Facebook messenger chatbot journey design

Awarded: Best Effective Technology Contact Center (Bronze), TCCTA 2021

Project Date

Feburary, 2020

Company

DirectAsia (Thailand)

Technology

Chatfuel, Maxar CRM

Who's in the party?
User Experience (My team)
  • Myself – User research, Journey design, bot development on Chatfuel, API connect
  • Charsiree S. – Graphic design
Customer service
  • Admin
  • Collect customer feedback
Brand
  • Align mood & tone with corporate branding
attachment-10
Automate Process and Generate Qualified Leads During the Pandemic

So why did we do this?

There was over thousands of incoming chats per month and we have only three dedicated agent to response. During COVID-19 pandemic, all agents were working from home, resulting is capacity being minimised from unstable internet connection, not supporting work equipments and other distractions.

attachment-08

Design Process

  1. Research
    Competitor, Interview with an agent, User survey
  2. Requirement planning
  3. Solution discovery
  4. Execution
    Character design, Conversation flow design, User Testing
  5. Outcome
  6. Learnings & Next Steps

Research (Existing)

I started my research by digging into current data sources: existing number of incoming chats, interviewing with Customer Services agent who responsible for social media. This is what I found from the research.

While the number of incoming chat increases, the abandoned leads also increases.

Insights from Customer Services agent
  • There were nearly hundreds of incoming chats per day and keep increasing in parallel to increased paid media spending.
  • However, we can categorised what customers usually ask into 3 categories: Requesting quotation, Payment/Financial inquiry and After sales services.
  • One of the main agents was hospitalised from office syndrome due to repeated ‘copy-and-paste’ 8-hour straight for months. People are asking same questions, so he had a sheet of common scripts to copy-and-paste from.
  • Customers tend to response more if agents response by female tone.
FBchat_trend2019
Insights from user research
  • Thai people get more comfortable talking with a real person about insurance more than an automated machine / a computer
  • With PDPA regulation is going to be active in June 2020 (now delayed to 2022), more people become aware of securing personal information and afraid to give away their info without talking to someone trustworthy
  • When stating chatbot, many people mentioned poor experience with IVR (Interactive Voice Response) and assume chatbot will leave them confused with many possible directions – just like IVR does. So they would rather stick with direct inbound call (which the number had been decreasing) or quick chat on Facebook messenger.
  • But why Facebook messenger, not LINE? In 2020, Hootsuite reported 94% Thais use Facebook and 64% prefer Facebook chat with Brands over direct call or email inquiry.
  • Asking anything on Facebook does not need to ‘Add-Friend’ or sign up. This make everything so convenience.

Because of PDPA awareness, more people only provide their contact info when they feel secured.

Research (Competitors)

FBChat_Competitors

Please note that this research was done during late 2020 – early 2021.

Requirement Planning

Human-like experience

Designing a dedicated character for bot services

Support both pre & post purchase stage.

Learn from existing pool of users and scripts. Then, make the bot scripts chat-like and keep it informal.

Separated into 3 journeys – Get quote, Find garage near me and FAQ/Promo inquiry

Automate with minimal human contact
(but can reach help at anytime)

Integrate with existing CRM/Lead management tool via JSON api

Solution Discovery

Card Sorting

We worked among Marketing and Services team to define main functionalities for our first launch, based on pre and post purchase journey.

And here is the output:

  • Request a quotation
  • Compare between policy types
  • Option to live-chat with real agent (for multiple incorrect inputs)
  • Claims service: Find dealer garage near customers’ location
  • PDPA consent

Additional functions for next sprint:

  • Answering FAQs (especially financial related questions)
  • AI learning based on keywords with most input
  • Real time motor quote (needs to enhance API server connection stability)
FBChat_ServiceOffering

Execution

Conversation & Process Flow

After we finalised main funactionality, me and Services team started to draft the journey before adding ‘as-is’ scripts into each journeys. Below is the overview of the whole bot flow.

A snapshot of conversation flow
A snapshot of conversation flow

Conversation Design

After adding an existing scripts/what they normally response in each situation from Services team, I worked alongside with Brand team to adjust the scripts in order to keep the mood & tone friendly enough for customers to think it is not a bot, but still formal to maintain DirectAsia branding. Using the right pronoun is very crucial when it comes to designing a conversation that drive conversions.

Commenting back and forth between me and Brand team (purple and blue sticky notes)
Commenting back and forth between me and Brand team (purple and blue sticky notes)

Character Design (Visual)

‘Noo-D’ Character was brainstormed by Customer Services, Brand and myself. After we settled, I had worked with my team to transform idea into visual and fine tune the scripts. The character design was done by my mighty designer, Charsiree S., who took up my sketch and create a meaningful Noo-D character to represent the bot.

Some of the draft we made (out of 100+ sketches)
Some of the draft we made (out of 100+ sketches)

Bot Development in Chatfuel

Due to limited resources and the need to keep the project as lean as possible, I decided to go with Chatfuel bot platform as it requires minimal coding and able to integrate with CRM via JSON api. It took me a total of 4 weeks to complete the prototype and complete UAT testing (I ran it on seperated test Facebook page).

Chatfuel backend screenshot. Censored some sensitive data.
Chatfuel backend screenshot. Censored some sensitive data.

Outcome

Over 70% of incoming chat becomes qualified leads (finished and passed Underwriting screening questions) without any assistance.

  • Minimise time spent to get all customer details from 3mins/person to almost none
  • Reduce man-power to take care of chat channels from 3 to 1 person
  • Comparing to prior year, lead to sales conversion grew by 568%, generating 3xx policies per month.
  • Took 6 weeks to complete with the cost of 3,000 THB per month
  • Able to use this pool of customers for further retargeting purpose (i.e. Sponsored Ads, Click-to-Message Ads)
  • Awarded: Best Effective Technology Contact Center (Bronze), TCCTA 2021. Read more.

Learnings & Next Steps

  • After a year of first launch, we have gained a pool of users. I would like to run a focus group interview on their exceptation during lead dropping. Anything we can do to improve?
  • Personalised scripts based on traffic/media — links with specific ad sets.
  • English bot