Chatbot troubleshooter: How to tackle the most common obstaclesby
In the latest in our new series where technology leaders tackle the challenges most commonly reported by adopters of their category of tools, Paul White, director of customer engagement at MPL Systems, shares his advice on how to overcome chatbot challenges.
1. “There are so many different types of chatbot! How can we identify the best one for our business?"
PW. Exactly what functionality each organisation needs from their chatbot technology is dependent on their customers. Understanding what they want from your service is key and being able to apply this through your software will keep you ahead of your competitors.
Do you want your chatbot to answer a whole array of questions like Siri or Alexa? If so, you need to ask yourself how broad you want these questions to be able to go. The more you want it to answer, the more data training you’ll have to ensure the software goes through. It is no good expecting the chatbot to be able to do everything from day one; which is often the mistake that many companies make; it is better to allow time for machine learning that will lead to a more coherent quality in answers further down the line.
The basic offering of a chatbot is to answer FAQs. Considered to be the bread and butter functionality of the software, it is imperative that your chatbot is able to search and filter through your data in order to be able to provide popular simple answers. Is it worth starting with a narrower coverage for the chatbot, identifying the most popular questions and allowing it to answer those, while routing more complex and irregular requests to agents? If this is a route that you decide to begin with, it is important to offer other self-service options so that you can collect data and use it to learn from and further develop you chatbot.
To be able to go one step further, by processing change requests from the customer, would positively affect your agents at the service desk significantly. For example, should they wish to change their address on their account, change details on a submitted insurance form or upgrade their membership status, the chatbot would be able to acknowledge, understand and process this request. All without disrupting the agent’s workflow, who might be completing a more complicated task at the time.
Finally, in order to provide a seamless customer journey, it is important for your chatbot to be able to hand over to an agent when/if necessary. If the customer has entered text that the system does not recognise, or has a more complex query, being able to hand off to an agent, with all previous chat history, is essential to providing a smooth experience whereby the customer needn’t repeat any information already disclosed.
How much functionality you wish your chatbot to have fully depends on the purpose you wish it to serve.
If used strategically, chatbots can be leveraged by the company to optimise agent efficiency, eliminate customer wait time and drive higher CSAT ratings, as customers expecting immediate, round the clock responses, are well satisfied.
As mentioned, how much functionality you wish your chatbot to have, fully depends on the purpose you wish it to serve. I would advise that whichever vendor you do choose to work with though, is able to develop the functionality of the software so that it evolves with the company’s scalability and your customers’ requirements.
2. “We want the chatbot to be engaging and friendly, but not creepy. How can we strike the right tone for our customers?"
PW. Through Natural Language Processing (NLP), the system learns to recognise real language. This engagement with language allows the software to utilise data to provide customers with personalised information. The software is also able to analyse the customer’s type tone, allowing it to interact and engage conversationally.
In many cases, customers have felt that they are receiving correspondence from a real person, rather than a robotic software. With this more colloquial tone, the software will extend the brand into an interactive experience for consumers.
The double benefit of this for the company, is that transcripts of conversations can be easily managed and analysed. Providing further customer insights, the software allows the company to understand what type of customers are visiting their website; what they are looking for and recurring issues that need resolving.
When answering requests, chatbots pull data and terminology from your company’s knowledgebase. These often have a factual tone and so this comes out through the chat messages. It may be worth asking yourself if you want to rewrite the knowledgebase to have a friendlier tone, or alternatively, incorporate slightly more colloquial data, e.g.: the company handbook or ideally answers that agents have previously provided customers with, to keep the tone friendlier.
3. “The bot doesn’t understand some basic commands and requests. How can we ensure the chatbot is intuitive?"
PW. The best thing to do is to start with a narrow coverage so that the few requests that the chatbot can deal with, it can deal with well. Following this, we would advise to have a way to hand off complex and broader requests to agents in an efficient manner so that the customer doesn’t feel any disappointment or frustration.
An essential skill is to listen well. A chatbot’s quality depends on its ability to make meaning of human language in written and spoken form – the quality of its natural language processing (NLP) technology.
With machine learning, the software will be able to continually learn from experiences. Further developing with each interaction, the ability will continue to advance and be able to recognise simple and more complex commands and requests.
4. “Customers aren’t using our chatbot, or are abandoning the interaction before their problem is solved. How can we ensure that we're using chatbots for the right interactions in the customer's journey?"
PW. Despite NLP and machine learning qualities, chatbots will not converse with humans the way an agent would for some time yet. The industry is accepting this more and more for now and so it is important to understand that its ability is also limited.
Simple queries are the best, as any question a chatbot can answer is monotonous work for a support professional. Therefore, it is most advisable to keep simple questions and change requests for the chatbot, making the customer immediately aware, if they have posed a more complex query, that the chat has been passed to a live agent, who will also be able to call them if they so wish.
This ensures that the most appropriate option for each customer type is presented within a matter of seconds, rather than them being passed from pillar to post within the customer service department.
Neil Davey is the managing editor of MyCustomer. An experienced business journalist and editor, Neil has worked on a variety of newspapers, magazines and websites over the past 20 years, including Internet Works, CXO magazine and Business Management. He joined MyCustomer in 2007.