5 strategic mistakes in implementing chatbots for Customer Service and how to avoid them

Chatbots taking part in customer service are slowly becoming an everyday occurrence. Study shows that up to 85% of organizations by 2020 will have made a decision about implementing a chatbot [1]. This means that everyone will at some point be served by a bot. For some people, having an opportunity to talk to a bot is exciting. They try to test the bot by asking it questions and trying to catch it making a mistake. The vast majority of us, however, prefer personal contact with a consultant of the company in which the service or product is purchased. Due to the rapidly growing costs of employing people, the possibility of talking to a real person is unfortunately slowly becoming a luxury, available only in specific situations. So, how to design a chatbot in such a way that it does not alienate the customers from itself – and thus from the whole company?

Underestimating the complexity of the task

It is easy to make mistakes when implementing chatbot solutions. The market is being flooded with more and more information about negative User Experience when contacting a bot [2]. Most users point out the bot’s imprecise statements due to an unforeseen path of user behaviour or lack of understanding of the context. Although it should be remembered that a bot is not a human being, some popular errors can be avoided already at the stage of implementation planning. We have decided to share with you the ups and downs from our experience in building conversation bots.

The most common strategic mistake you can make is the underestimation of the venture by the company that plans to invest in the chatbot. There are many myths about chatbots on the Internet [3]. One of them says that a chatbot can be built in three days for 10 thousand zlotys (a little over 2 000 EUR). On the one hand, it is true – there are so-called rule based chatbots, which can be created even in one day, but the range of issues they can support will be quite narrow. The preparation of such a bot also requires excellent knowledge of the industry to prepare a detailed description of the user’s behavior that the chatbot should be able to handle. Therefore, ladder chatbots require little technical work, but before it is done on the business side, much more time-consuming conceptual work takes place. Bots based on neural networks and Natural Language Processing [4] can handle many more queries and learn from historical conversations, but implementing them is a matter of weeks or months, not days.

Implementing a bot that does not meet your users’ needs

When planning to automate communication, which is what chatbots are basically used for, you should first answer a few questions, and even better – a few hundred questions, because this is how many unknowns come up when honestly assessing the amount of conceptual work on a good chatbot. After all, the chatbot devil is in the details. The first step to having a competent bot that really gives value to users is to define precisely what this value is supposed to be. It is the users’ expectations and lack of really understanding them that are at the root of the second strategic error that we encounter when implementing chatbots. If you don’t do good research on the user’s needs, you’ll waste a lot of time on training the bot to handle queries that never happen, while ignoring those that do. We are talking about so-called ‚garbage intentions’, which means teaching the bot not on the basis of real use cases – e.g. conversations between consultants and users – but on the basis of the presumptions and experiences of an individual person. Such assumptions are very often wrong and the company ends up with a bot that can handle 3,000 different questions about the size of clothes, while customers only ask for a return.

Skill overkill – too many tasks for one little bot

It is also very easy to succumb to the temptation of delegating many seemingly simple communication activities to the bot, which in consequence makes its training border on a miracle, due to its complexity. So the best starting point is to train your bot to handle simple, clear questions, which are much easier to work out. It is necessary to develop different behavioural pathways depending on the user’s response. Both rule based and AI based chatbots [4] need a precise range of competences. The more competences there are, the more difficult it is to reach a high level of coverage, as the number of answers to the possible questions grows very quickly. So it is better to design a robust chatbot that answers questions about flight delays within an airline than to additionally try to verify which airline, airport or date and time of flight it is about.

Trying to humanize the bot

All conversational solutions based on neural networks and elements of artificial intelligence require a huge amount of work to represent a modicum of human intelligence and humour. Suffice it to say that as many as 10,000 people are working on the development of Amazon’s voice chatbot Alexa [5] and although she can already have a conversation that makes sense, she still has serious problems with understanding the context, jokes or emotions. This results from the fact that the human mind interprets messages in thousands of different ways, not just as a sequence of concrete signs. Of course you can add many conditions and dependencies that will make the chatbot more and more ‚human-like’, but the truth is that for at least a few more years it will not be able to compete with humans. So, is it worth telling your customers that they are dealing with a human consultant? Our experience shows that a much better solution is to inform the user at the very beginning of the conversation that he or she is dealing with a bot. Thanks to this, their expectations change, they try to specify their questions better and inquire about one piece of information at a time to make the bot’s “life” easier.

Blindly trusting the supplier

There is no shortage of software providers on the market. After searching for the phrase ‚chatbot’ on the Internet, you are flooded with advertisements of rule based bots based on ChatFuel and other tools, ideally suited for example for Facebook sales. If you dig a little deeper, you will also find entities that have more comprehensive solutions in their offer. These systems, designed for larger Customer Service departments, usually have both a specific CRM for lead management and a dedicated chatbot. However, the fact that someone offers such a solution does not mean that they can be trusted blindly. The client should be present, active and aware at every step of the implementation. They should observe the work of the company they have engaged to do the implementation, make modifications, select the necessary groups of functions and reject the unnecessary ones. Close cooperation between the client and the solution provider for Customer Service is a guarantee of good implementation and subsequent production success.

If you are looking for a chatbot for your company, I invite you to a webinar during which I will talk to a bot designer about the pros and cons of bot implementation.

Sources:

[1] https://www.google.com/url?q=https://www.gartner.com/smarterwithgartner/gartner-top-strategic-predictions-for-2018-and-beyond/&sa=D&ust=1556110925364000&usg=AFQjCNFD7fLYHWq6-a0FP_3eGSJDeXhViA

[2] https://chatbot.fail/

[3] Two roads to chatbot implementation. How to choose the right one?

[4] Chatbot in Customer Service – facts and myths

[5] https://www.digitaltrends.com/home/10000-amazon-employees-work-on-alexa/

Our Blog

Blog posts

How to reduce customer service costs by as much as 30%?

There are many ways to optimize costs or increase the efficiency of your customer service department. If you are a…

Innovation ROI – how to measure it

Buzzwords Innovation, Digitalisation, Blockchain, AI – we all hear these words every day, yet they still hold some kind of…

5 strategic mistakes in implementing chatbots for Customer Service and how to avoid them

Chatbots taking part in customer service are slowly becoming an everyday occurrence. Study shows that up to 85% of organizations…