Northern Gas Networks (NGN) deliver gas to 2.7 million homes and businesses in the North East, Northern Cumbria and much of Yorkshire. They employ over 2,500 people and their HR team helps these employees understand NGN’s policies and procedures, payroll, expenses, rewards, and the health and well-being support available to them.
After helping NGN’s Colleague Care team successfully launch an IT System Management chatbot (known as KITT) on their intranet site, NGN wanted to use conversation AI to help support more teams. We saw some colleagues asking KITT HR-related questions and therefore choose HR as the next team we would support with this technology.
Although we could have trained KITT to answer HR questions, KITT was designed to reflect the Colleague Care team’s communication style and its content is managed by Colleague Care who are the subject matter experts in ITSM, but not necessarily HR. NGN did not want to create one large internal chatbot that answers queries across departments as this could become very complex and hard to manage and yet they have one intranet site that colleagues come to when looking for information. Therefore they decided to implement a master chatbot that would be installed on their intranet’s homepage to direct colleagues to the relevant chatbot for their query and install the department-specific chatbots on the relevant department-specific pages of the intranet. This way colleagues on the HR pages looking for information could interact with the HR chatbot, colleagues on the ITSM pages could interact with KITT and this approach would be scalable across the different teams within NGN.
Implementation
We worked with the HR team to identify common queries they receive from colleagues that the chatbot would be able to handle. We grouped these queries into ten topics of conversation and worked with the team to design the conversation flows and interactions the chatbot would have with colleagues to resolve these queries.
The chatbot opens with a clear welcome message that introduces itself and what the service should be used for. It presents buttons for popular topics as a place to start, whilst also telling colleagues that they can type their questions if they wish. The chatbot uses Google Cloud’s Dialogflow for Natural Language Processing and is trained to answer hundreds of queries, as well as respond to small talk. As HR queries can be very personal and complex, it was very important that colleagues understood they were interacting with a chatbot, not a human member of the HR team. The chatbot announces this very clearly in its welcome message and rather than give it a human-like name or persona, it is simply called NGN’s HR chatbot. We used a robot image as the chatbot’s avatar and designed a tone of voice that is professional yet non-emotive to emphasise to colleagues that they are chatting with an automated system.
The chatbot provides answers to colleagues’ questions that are relevant to the majority of employees most of the time, and links them to the relevant policy, procedure or online resource where they can find more information or tells them what the next best step is (for example links them to a form to complete). As resolving HR queries can be very situation-specific, sometimes the next best step is for the colleague to contact HR to discuss this further. In these cases, the chatbot provides them with the HR contact details, but also offers to pass their query onto HR for them.
As the HR team mostly communicates with colleagues via email, it was decided that the best way to hand over a colleague’s query to HR was via email. When the handover flow is triggered, the chatbot will ask for the colleague’s payroll number to help the HR team identify them and then will email a copy of the colleague’s chat transcript to the HR team asking them to follow up with this colleague directly. To send these emails, the chatbot is integrated with a third-party email service called SendGrid, which is part of Twilio. If the chatbot doesn’t understand a colleague’s question the chatbot will also offer to email HR and take the colleague down the same handover flow. By handing over queries in this way, the colleague doesn’t need to change channels to reach out to the HR team, the HR team can keep communicating in their existing channel and the HR team have the relevant chat transcript to hand when picking up the query.
The Master chatbot was built as a simple button-based chatbot. It opens with a welcome message explaining that it can take you to the HR chatbot or KITT the Colleague chatbot, and asks the colleague to select from two buttons if their query is related to HR or IT. The Master chatbot then takes the colleague to the relevant chatbot, which will automatically load open for the colleague. We decided to keep this chatbot very simple and turn off the free text input to remove the risk of colleagues asking the chatbot questions and remove the need for ongoing monitoring and retraining.
Results
Since launching in November 2022, the chatbot has answered 99% of messages correctly. The HR team have been trained to use our platform EBM as they are monitoring and retraining the chatbot when it has misunderstood a colleague’s message. As we look ahead to 2023, NGN’s vision of having a fleet of chatbots that are managed by the teams they have been built to support is starting to realise.
