Chatty chatbots and the ‘time to frustration’

Chatbots have been getting a lot of attention recently through a combination of trends: advances in natural language processing (NLP), a resurgence of chat and chat channels (think Facebook Messenger, Slack, Hipchat etc), ubiquitous mobility and advances artificial intelligence — including the emergence of A.I. systems that are beginning to understand the intent of spoken or written words. In addition, consumer and user familiarity with this low-hassle, unintrusive channel makes chat a preferred mechanism for several interaction scenarios.

As enterprises look to capitalize on chat, an option that offers connectivity and always-on communication and interaction channels, they have to be aware of the maturity of the channel and balance that with the need to maintain the quality of customer interactions and experience. This is hard to do, and it can conflict with the allure of a cheaper-to-maintain and scale-out customer experience and interactivity channel.

Chatty chatbots

Non-mature chat systems tend to be too chatty and can be detected quite easily. A chatty chatbot makes it harder for users to get the service they need. A chatty chatbot is unable to understand the user’s intent — or it may interpret the user’s intent incorrectly, leading to a repetitive and frustrating interaction.

‘Time to frustration’

“Time to frustration” becomes a key metric for designing useful and delightful chatbots. This is the measure of the time it takes for users to reach a point of frustration that turns them off from the chatbot — and, possibly, from the company’s product or service. Measuring time to frustration requires mechanisms that detect changes in the usage and interaction of a chatbot by a user in addition to post-chatbot usage activity and engagement.

Frustration can be detected, at a high level, by how a user reacts to the chatbot’s response. If a user resubmits his or her inquiry with slight changes, retries the same request or switches the channel of communication, it is a good sign of frustration. In fact, techniques used by search service providers to understand user queries and determine intent (and failure to do so) have a lot of relevance in the world of chatbots. This also means that excelling in understanding intent and responding to users’ needs with high quality requires a trove of good, highly curated content, interaction data that enables the chatbot to understand intent, and feedback loop mechanisms to measure and improve the quality of the chatbot.

Building chatbots that are neither chatty nor frustrating

To develop a successful chatbot strategy and build useful chatbots, enterprises should keep the following six tips in mind:

1. Determine the segment of users to target with a chatbot

Enterprises should expect that different segments of users will react to and adopt chatbots at different rates. This is critical, because offering chatbots in interaction scenarios that require high touch, highly personalized service such as financial advice for high-net-worth clients, can be detrimental to the experience and the quality of service. Enterprises also need to consider the familiarity of user segments with the chat channel and the suitability of their specific needs to be serviced through a chat channel. In short, will the users be targeted with a chatbot service show up on the chat channel?

2. Determine the type of chatbot required

Be sure to choose the right chatbot for the given purpose. Here are some examples:

• Transactional chatbots enable user transactions to be performed, such as ordering a cab or purchasing a product.
• Information-retrieval chatbots enable search and retrieval of relevant information, such as weather forecasts, movie times etc.
• Automation chatbots enable a series of actions in a predefined operational workflow to be triggered and completed automatically. Examples include planning trips or making arrangements for leisure activities (buying tickets, scheduling transportation etc.).
• Aggregation chatbots enable the automated collection, assimilation and reporting of information, such as building customer activity reports or automated NPS scores of the chat channel.

3. Determine if the conditions are right for a conversational chatbot

Conversational chatbots are much harder to build compared to transactional, aggregation, information-retrieval or automation chatbots. Natural language processing and A.I. have a long way to go before machines can understand natural language. For enterprises looking to get started, focusing on nonconversational bots not only provides a shorter time to market with this capability but can also begin generating highly valuable data relevant to training the conversational chatbot.

4. Build a knowledge base

Enterprises should ensure that their data and content is of high quality, curated and organized appropriately for algorithmic lookup and processing. In addition, the enterprise should mine all customer interactions across every other channel to determine the categories and forms of user requests and associated high-quality or low-quality responses. Depending on the types of chatbots that are required, enterprises should determine the transactions, automation, aggregation or information-retrieval scenarios and expose them as services that can be triggered through APIs or other similar mechanisms.

5. Build a user intent competency

Enterprises should analyze and categorize the intent of their users as evidenced from past interactions in other channels and the language that is used to express the associated intent. This can then be used to automatically extract intent from user-submitted text or audio.

6. Make the bots easy to find and invoke

Chatbots need to be easy to discover and invoke. Because there is an implication of privacy and data-sharing, explicit consent might often be required. In addition, the easy discovery of the appropriate chatbot driven through personalized recommendations and easy search (similar to the discovery capability introduced in Facebook Messenger based on user feedback that showed that users were having a hard time finding bots).

It’s a good time to build chatbots

With the conditions ripe and ready ,and a user base that is increasingly familiar with chat channels and chatbots, it’s a good time to explore this channel. Chatbots can offer cheaper, nonintrusive and often more satisfying and instant interaction channels for users, making their experience with your enterprise more delightful and satisfying.