Onboarding a chatbot: how to ensure a smooth start
As soon as a company decides to implement a chatbot in their service process, the actual development of the chatbot can be completed in a matter of weeks or even days if you are using a no-code chatbot platform. But what happens once the chatbot is launched?? When it encounters customers through live chat on your website, in Facebook Messenger or on WhatsApp for the first time. How do you ensure a smooth start and make sure that your chatbot will be successful?
The key for how successful your chatbot will be in helping out customers and being a valuable resource within the team lies directly at the start. By conducting a feasibility check at the beginning of your chatbot project you are able to clarify the following questions: In which business areas does the use of a chatbot offer an added value? Which problems should the chatbot solve? In which channels should it be used? Should it be connected to external systems? Answering these questions upfront plays a crucial role in the later success of the chatbot. NCOI also experienced the importance of this approach during their chatbot project:
“Think carefully in advance about which questions you want the chatbot to include and pay a lot of attention to the analysis part, so that you have a strong foundation from which you can continue to build on further. A choice that you make in the beginning will certainly have an effect in the long term”, Sarah Stoel, Knowledge and Quality Advisor at NCOI.”
But when looking at the implementation process the following 4 points should be paid close attention to:
1. Prepare employees for the new virtual colleague
The employees who will deal with the chatbot on a daily basis, mostly the customer service employees, should be involved from the beginning. They need to know exactly how the chatbot works, which tasks it should solve and what it is allowed and not allowed to do. This not only facilitates collaboration and greater acceptance of the new virtual colleague, but is also essential for the performance of the chatbot. After all, it is the customer service staff who monitors the chatbot’s work and helps it to become better. They can optimize its answers and/or add new questions to its repertoire, as well as point out its mistakes. Remy Christiaan, Customer Experience Manager at OBI4wan says: at OBI4wan says:
“Ideally, there should be a direct line of communication between the customer service staff and the product owner or technical team responsible for the chatbot. Customer service employees who work hand in hand with the chatbot are usually the first to notice when the chatbot is not behaving as it should. It is important that this can be reported directly.”
With the evolution of innovative chatbot platforms that make it possible to develop a chatbot completely without programming knowledge you can even take this a step further and handover the development of the chatbot to customer service agents. Read more about why customer service agents are actually the perfect chatbot developers in this blog.
2. Don’t expect too much of the chatbot at first
To ensure the success of the chatbot, its tasks should be clearly defined and limited to only a few when you start out. Instead of giving the bot a broad repertoire of tasks to solve on day #1, it is better to first entrust it with a specific task or the answering of a specific question. See our blog post “Chatbots: the most interesting applications for your organisation” for the most common tasks. The amount of tasks can be increased over time. In this way the chatbot can be better monitored and improved. Remy emphasizes:
“You shouldn’t expect a chatbot to be perfect from the start. He’s more like a human employee who’s constantly getting better and better every day.”
3. Introduce the chatbot step by step
The effects of a chatbot on user behaviour can be well monitored with a soft launch. A soft launch is also recommended to eliminate possible functional errors without having too many users exposed to them. Soft launch means that the chatbot is introduced step by step, so that only a limited number of website visitors (e.g. 10%) can see it at the start. While constantly monitoring KPI’s, the number is increased until the chatbot is finally visible to 100% of visitors. An additional possibility is to first show the chatbot on one website page only, for example on the FAQ page, and then slowly implement it on other pages. When the Erasmus University in Rotterdam launched their chatbot, they also experienced the benefits of starting step by step:
“We didn’t publicise the go-live at all. The bot was first live for three weeks before we launched it through a campaign. The funny thing was that people started using the bot right away. I’m very happy we did it that way because things came up that we didn’t see during the test period. For example, the call to action was not clear enough”, says Christa van der Kruk, Marketing & Communication Advisor at Erasmus University in Rotterdam.”
4. Regular maintenance of the chatbot
The implementation of a chatbot is not a one-time thing. It needs regular maintenance to make it better – by providing it with new input, correcting its mistakes and adjusting its answers.
“Don’t let your chatbot out of your sight – especially at the beginning it needs to be constantly monitored.” says Remy.
Customer needs can also change over time. New products may be introduced or external influences, such as the season, may change the nature of the frequently asked questions. A university or technical college that receives many questions about the choice of study subject in spring may receive completely different questions in autumn, e.g. about exams.
What metrics and KPI's should be monitored to check the chatbot?
To measure the effectiveness of the chatbot, the number of conversations it has been involved with is an important indicator. Also check, which part of the conversations was successfully completed by the chatbot and which part was handed over to a human employee. In order to determine how good the chatbot already is in recognizing the intention of the customer, the number of correctly classified requests and statistics on Natural Language Processing (NPL) should be looked at. In addition to these indicators, which reflect the performance of the chatbot, the effect on customers and employees should also be examined. Has the NPS score, or any other measure of customer satisfaction, changed positively? Have waiting times become shorter? Do customer service representatives have more time to deal with complex customer cases?
What effect can you expect in the first months with a chatbot?
Within the first months, the chatbot should become a fully integrated and accepted member of the customer service team. The workload of customer service employees should go down and they should have more time to advise customers on complex issues. Both the number of questions correctly classified (intent recognition) by the chatbot and the number of questions answered by it should have increased significantly. As a rule of thumb you can expect a chatbot at this stage to correctly recognize the intent of a message in 80% of cases and to participate in about 15% of incoming messages. Be aware though that these numbers strongly depend on the type of chatbot applied, on the number of tasks he is solving and on the input that it receives from the team to become better.