Chatbots can tell us more than we think.
Many developers and data engineers pour scorn on the buzz around chatbots. They might have played with them and because they can get a quick, isolated outcome up and running, they write them off as simple gimmicks.
I often compare building a chatbot with building a website. Once a website goes beyond a few static pages, there is a breadth of skills required to ensure a good digital experience. For instance, when a website expands from a few static pages to multiple pieces of dynamically served content within a content management solution, or a secure, multi-product, digital banking platform, it needs a lot more love and attention.
The broader team at Servian has created a number of chatbots recently using Google’s Dialogflow. We know that well-developed chatbots can act as a concierge, interacting with people through text (and increasingly through voice) to provide assistance in a conversational context — which, in many ways, is a more natural experience than using a mouse for a point and click interface. Chatbots provide answers to simple questions quickly, handing off to a human when more complex help is required.
At the very basic level, a simple chatbot uses language understanding to classify the input (text or voice command), interpret the user’s intent and then generate an effective response. However, making a well-constructed, relevant and (dare I say) engaging chatbot requires all the related disciplines (from data analytics to software engineering) to be in place, which is what I believe makes them a great organisational litmus test for machine learning (ML) and artificial intelligence (AI) readiness.
Nick Harrison and Deborah O’Neill’s article “If Your Company Isn’t Good at Analytics, It’s Not Ready for AI” articulates this perfectly. It explains why companies need to have strong basic analytics in order to harness the power of AI. It states that companies should automate repetitive processes with substantial data and then centralise this information so that, when machine learning or artificial intelligence is applied, it is able to draw off all relevant, correct and up-to-date information.
Harrison & Oneill use the example of a retail category manager to highlight their point. With basic data analytics, retail category managers can see a complete picture of historic customer data; which products were popular with which customers; what sold where; which products customers switched between; and to which they remained loyal. Only with this data already automated can a company layer artificial intelligence to predict customer behaviour.
The same is true for building a chatbot. Efficient analytics must be in place before adopting AI for the AI to work to its optimal potential. But that’s not all that needs to be in place.
To have a well-functioning chatbot, you really need to have:
- The ability to construct workshops and undertake the conversation design to develop an appropriate tone of voice and conversation flow
- The data analysis skills to develop, analyse and refine the conversation flow
- The front-end skills to make a browser or mobile interface engaging
- The software engineering skills to manage the context of rules/intents in the bot
- The software engineering skills to develop the integration points and ensure they perform at a level which provides feedback for a seamless conversation
All of the above capabilities are essential for ML and AI. Once you have these in place, you can then consider:
- How will a business process interact with the ML/AI capability — on both inputs and outputs?
- Do you have the data analysis and data engineering skills to feed the models and then take action with the outputs?
- Do you have the software engineering skills to support data and models evolving?
- Can the end-to-end solution iterate based on continuous testing and user trials?
A set of talented data engineers and data scientists who understand feature engineering and data science processes with commercial acumen can only go so far in isolation. There are many companies who have hired or developed data science rock stars to find out that what they need is an ecosystem around those rock stars to be successful. Developing a proficient chatbot can provide insight into where the gaps lie.
Instead of thinking of chatbots as a way to help users get answers quickly in their digital channel of choice, I challenge you to reframe your perspective and begin to consider a chatbot in regards to being a litmus test for a data science operating model. If your chatbot says you’re AI-ready, maybe you can start the serious stuff with the Data Scientists.