Chatbot Implementation in 8 Steps – Helpful Expert Tips for Getting Started

More and more companies are using chatbots to communicate with their customers. The motivational drivers for using bots are mostly increasing efficiency, enhancing the customer experience and reducing costs. But first, a brief introduction to chatbots.

What is a Chatbot?

A chatbot is software that allows users to communicate in natural language, via chat interface with a computer. Customers can communicate with a company at any time and request information without the need for a human customer service representative. The software takes over answering the customer’s questions.

In some companies, such as Telekom Austria or PostFinance, the use of bots brings the desired success and leads to enthusiasm on the part of the customer and the company.

However, many bot projects fail because they do not satisfy customers or have technical shortcomings. Therefore, consider the following points before you let your customers talk to a chatbot:

8 points of a chatbot implementation

1. Don’t want too much at once

By all means, don’t try to handle all processes and possible conversation flows via the chatbot right from the start. Rather, start small. Define your use case precisely and reduce complexity.

2. Predefined answers vs. open answers

Be careful with too many open response options. On the one hand, many users are overwhelmed if they are not given any answer options and on the other hand, open answers require very detailed programming and a high database from which the chatbot can learn.

3. Use Artificial Intelligence correctly

AI is often just a buzz word and it sounds good to say “our chatbot has artificial intelligence”.  However, it is important that the bot is intelligent in 2 ways. First, it should be able to intelligently understand the user’s queries. It must not be trained on direct input, but always try to understand the content.

For example, “I want a day ticket for my dog.” or “What ticket do I need if my dog rides all day.” Both times the bot should give information about the dog day ticket.

The second important factor is that a bot does not simply answer based on rules, but learns based on example conversations. In the case above, the bot would have been shown 3-4 possible conversations about the dog day pass beforehand and it would have derived the answer from that itself.

4. Identify cost drivers from the beginning

What a shame it is when you run out of budget in the middle of the project to develop further. So check the important cost drivers in good time. As a general rule, the more connections to internal systems, the more complex the bot becomes. In many cases, it is worthwhile to work without many connections at first, and instead, for example, simply link to the respective store.

5. Better to develop your own bot instead of using a bot framework

Of course, every bot framework promises only the best and supposedly has no limits, but you will quickly find the first limits. In most cases, it is therefore worthwhile to program the chatbot from scratch yourself (or with a partner). Particularly suitable here is, for example, the open source software rasa. With rasa, you always remain completely independent of a tool or development partner and can expand your bot at any time as you wish.

6. Precisely define the personality of the bot

One thing is clear, it must be clear from the beginning that users are just communicating with a chatbot. Nevertheless, many companies keep asking themselves to what extent the bot should take on human traits or really respond emotionlessly like a piece of software. The question cannot be answered unequivocally; it always depends on the use case. Should the bot rather entertain and make its users laugh sometimes or should the bot only serve to pass on information and this should be done as quickly and leanly as possible? Think about the role you want your chatbot to play right from the start.

7. Don’t forget testing

Once the bot is ready, you should test it with a selected group of users and make sure that it delivers the desired performance and user satisfaction. Start with a small group of users, test the bot internally or conduct focus groups. After a successful testing phase and a few possible adjustments, you can then do the chatbot live for the masses.

8. Maintenance, optimization and monitoring are not left out.

The bot is online and testing was successful? But don’t forget to continue keeping an eye on the bot’s performance. Use the first version of your bot to learn from it for the future and gradually add more and more competencies to the bot. This way you increase the complexity step by step.

Bottom line: there are 2 things you can do wrong:

1. Quick, quick – Without thinking, launch a simple bot that will only annoy customers.
2. Waiting forever – Wanting to cover all the complexity of customer communication from the start, thus spending endless time on development without getting user feedback in between.

Sophie Hundertmark

Nach Stationen in der Industrie und in Agenturen gehörte Sophie Hundertmark zu den ersten Masterstudentinnen in der Schweiz, die zu Chatbots forschen. Seit 2021 promoviert sie an der Universität Fribourg zum Einsatz von Chatbots im Banking und Insurance-Bereich. Dazu arbeitet sie als wissenschaftliche Mitarbeiterin am Institut für Finanzdienstleistungen Zug der Hochschule Luzern (IFZ). Sophie Hundertmark verfügt über langjährige Erfahrungen als selbstständige Beraterin für die strategische Begleitung sowie Umsetzung von Chatbot-Projekten. In diesem Zusammenhang sorgt sie für einen regelmässigen Austausch zwischen Akademie und Praxis zu allen Themen der AI getriebenen Conversational Automation. Anfang 2022 hat Sophie zudem den Verein Greenwishing Schweiz gegründet und engagiert sich seitdem aktiv gegen Greenwashing.