What are the best ways to employ chatbots to boost business? Our guide outlines why they are taking over and how to use chatbots in business.
Find more technology guides, tips and advice
Find more business guides, tips and advice
What makes a chatbot?
Chatbots are basically robots (abbreviated to “bot”) or software algorithms that react interactively to the input of a user depending on the level of expansion. Entries can be made textually, verbally via voice command or visually via gesture control – however, the user who enters something receives the answer from a machine. The answers are more or less communicative and helpful, depending on the complexity and level of development of the chatbot.
Chatbots can navigate to a product or answer questions about the weather in accordance with a filter in the online shop. There is then no longer a person on the other side of the chat, but rather the answer to the question asked is attempted to be calculated using given attributes or paths or artificial intelligence.
Chatbots are therefore a kind of “communication automation” – as a consistent further development of already existing automations such as email autoresponders or lead warming routes. The mere fact that the chat app usage intensity is overtaking that of social network activities suggests that the topic of “chat communication” will shift more and more in the near future.
What’s so new about chatbots?
After studying various sources, we keep coming to the conclusion that the mechanics of chatbots are basically not a new invention. At the end of the 1980s, many telecommunications companies introduced something similar to chatbots as part of the telephone hotline, only in the “very simple” version. If I call a hotline today, for the first few minutes I will be maneuvered into the correct “channel” using an audio guide and keypad selection. “For problems with the contract, please press 1, for problems with the connection, please press 2, for a conversation with your personal advisor, please press the 9 “- this is how the first few minutes of a call go. The telephone extension stage consisted of knowing the language, the digits to be pressed were replaced by specific commands: “Please say CONTRACT” or “Consultant”.
In the sense of simple chatbots, the foundations were laid many years ago. The use of automated chat solutions is due to the change in user behavior. Today’s applications show, in various forms, essential differences to automatic telephone systems.
On the one hand the medium. It is no longer the phone call that is primarily used, but the Internet and the chat function there. On the other hand, the possibilities in the context of speech recognition and evaluation as well as existing data have become much more extensive than 20 years ago.
In addition, it would be an illusion to believe that 1: 1 chat support can be set up for countless end customers. The changed user behavior, however, requires a communication channel on the 1: 1 level. The times when problems or questions are discussed on a company’s Facebook wall are coming to an end – the times of 1: 1 communication between customers and companies are just beginning. In order to adequately do justice to this change in behavior, bots or algorithmic tools must be used.
Why do companies use chatbots?
The great advantage of good and sophisticated chatbots is that they can be used for different purposes – but mainly with the aim of increasing efficiency, the reduction of human chat agents or the ability to meaningfully process numerous individual customer inquiries. You can ask yourself what would happen if you linked your customer hotline with WhatsApp and projected the current call volume onto the service. Is that workable today and with constant growth? Without support for the agents on the chat?
Thus, chatbots are a sensible extension of customer communication channels for various reasons, although there are different scenarios for their use. Bots can also be used without direct interaction with the user to categorize inquiries or to suggest answers or next steps to the human agent – these bots are known as collaboration bots. But collaboration bots can do even more. For example, you are able to book rooms, organize meetings and appointments, or support team chats.
A collaboration bot in a company for internal use can be used as a first step in order to understand the handling and above all the pitfalls and peculiarities of the machine as a preliminary stage to a public bot. An engagement bot, on the other hand, can already interact with the customer by automatically sending reminders and messages.
These bots are also able to work through a predefined process based on predefined questions and answers. A relatively well-known use is, for example, in the field of marketing automation. As part of an e-mail campaign, the customer receives an e-mail and, depending on the reaction, a predefined process is processed. The bot shows the user what it can do. Depending on the selection, there are other options that the user can activate. Thus, questions and answers are already given at.
Customer care bots try to go one step further by being able to request personal data from the user or responding to questions such as waiting times or opening hours. The difference to the engagement bots is that in addition to the functions of the engagement and collaboration bot, the customer care bot can also record, evaluate and react accordingly.
Distinctions between chatbot types
Chatbots can be divided into three different categories. There are textual bots, audio and video bots. Textual bots can best be imagined. In the implementation it looks like a WhatsApp or other messaging app. The user types his question or text into a chat window and the bot reacts to it. The situation is similar with video bots.
Here, too, is currently (at least still), the input by the user by text or by clicking. The bot then reacts according to the input with a corresponding video sequence or animation.
A look into the future gives us an idea that one could actually combine the gesture control of the Wii with a video bot, and thereby develop a visually controlled bot. The areas of application would be extensive. These bots could react to optically recognized commands over greater distances. Whether medicine, road traffic, care for the elderly. A lot would be conceivable.
Audio bots are not a look into the future, because they are very real today. The best use cases for this are Apple (Siri), Google (Google, Home Bot) or Amazon Echo (Alexa). These bots react to audio input, so they can see what the user wants from the spoken word. The answer is also given as audio output.
This type of communication with a machine is currently the closest to real communication. As already written, chatbots try to answer users’ questions. Depending on the bot’s “level of intelligence”, they are already able to memorize or remember user preferences. to integrate the answers from previous questions into the answer to a current question. Google, for example, is already able to combine the answer to a previous question with a current question – in real time. Siri and Alexa will certainly save my voice inputs and searches – so that I can understand and profile me even better.
I wonder when the bot will start asking me personal questions – in order to learn more about myself and then be able to anticipate my preferences and needs even more precisely – without having to draw conclusions from my search queries.
Functions of chatbots
In addition to the classification into different types, I also differentiate between chatbots in terms of their ability. Ranging from very simple chat bot solutions to complex and semantically reactive bots, there is currently a lot on the Internet.
Simple text bots behave in a similar way to filtering in an online shop. The user enters a question, the chatbot tries to identify the user’s needs based on the entered word or phrase using certain key terms. An example would be when the user in a clothing shop enters the question “Which shorts are suitable for a walk on the beach”. The machine now recognizes the terms “shorts” and “walk”. If there are enough good data and attributes, the bot can respond directly with product suggestions. For this purpose, the products should have the tags or attributes “Short” and “Walk”.
If there is no clear recommendation, but rather further criteria for a product selection are necessary, these are queried by the user. The bot then provides the possible options. He would answer. “Please enter your height. The options are 160, 170, 180, 190 ”. The user can now click on one of the given options according to his data – basically like with a computer console commands or the activation of different filters in a web shop.
This relatively flat logic is already known from guided selling, where customers in an online shop are guided to the right product through a kind of filter matrix (cf. SMARTAssist solution). The difference to the existing guided selling solutions is basically only the textual feedback from the bot.
However, the filter process is defined in advance by an attribute set and the control is carried out by a very restrictive input by the user. In terms of input, more complex text bots work in a similar way to the simple ones, only the specification of the answers is not stringently specified. This means that the user does not control the chat with clumsy “console commands”, but can write in whole sentences.
The chatbot then calculates the most likely answer option and reacts to it. There is more of a feeling of communication, although the hit rate and relevance of the bot response depends very much on the anticipatory skills of the developers and conceptualists.
A further expansion stage is to include the data from previous chats and their progress in the answer. The chat answer is then a kind of result from a “predictive analytics” calculation. This type of chatbot requires a lot of historical data from past chats, as well as the use of an AI or AI solution that can analyze and evaluate the data in real time.
An extremely good data situation is also a prerequisite for the chatbot to provide a meaningful and, above all, relevant statement. For the prediction, not only chats with this user but with all other users of the platform are included. It becomes problematic, however, if the bot gets out of control due to its integrated “learning function”.
This is what happened with Microsoft and the bot “Tay”. This was supposed to bring users closer to the technology via Twitter – but it only took a few hours for the bot to make racist statements and tirades. He had to be taken off the network. In the majority of cases, the result of these chat bots is a product image with the options “Buy”, “More from this brand” or “See similar products”.
The finalization of the chat then ends with a purchase within the Messenger app. Customer care bots, on the other hand, collect data and information about the user and save it – ideally in a central connected (CRM) system. In addition, these bots can also activate reminders or other trigger chains as described, depending on the content of the data collected.
Necessary basics for the operation of a chatbot
Depending on the characteristics and, above all, the objective, different prerequisites must be met. One of the most important aspects is the provision of sufficiently good and prepared data. The data should be both textually descriptive and contextually enriched (tags and attributes). All processes possible in the app should be described and modeled. Not every process or chat will result in a close, transaction, or positive customer experience – you should be aware of this and take the necessary precautions.
Chats with a chatbot often end in an endless loop (remember the first telephone hotline “bots”) or in a non-goal-oriented dialogue. It is therefore necessary to have process exits ready. If a transaction is to be triggered via the chatbot, the integration of a payment provider is necessary. If there is already an online shop, that will probably not be the big hurdle, just as processing and fulfillment should be relatively easy to map.
The big challenge will be for companies that have not taken the step of the online shop and basically want to enter the world of e-commerce with their own chat bot. All the basics of your own online trading are not yet available here, just as there will probably be little internal experience as to how the processing processes will be managed.
How to use chatbots in business effectively
Chatbot as a miracle weapon for every purpose?
Before introducing a chatbot, the question of the usage scenario should always be asked. Using a chatbot to reduce call center or e-mail inquiries can be a good reason to use the automatisms. Bots can also help to sort and preselect requests. Here, too, relieving the strain on possibly overworked service staff is a good reason. To develop your own bot or to integrate it into an existing messenger, it is advisable to first consider which target group the bot should serve.
Not everyone uses Facebook Messenger, WhatsApp or Telegram. The question is: “Does my (target) customer even use a chat?” If so, then the question follows: “Would the customer also use the company chat to communicate with my company?” If the answer is still “yes” then ask yourself what the customer would enter as a question or problem into the chat.
Do you have the right answer ready? Depending on the question, it is decided which type of bot should be aimed for. Engagement bots can be used to guide the user to the right product or topic. If you tend to work with target groups and problems that frequently raise questions about contracts or other existing information or contracts, a customer care bot is more likely to be more suitable.
Collaboration bots, on the other hand, can help classify and group inquiries or assign them to the correct person responsible. These bots find a suitable use for very complex and multivariate products, offers or services – if a purely mathematical calculation with too many dependencies would permute and human experience or advice is necessary.
Chatbot as a shop replacement?
“Everything in life begins with a conversation, whether you are buying things or reserving a table in a restaurant.” [Facebook manager Chudnovsky] – so far, so good. The only question that arises is whether the conversation has to start in a messenger app? If you look at the last 5 years, it has clearly emerged that large brands and brands, companies and networks have brought their own apps onto the market.
Often to map a single application or a useful function – up to and including your own customer loyalty programs. Much, however, is mostly superfluous or cannot be used properly. Especially not every day. The alternative to the company’s own solutions is to integrate an offer in platforms. For example the Amazon Marketplace or other sales platforms. An example of app integration can be found in the Google app, in which myTaxi is integrated as a service. Depending on the search query, Google, for example, suggests using the app.
Stand-Alone Solutions
Company-specific solutions often serve a use case or a solution to a problem. Since this is often not enough, To encourage the user to download and, above all, to use it continuously, further (secondary important) functions are offered. Apps are seldom extended with relevant and really helpful additional functions – mostly new designs or bug fixes come with the updates on the smartphones of the users. And especially with regard to customer communication, really good and new functions are often not built in.
Often the “customer service” consists of an app link to a contact website. This is the easiest way, which at the same time represents the greatest hurdle and thus almost prevents customer contact. However, since user behavior changes and 1: 1 communication flows more into customer loyalty and the customer relationship, the idea is obvious, use a chat to communicate with the company. This chat can certainly be an interesting way for a company to communicate with new or existing customers, but what added value does this chat offer and why should someone use this chat?
Briefly explained using the example of a company. The company has been selling a product for years and decades for security in old age or for savings purposes. The product on offer is well positioned and is indeed still profitable – also for the owner. Unfortunately, new sales are falling. So the idea of a chatbot came up. To reach new and especially younger target groups and to encourage them to buy the product.
But is a chatbot in the sense of an engagement bot really the right solution? Alone, The process of the end user has to be imagined. As a target customer, the first thing I have to do is download the app. Then I have to type my question into the chat, according to the motto: “How can I protect myself for old age?” In order to receive a definitely non-independent message in response, which introduces me to the three top seller products. You can do this via the Internet and research much faster. Investing in your own chat app should be reconsidered and not pursued. Only what then?
Integration or use of existing messengers
One feasible way can be to integrate your own products, offers and services into existing chat solutions. Probably on everyone’s lips at the moment, the WeChat app is from China. This messenger app works on the principle that users can add companies as “friends”. For example, you can add Amazon as a friend (i.e. connect to Amazon) and start chatting straight away. You will also get answers from Amazon – about the products you asked about.
In the background, WeChat offers companies a kind of interface through which they can then integrate answers to questions statically or dynamically. A question to Amazon therefore results in the question being transmitted directly to the Amazon data center. After calculating the answer options, these are delivered to the user in a format (interface) defined by WeChat.
In addition to displaying the answers, WeChat also offers a payment interface. The questioner can put the products suggested by Amazon into a shopping cart and pay within the app. Amazon delivers the order. The situation is similar with delivery services for food and other goods. The founders of WeChat understood that user behavior will continue to change. As already written, achieving the permanent and repeated use of an app by the user is not that easy and requires continuous demand.
The bundling of many use cases in one app is the optimal way from the user’s point of view. After all, who likes to open a separate app for every problem? Facebook and Google also seem to have recognized this. The Facebook Messenger is also included, to integrate third-party offers. It is currently already possible to book a flight via the connection to KLM – also with an integrated payment function.
Google is taking a similar path with Allo. Allo is a pure messenger app. The functions are comparable to WhatsApp or Facebook Messenger. You can also invite your contacts to chat in this messenger. In addition to the native chat function with friends, Google offers an additional service. With Google you can formulate questions to Google in the same way as “Google Now” or “OK, Google” – and get answers. Allo does not have access to the device control, but it is possible In a chat with friends, briefly formulate another message with “@google” in advance. This message will then be understood as a question to Google and you will get the answer from Google.
Anyone who is only now getting into e-commerce and would like to try it with a chatbot should choose an integrated solution. If it is only about selling products, you can also use marketplaces such as Amazon in the bot market. A relatively simple integration of own products in Amazon Echo in combination with a fulfillment logic through Amazon Marketplace creates many “challenges” out of the way – but at the same time closes the door to independence. Amazon listens, learns and makes money.
Speech recognition as an evolutionary step
According to the first classification, the chatbots described so far are rather pure engagement bots. The answers that a user can give are suggested by the system so that the user basically controls the chat as if using console commands on the PC. Incorrect entries are attempted to interpret what works sometimes more and sometimes less well. Evolution can manifest itself in the form of voice chatbots. The currently best known voice chatbots are Amazon Echo with the product “Alexa” and the Google “Home Bot”. Less known, but deeply integrated into the Windows world, is “Cortana” – Microsoft’s search.
A voice chatbot works similarly to the textbot, only that there is no reaction to a written text, Instead, before searching for the right answer, the verbally entered question must be transformed into text (speech recognition). After the transformation, the question must also be semantically evaluated. The aim is not to break down the question into keywords, but to logically recognize the formulated content in order to essentially enable a natural conversation. The bot’s feedback is not sent in text, but also in audio format.
Current developments even go so far that the results are not only output verbally, but also visually at the same time. Google uses its Chrome Cast system for this, while Amazon integrates directly into new hardware such as screens and televisions. Here, especially with Amazon’s Alexa, the needs of users meet, the continuously changing behavior and experiences from current text chatbots.
Amazon Echo uses speech recognition to create the most natural communication experience possible for the user, and Amazon also offers an interface through which companies can import data and offer their own apps. This offer is currently being used extensively in the home automation sector. Amazon Echo is already a leader in the networking of elements of the automatic home. The reason for the predominance is the range of extensive interfaces to the manufacturers of home automation components.
The Google Home Bot is nowhere near that advanced. The presented system answers semantically more logically and can create connections between individual questions, but the interfaces to the outside, to providers of other useful apps or data, are rather limited. As a result, Google answers more intelligently, so it comes closer to natural communication than Amazon’s Echo, but the range of applications at Alexa is much greater and the lead grows with each report about it. Microsoft’s Cortana tries to be the helping assistant via the Windows operating system, although the use of voice search on the Windows system is not used extensively.
When trying to launch a self-learning chatbot “Tay” (tay.io), Microsoft had to backtrack after the chatbot learned a little too quickly and started Making racist comments and remarks. Still, Microsoft shouldn’t just be ignored. Using the botframe-work.com platform, it is relatively easy to develop your own bot in .NET and then connect it to existing messengers such as Facebook Messenger, Slack or Telegram using the botframework. Depending on the offer of existing interfaces, the list of possible bot integrations will certainly continue to grow. Microsoft is therefore not necessarily positioning itself as a provider of a chatbot, but as a technology partner through which developers and companies can gain access to existing apps.
Make, buy or integrate?
When introducing a chatbot, it is advisable to to study the current situation before developing. We personally recommend looking at the company’s incoming and outgoing communications over a period of one year. Not only the e-mail traffic (inquiries and answers), but also the communication on Facebook or other social networks should be considered. The analysis quickly reveals patterns. It can often be seen that the same or similar questions are piling up, and the answers to these questions are mostly similar.
Classify and group the answers, also try to sketch the way how the inquiring user got to the answer that ultimately fits. Here it can be helpful to use a ticket system, a BI or a CRM system in advance. A great advantage of these systems lies in the structured and historical mapping of the relationships between questions and answers.
After analyzing the communication flows, it is possible to write the first text blocks as “templates” depending on the question category. Use these templates in manual e-mails for a while to test how well these pre-formulated answers fit the questions and follow up if you notice that the questions tend not to fit. If you have enough knowledge, you can think about which type of automation or chatbot is right for your business. Integration into existing systems may make sense if your target customers use existing services.
But it may also make sense to develop your own company chatbot, which can be accessed via the company website and is not necessarily available as a native app. Basically, however, as soon as you notice that you often give the same answers to similar questions, send them automatically or give the answer before submitting the online form via the FAQs (or “proactive chatbots”) provided .
A simple “chatbot element” that is already used in many places is the autoresponder for a contact request. You probably know that. After you have sent an email to a company, you will receive a confirmation promptly. “Thank you, we have received your message and will be in touch shortly” – if you will, this is the first stage of a chatbot. Aim of the automatic answer: “Don’t call, everything is fine. We have received the message and we take care. ” With this message alone, you will noticeably reduce the number of calls or inquiries that only deal with the question: “Did you get my email?”
As a small digression to the “Autoresponder” email, you can use frequently used lead warming routes describe. These are not chatbots in the strict sense, but communication automation takes place in lead warming. A so-called “lead route” is anticipated in advance of a mail campaign. Recipients who have responded to the first email will receive another email after x days. Recipients who did not respond will also receive a second email with different content. Again, depending on the reaction, various additional emails are sent to the recipient.
Each of these e-mails has a goal, be it a download or the submission of personal data. What has been automated is the selection of which recipient receives which type of address and thus communication content. In contrast to chatbots, the lead route mainly reacts to opening and the respective reaction to an e-mail. Triggers here can be:
• Was the e-mail opened
• Was a link clicked in the e-mail
• If it was clicked, which link was clicked?
• If the recipient landed on a page (e-mail link), did they perform the intended interaction? (Download, registration, participation, etc.)
These “triggers” are to be understood in the chat context analogous to questions and answers of a chatbot, although the complexity is much less. than with a chatbot, as the amount of data to be evaluated is much lower – as is the number of possible next steps and answers.
Assessment of the future – a daring outlook
Voice bots and assistants will increase and will soon become indispensable. Star Trek’s vision of operating everything by voice control is now within reach. The connection of language and control will not only reach the automated home, but will also extend over all possible areas of life. As early as the late 1980s, a man in a black leather jacket called his car “KITT” by voice. What seemed like a utopia at the time will represent our future – whether you like it or not. In my opinion, however, the application scenarios will differ.
So you can imagine using an Alexa or Google both in looking after the elderly and in education. Why does learning always have to be linked to the ability to write? Why can’t you ask the same question as often as you like and always get a nice answer? It will be presented with various other scenarios – the technology will make it possible to connect a wide variety of use cases by voice in connection with the “Internet of Things” – in order to create “intelligent” assistants, although the probability is high, not many different ones to use wizards for one task at a time, but rather one wizard that can assist in many cases.
In our opinion, textual chatbots are only a small intermediate step in collecting information and creating an understanding of the use of human-machine communication. Voice control will dominate in the medium future. Siri was a nice start, with Google Home Bot and Alexa pointing the way. From a second perspective, we see the integration of many services in one app. The days when every company had their own app for their own use case are coming to an end.
For specialized questions, company-specific apps will still be indispensable, but when it comes to reaching a broad masses, the development will be similar to that of the marketplace business. If you want to address customers, you have to be where the customers are. That is the reason, why Amazon Marketplace and eBay are so successful – a large number of people move around on these marketplaces. The same will happen in the context of chat and chatbots. Anyone who wants to communicate with people digitally should therefore also be where the people are.
Alexa has created the best starting conditions for this thesis, Facebook is massively following suit, the opening of the messenger to external bots is, in my opinion, a pioneering step in the right direction. Microsoft is not even trying to expand the messenger market; they are investing in the still open or poorly used field of technology and interface providers. From a strategic point of view, the role of the gatekeeper is just as relevant as that of the messenger platforms themselves.
And what about Google, Apple or eBay?
Google will likely secure the technology field in the Android context, better that than nothing. In addition, Google will integrate its assistant further and deeper into the Android system in order to possibly revolutionize the verbal system control – at least in the smartphone / mobile device segment – to master the gate of the access technology.
An opening for a wide variety of providers (similar to WeChat, Facebook Messenger) would then save the way to the messenger apps to solve any kind of problem. If Google opens for the connection of external apps and at the same time uses the power of search know-how, the knowledge graph and possible device control (which per se no other app would allow), it becomes a real danger for everyone who has ever dealt with chatbots – these providers could then quickly become obsolete in the Android context.
Apple is trying to continue on the path of virtual reality, Siri will remain as it is and with this, in connection with Google developments, there will be a further decline or stagnation of Apple in the field of communication automation. In my opinion, Ebay has no chance in this market without massive investments.
Due to the fact that eBay receives a lot of “product data” from end users (user generated content), there is a certain loss of control over data quality, which can lead to problems per se. Similar to Amazon Marketplace, eBay tries to work with product templates, only the internal knowledge of the products cannot be compared with Amazon’s knowledge. A chatbot for the eBay app would be conceivable, which works semantically and can thus evaluate the product descriptions, some of which are available as free text. According to a report from 2017, eBay laid the foundation for this, although with a different motivation, through the massive integration of AI (artificial intelligence). An eBay connection to Amazon Echo would also be conceivable.
The interface from Amazon should allow from a technical point of view. In addition to the advantages for eBay of reaching an even wider audience, Amazon would further expand its own range. Whether the model pays off is left to the design and negotiation.
Author: Alexander Bittern is a technology expert specialising in business and ecommerce.
