Creating Effective AI Chatbots

A chatbot in its simplest form is a way for a human to ask questions of a computer. Often the chatbot exists on a website for customers but might also be on internal company web pages for employees. Chatbots can also work over mobile apps, mobile phone text messages or even email. They can also exist as interactive voice responders (IVRs) with speech recognition and speech synthesis.

The simplest chatbots are guided interactions where the user selects from a list of options. The chatbot then presents further options to guide the user’s journey towards the information that will answer their question.

More sophisticated chatbots interpret free-form text entered by the user. AI techniques can parse and understand the question. The responses are generated in a conversational style in the desired tone. This allows a much more natural interaction between the user and the computer.

Artificial intelligence allows a much more immersive experience than was possible for simple web chatbots and IVRs. Responses can be influenced by the whole conversation thread rather than the single most recent message. Information provided by the chatbot can be generated from a much larger body of knowledge than was possible until recently. The understanding of questions and the more flowing nature of the conversations can feel like magic and uncannily realistic.

A well-trained chatbot can deliver great experiences for potential customers, existing customers and internal employees.

Customer and end-user chatbots

Pre-sales chatbots can be embedded into a customer-facing website to handle pre-sales enquiries and answer frequently asked questions. Chatbots can provide a better experience than typical search features found on websites, though can be integrated with them. When implemented successfully, they can improve conversion rates and reduce costs.

Post-sales chatbots are useful for front-line support to end users. They can be used for initial customer service enquiries and support queries. They can reduce the burden on first-level customer support teams and reduce the time to deliver the information needed by customers. They can also be used to gather the information necessary before escalating to more senior support staff, such as product information and customer contact details. When the chatbots are trained with enough support documentation from a technical knowledge base, they can often provide all the information required by the customer without further human involvement.

Internal and employee chatbots

HR chatbots can be used to reduce the burden on HR teams. Chatbots can be used as part of the recruitment process, for example for a first round of candidate screening questions to ensure they are broadly suitable for the role, meet the right-to-work requirements etc.

Chatbots can also support the employee onboarding process. For example, they can be used to manage the question and answer flow during the identity verification process. They can also be used to answer questions from new starters such as benefit entitlement and registration.

Ongoing employee support can be delivered by HR chatbots to reduce the burden on HR call centre teams. The chatbots can be provided with frequently asked questions for employees. The natural language features of chatbots allow users to ask questions in the way that makes the most sense to them.

• Good training data.

  A great chatbot is only as good as the information it has available to it. The better the training data, the better the responses it can give. Create a knowledge base that has the most relevant content. The AI can select the most relevant parts according to the questions it receives. It can even cross-reference multiple sources to create a combined answer.

• Adequate infrastructure.

  Make sure that your server-side infrastructure can handle the number of requests it receives. Modern cloud platforms can scale up at times of peak demand, and then back down again to keep costs down. The more computing power available to the chatbot engine, the more responsive the conversations will be.

• Artificial intelligence.

  Consider how AI can make your chatbots more effective. AI can enhance your chatbot's understanding of natural language and provide more human-like responses. It can create responses that build on answers to previous questions, creating more natural flowing conversations. AI can also be used to efficiently work with much larger bodies of training data than would otherwise be possible.

• Deploy your chatbot to different channels.

  Make your chatbot available where your users will find it. Chatbots work well on public-facing websites, but they can also work well in mobile apps, text messages, and emails. They can also work with ordinary voice calls, utilising speech recognition and speed synthesis. The same chatbot engine (with the associated knowledge base and rules engine) can be used across multiple delivery channels.

• Integrations

  Behind the conversation, the chatbot engine is more than just text analysis and generation. Chatbots will be trained (or grounded) with a knowledge base of documents and information.

  The most effective chatbots will also connect to other systems. This allows a chatbot engine to fetch up-to-date information from different employee and customer databases. The chatbot becomes an easy-to-use interface across systems that would otherwise be difficult to communicate with. Many business systems have APIs and SDKs that chatbots can connect to.

• Personalised experiences.

  Users will find the chatbots most effective when they contain content that is relevant to them. Chatbots for customer support and employee information can be provided information about the user they’re talking to. This can guide the chatbot to support material for previous orders, the status of refund claims or employee absence allowances. Personalised responses will make the chatbots more engaging.

• Ongoing improvement.

  The best chatbots learn from their previous experiences. Continuous monitoring and feedback can drive improvement. Positive and negative feedback reinforces the behaviour you want to see. It is often helpful for chatbots to ask whether the user whether their problem has been resolved. Users can also be asked how well the chatbot has performed. More detailed analysis of the conversations can also offer insights into performance.