We naturally think of “intelligence” as a trait belonging to individuals. We’re all—students, employees, soldiers, […]
What is the business value of a chatbot?
Bot technology is a promising development, but you have to really examine why, where and how you can apply it. There are three aspects you can use to assess added value to your chatbot: customer-centricity, quality and efficiency.
To determine the business value of bot technology, it is useful to keep all the current possibilities in the back of your mind. That’s because the technology has various manifestations. The simplest bot is scripted. It signals certain keywords or sentences in a conversation and provides pre-set answers. This bot is therefore pre-eminently suited for answering FAQs, for example.
Bots that can recognise a customer’s intention are a bit more advanced. They use machine learning to analyse the relationships between words and find out what the website visitor wants. You encounter this type of bot often: in restaurants, the retail sector, the travel industry, etc.
Virtual agents go a step further. They understand what a person is trying to achieve with the interaction and can carry on a conversation. This type of bot can retrieve data from other systems in order to obtain the right information and learns and improves over time. An example of such a bot is the virtual health care provider that a hospital in Maastricht recently introduced. The ‘Molly’ bot teaches patients with heart problems to monitor and improve their health at home, so that they only need to come to the hospital if this is really necessary.
The most complex bot is the humanoid advisor, who can reason like a human. But this is still off in the future. The development of artificial intelligence is not yet so advanced that the neural network of a human can be equalled, and it is unclear whether and when this will be the case. The first three bot types are already available and are being used more and more.
Added value in three areas
Of course, whether the technology is interesting for a company depends on the problem that needs to be solved. In practice we see that organisations can bring about improvements with bots in three aspects:
Customer interaction quality
With bot technology you can serve the customer 24/7 in a consistent and targeted manner. The answer that the customer receives to an answer does not depend on the person he or she is speaking to – bots always answer the same way. You can, however, introduce variables to the process in the form of personalisation. Then your customer profile determines how you are helped. For example, someone who has achieved a high status in an airline’s frequent flyer programme may receive a different answer to the question of what happens in the event of overbooking than someone who is not registered as a loyal customer. Of course, all of this happens within the applicable privacy laws, in other words with the customer’s permission.
If you use the bot to handle frequently asked questions or commonly occurring problems, then you can free up service desk employees for complicated issues. This way customers will not only be helped faster, the service representatives’ work will also be more interesting.
Customers leave a digital trail of their interactions. By automating conversations, companies have a better view of the questions asked and the quality of the answers. What is trending? Companies can respond to customers’ needs and detect patterns, which is interesting from the point of view of business development as well as customer satisfaction. If you, as an organisation, are able to track your customers’ digital trail in their interaction with your company, then you can always pick up the conversation where it left off. This prevents customers from becoming annoyed and feeling like they are being sent from pillar to post because they have to constantly repeat their story.
Once the business case has been generated, you need to decide which type of bot can best help you achieve your goals. You don’t need an almost-humanoid bot by definition. The problem dictates the solution, as the saying goes. A scripted bot can resolve 80% of incoming questions, and it is very possible that this will be enough if it is your objective to ease a call centre’s workload, for example. What’s more, you can see such a scripted bot as the basis to which you can add intelligence, such as interpreting the customer’s intention.
The trick here is to precisely determine what the bot can do by itself and when an employee has to intervene. A chatbot can help a customer get quite far in the sales funnel, but it may be desirable for an employee to complete the purchase or answer complex questions about it. The bot can also give recommendations, thereby being used for cross-selling and upselling. The algorithm retrieves the necessary information for this from the conversation and previous transactions. A sentiment analysis can also be performed so that the context can be better determined, based on which the bot will propose actions to the backoffice.
Which service providers?
Bot technology is maturing, and users’ acceptance is increasing. Due to this, the bot landscape is significantly splintered. Bot Framework & Deployment Platforms are the ones who are developing the infrastructure and functionalities. There are also numerous specialty companies that have designed plugins with which the bot can recognise intent or emotion, identify images or talk. But there are also all-in-one solutions that develop both the infrastructure as well as the competencies of a bot.
Then you also have to deal with channels where the bot is used, such as Facebook Messenger, Twitter and Skype. And then there are the generic platforms that are developing APIs for speech, image recognition and natural language processing that are not specialised in bots but can make a contribution to this technology.
Once you’ve assessed the problem and know what the technology can do, the next step is to determine the scope by selecting a sub-problem on which to test the bot. Instead of answering all FAQs, you can first train the bot in a certain product category, for example, so that you don’t get stuck on technical and methodology issues. Training a bot is, after all, an enormous job that costs a lot of time, data and expertise. It is emphatically not an automated process. You have to be knowledgeable in data science and artificial intelligence.
You must make sure that the foundation you choose is scalable and flexible. You start with the sub-problem by way of proof-of-concept but ultimately you want to develop something bigger. In addition, it is very important that you comply with the General Data Protection Regulation (GDPR) and securely save the personal data of bot users.
As soon as the foundation is in place, you subject the bot to comprehensive testing. People check the results so that you are sure the interaction complies with the requirements and needs of users. At the same time, you have to manage customer expectations – like telling customers that they are talking to a bot who is not yet very intelligent.
This is a good strategy to prevent being dragged along by the hype and instead quickly achieve results and determine whether bot technology really has added value for your organisation and the users.