My company Freightwalla is a technology-enabled Freight Forwarding platform. Our mission is to bring transparency and efficiency to the freight forwarding industry, where most processes & practices are decades old and are due for a digital evolution.
We are building Freightwalla-Bot to target businesses in emerging markets that aren't technology forward and still do business on paper ledgers and phone calls/text messages/emails. We see natural interfaces like chat bots as a gateway that will easy the transition from the way they currently do business with a friendly easy to use interface for performing the tasks they need.
What it does
What we’ve done with Amazon Lex, is build a digital freight forwarding assistant that helps the customer search, compare and book ocean freight. The bot guides the user through providing the basic information to do a search, let's them query the results through a number of intents, and finally place a booking for an ocean freight shipment.
How I built it
For the core business logic we used our internal Freightwalla Apis. The rest of it was built with Amazon Lex/Lambda and integrated with Slack for now.
As part of the building it was important to ensure that the conversation flows smoothly form one intent to another, and that we thoroughly tested and iterated on the utterances/slots with our operations team and select customers. It was key to ensure a good understanding of the conversation flows and that we were delivering the appropriate details at each stage of the conversation.
Challenges I ran into
Aside from a lot of tuning, one of the larger design challenges we had was how to structure the Intents. We knew our goal was to get the user from a Search to a Booking, however decided against defining that flow as a single intent with a number of slots to fill out. This would result in a forced conversation resembling the user picking a list of options.
In an actual conversation, the customer should be free to guide the conversation and not the other way around. The way we accomplished this freedom to to define several search & filter intents, which enabled the user to query the system for the information they wanted, while each progressive query enabled us to further narrow the results.
Accomplishments that I'm proud of
I'm pretty excited about the potential of Lex and what we can do in this industry. On demo'ing the bot to our operations team and customers (non-tech folks), it was amazing to see the look of awe in their eyes as they were exposed to ideas and possibilities that they had never considered before.
What I learned
Test and Iterate.. Test and Iterate.. Test and Iterate. With interfaces like this it is super critical to tweak the conversations and ensure they flow well. Unlike a website or mobile app you don't get a lot of visual queues and other affordances to direct the user to where they need to go, so it's important that you continue to test and iterate.
What's next for FreightwallaBot
For purposes of this prototype we’ve left out a number of services like booking transportation or digital track and trace, but are things we're pretty excited about building next, as well as integration with SMS and voice.
Another challenge for us to tackle is enabling Lex to support the Indian English language. While most business is done in english these days, Indian colloquialism and business vocabulary in the FF industry is quite different. We'd love to see more features in Lex to help tune the utterances and grammar rules as well as tools to help better understand our bot performance.