bot, which is beneficial for fitting the bot to people’s mood. According to Jens Dahl Møllerhøj, Lead Data Scientist at BotXO, “being able to measure how happy customers are with the bot means we know how to change the bot to move into the right direction“.
Most humans would agree that real conversation is rarely limited to achieving just one response or one goal. Humans are social beings so besides exchanging relevant information they might try to have a little fun with their bot. Therefore, to have more stimulating conversations, the smart bot should integrate social talk into the conversation. This can give the conversation a lighter and more natural form. Enriching a chatbot with a “personality” (as close to a “person” as it can get) might enable the bot to better engage its users.
Having all of the above in mind, the biggest takeaway about the difference between dumb and smart chatbots is that the first act as collectors whereas the second act as helpers instead. This is indicated by tech professionals from Maruti Techlabs believe that “a chatbot acting as a helper is considered to be smarter than the chatbot that serves as a collector”.
This means that the smart AI chatbot lets the user lead the conversation while through its operator, it learns from the user as much as it can. So, the more data it receives, the better! Thereafter, the smart bot can show off its Natural Language Understanding power with the aim to assist its users closer to the users’ needs and desired goals.
In the same situation, the dumb bot aims to stick to its script by targeting the user with pre-defined questions that always expect a certain answer. The technology behind dumb bots does not support them to calculate how close a user input is to an intent. The lack of this understanding does not allow dumb bots to help their users in a clever and personal way, as smart bots do.
How can a bot deliver lovable automated conversations?
If we fuel a bot with NLU and a sense for the context of the conversation, this seems to be possible as it helps to deliver more personal experiences.
Any bot’s biggest role model can only be a human, so for a bot to make sense and become a smart conversation partner, human involvement is needed and crucial to their development and success in any application.
More personal interactions have the potential to trigger a bigger amount of engagement and excitement, and if you’re a business, a much better customer experience.