A little while ago I recalled a former agent telling me how excited she was about her new role as a chat agent. She later went on to tell me about this story she had using chat solutions.
When a member logged in to the company’s portal with their assigned credentials they saw details about their medical claims, benefits, etc.
Members can also contact customer service through chat, but a chatbot assists them before agents can be assigned.
The chatbot lists several examples of questions it could help answer. If a customer’s question was challenging to answer, the chatbot would assign a chat agent to the customer’s inquiry.
In this incident, the chatbot assigned my former agent to the customer’s inquiry. A minute into the interaction, the agent identified that she needed the customer to provide a few pieces of sensitive information. However, the customer was reluctant to do so. They thought she was still chatting with the chatbot and not to my former agent.
I later found out that even though the chatbot mentioned connecting the customer to an agent and provided the agent’s name, the customer felt it was a chatbot asking for her personal information, which caused the customer to log off the session and call customer assistance to get her questions answered.
Deploying Chatbots Right
A few months ago, I decided to develop a chatbot for the call center team. I used Power Virtual agents, which are available on MS Teams. I learned everything about it and created a few conversations or flows.
Before getting into the actual developmental stage, I created a flowchart of the top questions and broke it down by each line that the call center took. I also made power automated flows that retrieved information from external sources.
However, as I was building the bot, and from my personal experiences with chatbots, here are a few tips I learned:
- Keep it simple. Chatbots can help, but if they span across too many sources of information, they are confused about how to assist their customers. This leads to customer frustration.
- Employ user testing. Test internally and then with actual users. Add a brief survey at the end, with ratings and with the ability for the customers to provide open-ended feedback. However, even after the chatbot has been deployed, keep an eye on the analytic reports and review random chat conversations to identify bugs or errors, fix them, and test them.
- Once deployed, check to see if your chatbot serves its purpose: to answer questions. If it’s unable to do so, it’s time to go back to the drawing board to find out what needs to be done to help it serve its purpose.
- Map it out. If your bot’s tasks are relatively simple, such as ordering food, or providing information about a store, such as hours of operation or location, you can create a simple MS Word document of the customer’s journey.
- Suppose your chatbot assists with complex questions, such as “when my shipment will arrive” or “details about my purchase,” then use MS Visio or something similar to draw out the whole flow. When I created my chatbot, I used a lot of sticky notes and a whiteboard to identify the flow and determine if it made sense.
- Review the map. Sometimes developers may miss things, so have someone else look at the flow as if they were using the chatbot to assist with the various inquiries. This way, the flow is relevant.
- Agent hand-off. Introduce your chatbot as a chatbot, and don’t make my former colleague feel that she is a bot and confuse your customers. Also, make it easy for a customer to stop working with the chatbot and get handed off to an agent. Further, let the customer know that they are being handed over to an agent for assistance.
- Document the processes. Train and document what chat agents can say and do. Chat agents need to use the same knowledge resources that your call center agents use; this way, everyone is on the same page.
- I recall a few years ago when I was chatting with my internet service provider (ISP), where the agent provided me with a program that would increase my internet speed while reducing my monthly payment, a deal made in heaven, I thought. The agent stated that I was enrolled in the program, and it would take 24 to 48 hours to take effect.
- I felt it was too good to be true, so I logged into my account after three days, and there were no changes. I reached out via chat and phone, and every agent denied that the program existed, but I had the entire chat conversation saved to my email. In the end, the ISP apologized and stated that they couldn’t honor it.
- Use clear terminology. Make sure your bot can recognize the words used by your customers. If one customer calls a food item pizza, and the other calls it a pie, make sure your chatbot can recognize what your customer is inquiring about. Your bot needs to take advantage of natural language processing (NLP) to understand your customer.
- Personalize the conversations. If the chatbot appears within a customer’s profile, ensure that your chatbot uses the customer’s name and other information to assist with the customer’s inquiry. If the chatbot appears on a public-facing website, collect the customer’s first name, email address, and the reason(s) for using the chatbot for assistance.
- Keep the knowledge base accurate. Ensure that the knowledge base the chatbot pulls from is up-to-date with the most recent changes and that the communication flows also reflect those updates.
- Simplified user interface. As I mentioned before, keeping your chatbot simple is vital. Avoid responses with lengthy paragraph messages or providing the user with too many options for a single inquiry.
- For example, a user submits an inquiry and is provided with four or five chat messages, one after the other. Ensure that the user interface is easy to understand, such as the icons used to send messages, and there are minimize, cancel, and clickable options within messages, to name a few.
Test internally and then with actual users.
I don’t believe chatbots are evil or that their primary purpose is to annoy customers before getting them to an agent. I feel that they need to be designed correctly with the end-user in mind and not another tool for the customer to play with to keep themselves busy.