Chatbots are the new apps; every brand worth its salt has one. Tommy Hilfiger had TMY.GRL launch a Gigi Hadid collaboration. The Singapore government has GovBot, a bot that allows citizens to contact civil servants and report concerns of public interest. Almost every week, a retailer rolls out a recommendation or shopping chatbot.
Chatbots are on the ascendant and arguably at the exuberant peak of the Gartner Hype Cycle which models the maturity of a technology and its lifecycle. And for good reason too; Facebook has made it easy for any brand to build its own chatbot through its Messenger Platform. What’s more, the use of messenger apps (on which chatbots thrive) has surpassed that of social media.
But try interacting with a chatbot, and you will quickly realise a simple fact: Chatbots cannot sustain a real conversation. Chatbots are perennially question and answer based. Ask a question or provide a trigger, and the chatbot gives an answer. Often, conversations do not extend beyond this scripted give-and-take. The chatbot is also not inherently curious or interested (some might seem so at the beginning, but only to collect inputs for what seems like a search-like query).
Maybe we are disappointed because we are anthropomorphising the chatbot. The conversational manner of the user interface lulls us into imagining we are talking with someone. We expect it to display empathy. Or maybe, regardless of how the chatbot is designed, we want to break them to test the limits of their ability to sustain a human façade (as famously happened with an experiment by Microsoft).
Conversation or not, it’s in the design
The truth is, most chatbots are just not built for conversations. Early chatbot interactions were designed based on rules and decision trees. In fact, these chatbot designs borrowed heavily from call centre scripts to handle simple enquiry-based interactions. The user interfaces will often feature menus and selectable options as opposed to free-text based interactions. The chatbot understands a specific command and nothing more.
To make a chatbot conversational, more technology needs to be incorporated into its design. The chatbot needs to understand language using a Natural Language Processing system (NLP). This gives it the ability to recognise keywords and ad hoc phrases in natural conversations.
The chatbot will also need to learn from experience and get smarter over time. This requires advanced machine learning and Artificial Intelligence (AI). Over time, an AI-powered chatbot that is exposed to conversations with real people is able to learn which of its answers elicit the best response and continue to hone the answers it provides.
Some chatbot designs take it further by understanding complex queries, maintaining context while in a conversation, or analysing sentiment and tone to optimise responses. These chatbots demonstrate what we call cognitive intelligence and often need sophisticated AI, such as IBM’s Watson. Recently, Facebook has started taking steps in this direction with the design of a chatbot training ground known as ParlAI.
It’s often unnecessary for a brand’s chatbot to take it this far. As a brand, it is not about making a chatbot to pass the Turing test—an assessment of a machine’s ability to exhibit intelligent behaviour. Functionality takes precedence over conversation. Chatbots need to be effective at conversation commerce, rather than making human conversation comfortable.
Working a chatbot
So what’s a brand to do in building its own chatbot? Here are three considerations to think through:
- Design the chatbot to serve a purpose: A retailer will likely create a chatbot to help people purchase goods. A telco might create one to deal with customer service. Knowing the purpose of the chatbot means knowing which conversation threads are essential, and which might be extraneous to the design. In our experience, chatbots can be designed for four purposes: content distribution, customer service, recommendation, and transaction.
Approach it as one would a product design project: In a chatbot, content is what the user interfaces and engages with. Content design entails knowing the product opportunities and limitations that can or cannot be resolved by the chatbot.
For example, a logistics company helping customers to track deliveries will need to provide access to real-time package-location data. If this is not technically feasible for a chatbot, it should not be designed within the interface.
- Have a beginning, a middle and an end: A purpose-driven chatbot will guide a user through to an intended end goal (conduct a transaction, serve a piece of content). Consider the journey a user will take when they encounter the bot, how the customer will move into an interaction and out of it. Off-script conversations are also inevitable, and good chatbot designs build in paths to navigate such conversations back on-track.
With the chatbot, we are quickly moving past a period of ‘inflated expectations’ and entering a stage of disillusionment over its conversation ability. Before you jump on the bandwagon, start with knowing whether conversation is essential to the chatbot you want to build first. It will save you from a lot of pain.
Yeong Yee is senior consultant with Ogilvy & Mather Singapore.