Conversational banking proved to be one of the most complex projects I managed to date, with highly technical and analytical process at the fore front of day to day activities, including:
Multichannel data analysis
Tone of voice analysis
The image below is an example of the chatbot I launched while at HSBC.
With many different implementations of chatbots located across international regions, I led the way in defining global standards for measuring chatbot performance by creating guidelines that provided a clear, consistent and accurate method for defining how to measure the effectiveness of chatbots.
Below are extracts from the paper I wrote on measuring chatbot performance.
First contact resolution (FCR) is often measured incorrectly with both humans and AI as it directly assumes that from this experience that the problem the customer discussed is resolved without doubt. It does not take into account that the customer may have immediately called to speak with an agent or visit a branch to resolve the issue.
Average handling time (AHT) is used frequently as a measure of productivity in business but a high AHT does not directly correlate with a drop in productivity as AHT it is dependent on may factors such as lengthy delays between responses, complex queries or users having small talk.
In addition to this, comparing humans with bots is not recommended one obvious reason is that a bot can deliver can process and deliver answer in milliseconds while a human by nature is more conversational and continually assesses the chat session altering the tonal response applying humour and empathy making the conversation by nature longer than compared with a bot.
Containment rate (CR) is widely used as an indicator of cost saving but quite often it does not take into account a number of scenarios that...