ICMI is part of the Informa Tech Division of Informa PLC

This site is operated by a business or businesses owned by Informa PLC and all copyright resides with them. Informa PLC's registered office is 5 Howick Place, London SW1P 1WG. Registered in England and Wales. Number 8860726.


How to Develop A Successful Customer Service Bot Strategy - Part 2

AIThis article first appeared on HDI

In the first part of this guide on how to implement customer service bots within an omnichannel experience, we discussed how to set parameters for what you want the bots to accomplish and identify how the bots are needed. In the final part of this series, we discuss how to measure bot interactions and use customer feedback to build a better self-service experience. Here are the final steps in this 12-step guide:

Step 7: Define Measurement Framework

While work to develop a bot is significant, your organization must also develop a measurement framework that includes measuring the success of the bot and measuring the success of the customer journey. A bot is just another channel of support that must be measured for its success in serving customers. However, developing a set of channel-based KPIs only gives a partial indication of success. A measurement framework must focus on understanding the success of the overall customer journey.

An important bot metric is based upon a successful outcome or completion rate. Traditional transaction type metrics in other channels will not be appropriate. For example, AHT (average handle time), MTTR (mean time to resolve), will not apply to a channel that can handle many different transactions at one time and where the transaction can take as long as needed for the customer to get the desired result. Tracking the rate of successful outcomes is an excellent start in understanding customer success.

Another metric to consider is the bounce rate or fallout rate. This metric tracks how often a customer bails from a chat with a bot to use another channel, or the customer is moved over to an exception path where human intervention is required. To understand customer adoption, another metric to consider is reuse rate or how often a customer returns to using the bot again after a successful outcome or result.

Remember, customers are likely to pick the path of least resistance, and this often means starting in one channel but may involve multiple channels until a successful conclusion is achieved. A customer may start in a self-service portal but then may use the bot and eventually need to call in to speak to a live agent. It is essential to understand why a visit to a website was not successful and what ultimately led to a call to a live agent. By studying the customer journey with click-stream analysis or digital tracing, the support organization can better understand the bot’s effectiveness in serving the customer, and exceptions can be used to improve service across all channels.

Step 8: Build Intents/Responses

The next step is to build the bot by developing the intents, responses, and customer contexts based upon the defined use cases. Each use case should map to a unique intent. For example, ordering a computer is a common request. The use case—how a computer is ordered today within the request fulfillment process—is used to build the underlying conversation that helps the customer to accomplish the outcome efficiently.

In this step, it is critical to have an in-depth understanding of how the conversation engine works in your chosen platform. Each intent or customer outcome will have questions, responses to questions, and also entities—facts or pieces of information that are needed to understand the specifics of the outcome.

For example, does the customer want a laptop, desktop, or mobile device? What operating system is required? The entities are built into additional follow-up intents. The input of one intent into another intent is an integral part of building a realistic conversation. Data from existing channels can provide significant visibility into how those conversations take place in existing channels and will provide breadth in adding multiple training phrases to the bot's initial intent development. It will also be important to identify points in the customer journey where the customer will fallout of the interaction and work with a staff member, an exception occurs, or when the desired outcome is reached. The journey maps, use cases, and data analysis of existing knowledge will support the development of well-defined intents and response.

Each of the intents requires testing to ensure that all paths work as intended, and any integration with systems outside of the bot platform is working properly.

Step 9: Crowd-Source Bot Learning

The bot will have a lower degree of accuracy and, thus, a lower containment rate at the beginning of the adoption. To ensure that customers are not subjected to the steep learning curve, additional testing by staff and select customers will improve the artificial intelligence and learning of the bot before launching to customers. The process is not entirely manual. Using the natural language understanding, the bot will learn to handle new phrases and map them to the existing intents. Alternatively, additional paths or intents may be identified and required before use directly by customers.

If possible, a short period of testing by a broad customer base will provide a lot of additional data to help the bot learn. Remember, it isn’t just about helping the bot to learn. The staff also need to practice their new responsibilities of monitoring the progress of the bot’s learning, identifying exceptions, and measuring its success.

Step 10: Use in Assisted Service

The final step prior to rolling out the bot to the customer is to use the bot in assisted support. What is the containment rate when used by a staff member when walking the customer through the intent in a different support channel such as phone or chat support? The goal is to improve the design of existing intents to ensure the results are being delivered successfully before launching to customers. Here is where the final tweaking occurs, and staff is supporting the bot development and practicing the skills needed to improve the bot continually.

Step 11: Launch to Customers

With the above-outlined level of planning, the organization can, with greater confidence, launch the new channel. Measurement is vital to understanding customer adoption, completion rates, fallout rates, and the impact on support volumes in other channels of support. Be sure to market the new service and celebrate the successful journey of launching a new service!

Step 12: Multi-Channel Continuous Improvement

While substantial work is required to get to a bot launch, remember that the work has just begun. To improve the bot, the staff need to identify additional intents, gather data, and repeat the process to increase the scope of the bot gradually over time. But continuous improvement should not focus on just the bot. Remember that a customer journey may cross many different channels. The organization must ensure that the underlying processes are consistent across channels, that the knowledge of intents and responses does not vary depending on the channel of support.

Keep Learning and Improving

As we wrap up our discussion on developing an effective chatbot strategy, it is important to note that artificial intelligence is an exciting, emerging field that can help us not only serve customers in new ways, but also provide incredible insight into the customer experience. The implementation process of the chatbot is relatively easy, and the tools have well-documented implementation guides. While adoption is easy, it will be essential to establish a strategy that will allow the organization to experiment with this emerging channel, learn from the experience, and develop a more comprehensive approach over time.