On 1st January 2021, the UK officially left the EU with a finalised trade deal. Providing businesses with more certainty on how to operate within the EU moving forward.
However this will introduce 1000s of traders to customs formalities for the first time and generate an additional 200 million declarations per year!The Brexit information chatbot was created to answer questions related to importing from and exporting to the EU. Providing vital customs declarations knowledge.
Introduction
The Customs Clearance Consortium are customs and border specialists with huge experience in this field.
Brexit will expose huge numbers of traders to customs formalities for the first time and generate an additional 200 million declarations per year! How, when and where you submit those declarations is vital.
Having the right document in the right place at the right time is critical if you want to sail through the border!
(Customs Clearance Consortium)
For the Customs Clearance Consortium team, the challenge is getting to the correct customs and excise information quickly therefore enabling them to provide an efficient advisory service to their customers.
Navigating through a bewildering array of acronyms and jargon and providing the right route information was the perfect use case for chatbot technology.
We worked closely with the Customs Clearance Consortium and our partners Ether Solutions to make this happen.
Chatbot brief
The Customs Clearance Consortium wanted a brexit customs information chatbot which they would easily be able to add to their current website in the customer facing and Consortium members areas.
The chatbot will work on any website and provide a responsive mobile friendly interface. Utilising free text and rich UI elements to aid conversation flow and offer meaningful and accurate responses.
The chatbot persona was to be helpful, upbeat and be called CEECEE.
The primary use-case’s for the chatbot are:
The chatbot should be able to respond to the more common domain specific questions related to customs such as brexit faqs, transit information, safety and security. The chatbot was also to take on the role of a jargon buster; to provide information about the terminology that is used when discussing import/export after Brexit and the required preparations. Answering these questions the chatbot would be able to provide its own succinct answers and/or point to the correct page on the CCC website.
Customer: what does wto mean Chatbot: The World Trade Organisation. A group consisting of 168 trading territories worldwide....
Other key areas of information which the chatbot needed to concentrate on was to provide route specific information. The chatbot would need to understand queries regarding documentation for routes and understand country names to extract them from the conversation and select the correct information for that route. These enquiries could be straightforward or more complex enquiries where CEECEE would need to ask further qualifying questions to provide the most accurate information for the user enquiry e.g:
Customer: I want to import from France Chatbot: where are you importing to? Customer: UK Chatbot: do you want to come direct, via EU or via NI Customer: direct Chatbot: ok let me look up that information Chatbot: Please follow the following link for importing Direct from France to United Kingdom of Great Britain and Northern Ireland.. Chatbot: link to correct route>>
Despite being able to answer common questions the focus on the chatbot was to be usability. With effective conversational handling and always an easy provision for users to contact the CCC team if necessary.
The chatbot will be part of a solution where Customs Clearance Staff could access chat transcripts, chatbot analytics and reporting.
How we solved it
Fact finding sessions helped us to scope the high level chatbot functional requirements. Customs Clearance Consortium and Ether Solutions provided us with user enquiry data from their teams. We collaborated on our chatbot script spreadsheet to flesh out the intents and training phrases needed to answer the more common questions; concentrating on the following key areas:
We also had access to PDF documents covering specific subject areas: e.g Jargon buster documents. Leveraging the functionality of the Dialogflow knowledge connectors we were also able to parse some of these into the Dialogflow agent to aid in training the agents conversational ability.
CEECEE was also fed with some of our standard customer support training data. Providing common customer support intents: requests to contact the team, to speak to a real person or handle complaints are supported by 100s of training data sentences from our canned datasets.
For the more complex multi turn enquiries we handed over to our conversation designers to streamline the chatbot responses and map conversational flows.
Training the data was an iterative process. Matching user intents to core functionality and features and training the natural language processor to understand users and handle conversation failure scenarios gracefully.
You can read more about how we run our chatbot projects here.
The PDF documents and training data was invaluable for perfecting the chatbot conversations. The process highlights the need to monitor the chatbot as part of a continuous cycle of continuous learning.
The natural language processing technology we used for this project was the Dialogflow ES platform. As Google Dialogflow specialists this was the natural choice for this type of chatbot project.
Chatseer provides the analytics and chatbot monitoring platform. Allowing our client to extract the relevant chatbot usage metrics and our team to maximise the chatbots effectiveness as quickly and efficiently as possible.
One of the great benefits of a chatbot is the constant availability. Customer expectations are high expecting a quick response to enquiries. With CEECEE the information is available 24 hours a day even when there are no employees in the Customs Clearance Consortium office.
CEECEE can handle 240 + separate intents (separate enquiries) with multiple training phrases(utterances) for each one of these. Some intents can be answered simply whilst others needed a more complex approach to the conversation.
CEECEE remembers context and leverages entity extraction, asking further questions to provide more relevant answers and understanding standard country and custom route entities. For example handling route specific enquiries and extracting the relevant to and from countries as well as the route specifics so the chatbot can respond correctly.
The chatbot also provides Smalltalk and automatic spelling correction.
The chatbot is supported by a persistence layer to provide the correct route information dependent on parameters extracted from the conversation.
The key as always to a awesome chatbot experience is handling fall-backs gracefully and providing the functionality to hand over to human operators as needed.
CEECEE handles multiple complaint and contact request intents providing the ability to request further customer information and relaying the enquiry onto Customs Clearance Agents via email.
Fall-back mechanisms for misunderstood intents enable changing responses so users don’t get the dreaded repeat response “sorry I don’t understand the question”. Instead the chatbot changes its responses over 3 turns providing other options.
Training is ongoing with the CEECEE chatbot Leveraging machine learning we are constantly improving responses and adding new ones so that she is always getting smarter.
CEECEE is currently supporting the Customs Clearance Consortium agents.
CEECEE has a helpful persona with an informal personality which works well with the subject matter and her role of Brexit customs assistant.
