As with any new recruit, chatbots work better if they are trained before interacting with customers. The larger and more complex the bot, the more training is needed, and the greater amount of business data required to ensure the chatbot does a good job.
Feeding good data to your bot will improve its responses and approval rating, and is a key consideration before falling for the chatbot and AI hype. Chatbots might be simple to create and launch, but a good chatbot requires a fair amount of effort on the part of the business. How much? Check out this piece discussing Verizon’s in-house customer support bot on My Verizon.
‘My Verizon engineers did the initial development with months of chatbot training. Today, a team of 50 people maintain the bot with a team of computational linguists monitoring conversations for what Verizon calls “fall-out”: words and expressions the company chatbot doesn’t yet understand.’
Basic Chatbot Training
That’s the level of effort required for a global enterprise-class chatbot. While your business might not have such globe-spanning demands, good training is still a key part of the process.
The bot can be created fairly rapidly, and training it, can take many forms. For a simple script-based bot, testers can feed in all the responses they can come up with to see what happens.
But that runs only as fast as your users can type. Data-based training speeds this up dramatically. Using sets of business data, they will help validate it and find weaknesses before a customer comes across them.
Your business should have much of that data, but much it can be stored in unstructured forms. That might be through customer service transcripts, message logs or other types. Getting that into the right format for the chatbot will be a first step, then running the training through the responses to see what happens.
Chatbots handle data and responses in different ways, scripted bots can only offer a limited set of functions or responses. The use of machine learning, natural language processing and other AI tools helps bots develop a growing set of knowledge and understanding.
AI-powered bots can be trained by studying live conversations with agents or existing bots, and learn from these over time to suggest new responses that can be approved by bot owners to improve the conversation. As with Verizon, smarter bots can also report new words or phrases that may be germane to a conversation, while a scripted bot can only report any non-standard phrase.
How Bots Learn
Machine learning bots can be provided with those sets of data. If your business has none, perhaps as a startup, then it is possible to buy data in for training purposes.
SnatchBot’s NLP, for example, functions with a declarative approach to intent and entity recognition. In this case, it uses large numbers of example sentences that show the bot what terms are important in the conversation and what users want to achieve. If there are problems, a ‘supervised machine learning’ approach allows builders to add new sentence examples manually to help the bot learn.
Once up and running, analytics from the conversations allow bot builders to spot where the bot has difficulties correctly analyzing sentences and, in addition to the automated machine learning that takes place for their bots.
As bots become smarter, they can use techniques like sentiment analysis, using NLP to examine the language used. This allows them to establish a user is cross, unhappy or upset (and can learn what to do that will make them happy), or perhaps confused by the questions (and lay out a simpler set of options.)
As smart bots gain more knowledge over time, they can expand the range of features they offer the client or customer. This will make them valuable to both the business and customer by saving time, or doing more work. That frees workers up for more important tasks, and saves revenue. And, as AI becomes smarter, chatbots will be able to continuously learn and improve the service they offer.
Whatever bot your business needs, training and learning will take a lot of the strain in making it fit for purpose and give it a greater value.
Chris Knight writes about where technology will take us next, from the power of neural networks, artificial intelligence and chatbots, to the endless worlds promised by augmented and virtual reality. From the latest in gadgets and hardware to how digital businesses can use technology to grow, Chris makes the future clear and understandable to all.