How to 'crawl, walk, run' to improve the customer experience with chatbots

Industry professionals at EY break down the 'crawl, walk, run' approach to implementing chatbots.

(Photo: Shutterstock)

In today’s digitally-driven insurance industry, customer experience has emerged as the most important source of competitive differentiation. In the race to develop digital capabilities that can improve the customer experience at a large scale, one of the most promising new technologies has been the automated chatbot.

Chatbots have the potential to automate a wide range of inquiries and service requests, which improves the customer experience while lowering the cost to the insurer for servicing a large customer base.

Chatbots have already been embedded in consumer products that provide voice control, and numerous industries are making chatbots available throughout their customer touchpoints. These benefits have widespread applicability throughout the insurance business, whether in policy origination, servicing, or claims.

Yet insurance companies should not rush ahead with chatbots before mapping out a staged progression that defines movement toward a desired future state for the customer experience.

Lessons learned from carriers who have deployed chatbots include:

  1. Moving too quickly may lead to clumsy and immature implementations and failure and subsequent dismissal of technology.
  2. Pairing Natural Language Processing with advanced text analytics.
  3. Don’t leave humans out of the process.

To best leverage these lessons learned, insurers should take a “crawl, walk, run” approach to implementing chatbots.

Related: Property claims, customer satisfaction, and good business

Crawl: knowledge bot

A knowledge bot delivers the benefits of a search engine, focused on a given domain. By starting here, insurers can gain hands-on experience with the capabilities of chatbots and then slowly expand.

To crawl:

  1. Curate a content library. Analyze transcripts of call logs to determine what knowledge to include. The library should span a sufficiently broad range of subjects to respond to the range of searches and situations a bot may encounter.
  2. Build a “frequently asked questions” (FAQ) bot. Drawing upon information in the content library, deploy an “FAQ” bot that can answer simple customer queries. For anything that cannot be answered by the bot directly, provide relevant links to curated content.
  3. Use bots for targeted self-service opportunities. Enable customers or agents to use natural language queries to search through product documentation and policy manuals, which are notoriously difficult to navigate through traditional libraries.
  4. Design intuitive conversations. Predict the various ways that people may ask the same underlying question, and script separate responses to make bot interactions seem more natural.
  5. Pilot bot with a limited group. Test and learn from the experience of pilot-test users before rolling out to a wider audience.
  6. Measure and assess outcomes. Based on the results of each interaction, actively update the knowledge base to remedy missing or unclear information. There’s no shortcut for replaying actual conversations.
  7. Collect user feedback. Follow up bot interactions with quick surveys to determine what is working and to find out what other knowledge areas should be covered by the bot.
  8. Keep bot up-to-date. As the business changes with new products and processes, organizational change management processes should make sure that the bot stays in sync.

Related: How artificial intelligence will affect the P&C insurance industry

Walk: human-chatbot collaboration

Human agents and chatbots can work in coordination to serve customer needs. Whether it’s a human-aided chatbot or a chatbot-aided human, the combination of human and chatbot can improve customer interactions.

To walk:

  1. Digitize desktop procedures. By digitizing and automating the top desktop procedures, organizations can improve outcomes for 80% of their call volumes. Codify the steps required for each standard interaction, and then evaluate each step for partial automation with chatbots.
  2. Human-aided chatbot. Let a human manage a queue of chatbot conversations with customers. When a chatbot has reached its limit, immediately patch in a human, who avoids having to escalate to a phone call.
  3. Chatbot-aided human. Provide human customer service agents with chatbots that understand content and context. This is aided by robotic desktop automation, which can automatically look up info, populate forms and launch procedures during the discussion.
  4. Be context-sensitive. Ask clarifying questions with an understanding of the expected prior knowledge for any given conversation. In other words, don’t have a bot asking questions that the customer spent 30 minutes describing earlier.
  5. Monitor bot metrics. Track bot usage, which shows the extent to which new and returning customers are using bots, as well as conversation success rate, the proportion of conversations that lead to successful outcomes rather than frustrated kick-outs to a human agent.
  6. Conduct usability testing. Thoroughly test human-chatbot interactions by including a limited number of people from the target audience. While internal team members should also be heavily involved in usability testing, they should not be solely responsible for determining whether the bot hits the mark.

Run: chatbot conversations with customers

Once a sufficient knowledge base has been built with knowledge bots and success has been achieved in automating desktop procedures, you may be ready to “run” with chatbots that speak with customers without human oversight.

To run:

  1. Build on prior bots. As more bots are deployed to cover different use cases, use the knowledge gained and the assets developed with earlier bots.
  2. Use speech recognition/transcription for well-defined areas. Minimize difficulties by limiting scope. Start with well-defined, non-ambiguous areas such as requesting new quotes for an existing policy, ordering extra insurance cards or making modifications such as address changes. More complex processes can be added later.
  3. Focus on insurance-specific capabilities. The tech companies’ voice services are exciting, not because of the voice aspect itself but because users can easily make purchases, access media and locate information. Insurers should focus on equally compelling use cases based on their own capabilities and offerings.
  4. Activate “self-learning” capabilities. Configure bots to automatically incorporate new data, forming new connections between questions and relevant results. Over time, the bot should improve its ability to locate relevant information.

No matter what stage of adoption an insurer is at with chatbot deployment, the “crawl, walk, run” approach can help customers have high-quality, low-frustration experiences with their carriers.

Related: 4 emerging risks in the world of artificial intelligence

Shekhar Mahadevan, Executive Director, Ernst & Young LLP, can be reached at shekhar.mahadevan@ey.com. Jeff Scheerhorn, Executive Director, Ernst & Young LLP, can be reached at jeff.scheerhorn@ey.com.