Powered by data-driven logic and specialized algorithms, MC converses with job seekers much like a human. The chatbot guides each user to easily apply for open jobs based on discipline and location, and provides education on the company’s culture and values.
Why a chatbot?
An entire generation of job-seekers was essentially born with a smartphone in their hands!
- 01Globally, chatbots rank highly as the preferred way to interact with businesses.
- 02Among Gen Z mobile device users, 52% spend more than 3 hours per day using messaging apps.
- 0353% of people are more likely to interact with a business they can message directly.
- 04Bots deliver information directly to the user, are on-call all the time, and often "live" within messaging apps
Why on Facebook Messenger?
As a tech agnostic company that puts users first, Five decided to leverage Marriott's large Facebook Community (with over a million followers on the Marriott Careers Facebook Page) to provide a solution that was fully integrated into an existing process and that users wouldn't have to download separately.
How does one design a bot?
Mapping out conversation trees & user journeys
Mapping each step of a user journey from the welcome message to final outputs, conversation trees & user journeys serve as the chatbot equivalent of programming flowcharts. We mapped out all the possible scenarios of our client’s employment process.
Identifying Keywords and phrases
Because users love free-form input and typing in whatever they want, identifying proper keywords is essential for triggering different user flows. For example, if a user mentions a particular word, the bot will trigger an event which will give the user multiple options, depending on the context of the word.
Five combined more than 200 Rich Media assets to keep users entertained and help them navigate through the content smoothly and in a fun way.
To make a bot understand human sentiments, we needed to integrate a sentiment analysis library. That allowed the bot to identify whether the user’s input is positive, neutral or negative in order to provide more robust analytics and an appropriate answer like: “Glad I could help” or “Oh, I’m sorry to hear that”. We used the Sentiment JS library to return a sentiment score for a sentence and attached this sentiment score to the result and send it back to the bot.
Building a Bot
After everything was mapped out, we moved to the development phase.
The architecture combines:
- 03Bot Building Platform
We built the backend ourselves, implementing database sources and social media API hooks. For the content sources we used Marriott Jobs, Marriott Careers, and Marriott Careers Social Media.
Facebook Messenger’s platform enabled us to use text messages, multimedia and multiple choice messages (with buttons) as modules.
Bot Building Platform
Using Dexter we were able to leverage the power of Rivescript (a scripting language made specifically for writing chatbots) instead of hosting our own Rivescript parser/server. It also enabled us to connect to various analytics and social platforms. Dexter's platform also features built-in broadcasting mechanisms which enables user re-engagement after a certain period of inactivity.
For the purpose of AI, we chose Google api.ai (now called Dialogflow) to leverage Natural Language Processing. It uses machine learning to figure out what users are saying and works on virtually every platform and user device out there.