SSD Manychat Flow
Chatbot User Study And Segmentation Flow

Chatbot Flow, User Research, User Segmentation

Intro

During my time at South Street Designs(SSD), I was working on different products across three e-commerce brands that belong to the company (Versachalk, Silver Phantom Jewelry, Ballet Bracelet). This time I had a chance to work on a conversational flow for a chatbot-based user study.

The ultimate goal was to perform user segmentation to deliver the most relevant content and marketing promotions based on users' choices. The potential to reach customers through the chat was massive and companies like Facebook were investing heavily in not only chat platforms but AI frameworks that allow them to respond in more human ways. The interactive flow I designed contained a visual map of queries and responses between the users and the company’s chatbot that provided insights about our audience demographics and content preferences.

Chatbot flows are intended to provide an additional channel of interaction with a business for purposes other than customer service. Such interaction bots are available on a variety of platforms, the most prominent being Facebook Messenger. However, they can also live within an app or be available through text messaging. VersaChalk project was a Facebook-based interaction chatbot. The company already has been using the social platform for providing customer service, quick access to the company’s website and products.

My Role: UX Designer. Collaboration with a content manager.
I was responsible for building the flow using ManyChat framework, creating logics, segmentation, tagging, testing, collecting and organizing results.

Process

Users can interact with the chatbot either by typing text in the Send a Message textbox or by selecting one of the options displayed on the screen. The first round of testing proved that if options for answers are suggested it makes the conversation easier for users as opposed to having to type an answer from scratch. A flow that is based on that principle receives a higher percentage of answers. To test the usability of our chatbot, I’ve picked 2 groups of our engaged customers (people that were active on the website less than 2 months ago), each group included 250 people. The participants were asked to perform a set of chat-related tasks on mobile:

• chatting for customer-service purposes with either humans or the bot

• website browsing, and products purchasing from the messenger view. 

Besides regular buttons and links, I created a menu element, that, when selected, displayed a set of possible tasks. This menu allowed customers to view a website, learn more (which took them to customer support flow), and shop now (allowed browsing products in the messenger window).

In addition to it, I set some listening properties for the chat. Whenever a customer typed one of the trigger words the chat reacted accordingly. For example, words "stop", "enough", "unsubscribe", "sign out" would trigger the bot to ask a question and offer the “unsubscribe” option. This feature was very important in case the flow didn’t seem to be clear for a customer, and the person was having a hard time escaping the sequence. Such words as "help", "support", or "representative" would automatically send the user to a customer support flow. The same rule was applied to the most common typos. Whenever a word similar to the proper way of spelling was typed, the system would try to clarify it. (Example: "custmr suport" = "customer support" )

Choosing the language for the bot was a very important moment. With user personas in hand, that were created for the VersaChalk beforehand, my team decided to keep the chatbot playful and light. I've highlighted some of the techniques that tend to be successful:

•  Predetermined links and buttons saved users from typing. People appreciated having these options and even expected them for common inputs.    

• Text allowed users some flexibility in choosing the types of questions they wanted to ask and enabled them to deviate from the (often too strict) script of the chatbot.

Based on the interactions with these 2 groups (250 people in each group) my team decided to go further and conduct larger research with 7000 engaged customers. Insights from previous experience were applied when I created the new segmentation flow. This sequence was meant to gather valuable data from our engaged customers about their content preferences, work industries, shopping style, etc..

After mapping out the flow on the window glass I've moved to the ManyChat interface that allowed me to create a dynamic scrollable flow with logic jumps, trigger actions (such as tagging customers), saving answers to a spreadsheet for future analysis. The segmentation questionnaire asked users to answer several questions and offered a discount for the newest product as a reward. All answers were automatically recorded in a spreadsheet for future analysis. Specific tags were applied to the customer with different interests and metrics.

Among other questions, we scanned users' interests for being recruited for a Review Group for testing new products. I separated regular customers from Chalk Artists interested in collaboration with our company. As a compliment at the end of the survey, everyone was offered a significant discount for the newest product. This decision not only pleased the participants but also brought the company additional sales.

Outcome

Some of our users were amazed that they could buy art supplies, get the customer support team’s attention, or just browse the website so easily with just a messenger interface. With a chatbot, there are fewer distractions and less information to attend to — the experience was straightforward. In general, people’s attitudes toward the bot ranged from neutral to slightly positive. I was able to collect valuable information about our users and sort content and offers they received.

Further Opportunities

Reflecting on the project I defined a few areas for improvement for the conversational flows:

• Adding such things as help messages and suggestions for when the user feels lost.

• Users should be able to reset the conversation at any time during the interaction.

• Neither conversation interfaces nor humans are perfect, so "undo" and "cancel" are essential functionalities for a smooth experience.

• Upfront disclosure to customers that they are interacting with a bot is a must-have.

I think that using chatbots for studies and research is a great idea. Comparing to any email campaign chatbots have a higher opening percentage and click rate (83% and 32% from the segmentation flow), less information overload, convenience for users, and instant results.