The increase in cross-border e-commerce and the removal, on 1 July 2021, of the VAT exemption for the importation into the European Union of goods with a value under €22, have led many individuals to interact with Customs administrations for the first time. At Finnish Customs we anticipated this major change and asked a data analyst working in another public agency to help us gain insights into the number of new customers to expect. It was obvious that our current customer service resources were not enough; it was time to explore new ways of doing business.

The good news was that customer service had evolved, with an increasing number of “intelligent solutions” becoming available. There is no single way to define “intelligent” here, but we could say that such solutions: operate in an environment with other solutions; possesses cognitive abilities such as perception, action control, deliberative reasoning or language use; and have the capacity to adapt through learning.

A strong driving force behind the advances is artificial intelligence (AI), and especially natural language processing (NLP), the branch of AI concerned with giving computers the ability to understand text and spoken words in much the same way human beings can. NLP is an area where much has happened. One manifestation of this technology is the chatbot, a software that simulates human-like conversations with users via text messages on chat. Chatbot can provide information to customers 24/7. It does so by pairing questions and answers, so that questions do not have to be precisely formulated in order for the chatbot to provide the right answer. Even though it is less than perfect, it is a great tool to have in a customer service palette.

There were a couple of commercial solutions to choose from, and we opted for an off-the-shelf solution developed by a Finnish company; this meant we only had to add Customs-related information to the tool and then train it to pair questions to answers, using about 60 interaction scenarios. The chatbot we chose was easy to connect to our contact management system. This allowed us to transfer customers to our chat service if the chatbot could not answer the question or the customer wanted to interact with a person.

We named our chatbot “Hippu”, which means “Nugget” in Finnish, in honour of one of our drug sniffer dogs who retired. We piloted Hippu at the end of 2020 and deployed it at the beginning of 2021. Hippu was developed primarily to help individuals who are viewing the information on our website related to the clearance of parcels. After 1 July 2021 there was a spike in the number of individuals interacting with Hippu. We reached 6000 connections per month. Since then, the number has evened out as everyone has grown accustomed to the new legislation. Hippu served nearly 32,000 customers in the first year of its deployment. 75 % of the conversations were successful, with customers reporting that the chatbot provided an answer to their question. Only 8 % of them had to be transferred to our chat service to interact with one of our staff.

Hippu processes information in Finnish and Swedish, but for NLP solutions the best is yet to come, especially in terms of language management. Large language models, or “LLMs” for short, are algorithms that learn statistical associations between billions of words and phrases in order to perform tasks such as generating summaries, translating, answering questions and classifying text. But historically they have been costly to create, keeping them in the hands of large IT corporations. This may change following the publication, on 6 July 2022, of an open source language model called BLOOM. According to its developers, “BLOOM is able to generate text in 46 natural languages and 13 programming languages[1]”. It is not a chatbot but a text completion model. You can start a sentence as if you were writing a text, and BLOOM will generate a coherent follow-up. It is really difficult to imagine and predict all the uses of this new language model, but it will undoubtedly have a positive impact on the development of NLP solutions.

Commercial solutions used to manage customer service operations have also evolved to enable employees to address customer needs quickly through user-friendly tools. Contact centers usually handle queries across multiple channels, from digital and social media to the more traditional phone support. We realized that we needed a contact management system which would enable us to manage all our contacts, view conversation histories, and filter customers’ data into contact lists by title and a range of other attributes.

The new system we have acquired is very informative. It shows a customer’s past interactions with the customer service across multiple channels in a single database. We can also add notes about a customer’s needs and issues. This information enables us to serve customers better if they contact us again.

The system also provides for multichannel customer support, which enables us to switch easily from one communication channel to another which is more suitable for the issue at hand. For example, a customer with a simple question can use a “live chat” option rather than calling and having to wait for an agent to respond. A customer who wants detailed answers, or has a query which is major and too difficult to articulate through typed text, may be invited to call and speak to an agent directly. Last but not least, we have built a list of common questions and answers to help our staff deliver a high-quality service.

This is no longer simply about providing a service to customers – it’s about understanding their preferences and their overall experience with customer service. Ultimately, the objective is to ensure they are satisfied and to improve compliance with regulations.

More information
Mikael.hyovalti@tulli.fi

[1] https://huggingface.co/blog/bloom