Using conversation AI to support colleagues with IT issues
Northern Gas Networks (NGN) identified conversational AI as an opportunity for automation as part of their 2021 Digitalisation Strategy. They choose EBM as their chatbot partner to provide expertise, development services and a chatbot platform that could be scalable across their business.
To start using conversational AI, NGN decided to focus on a high-value use case that would support their 2,500+ employees by giving them access to IT support via chatbot installed on their internal intranet site and mobile app. IT support was chosen because NGN’s Colleague Care team receive a high volume of similar queries that can often be troubleshooted and resolved by the employee in a few simple steps, making these queries easy to automate. By developing a chatbot to handle these common queries, employees would be able to resolve issues faster, whilst the Colleague Care team would have more time to work on complex issues.
We kicked off by introducing NGN to how chatbots work, chatbot best practices and conversation design. We worked with the Colleague Care team to design conversation flows and interactions users could have with the chatbot. We started by reviewing their existing common queries and categorised these into two types: IT issues and new requests. We designed two types of conversation paths. For IT issues the bot will try to help the user resolve the issue themselves and for new requests, the bot will collect the required information and raise the request on the user’s behalf.
When designing the chatbot’s persona and copy, Colleague Care wanted their bot’s tone of voice to mirror their communication style. It needed to be concise and to the point, as well as helpful and personable. They named the chatbot KITT, as IT is your work kit but also as a playful nod to the Artificial Intelligence in the Knight Rider film. KITT introduces itself as NGN’s ‘virtual assistant’ that was built to help with ‘your’ IT issues and help ‘you’ raise new requests ‘quickly 24/7’. The message is short and clearly identifies what the service is for and why it was built. The welcome message also establishes what is not there for and tells colleagues to call Collguge Care directly if their issue is a ‘potential major incident’.
At the start, users can select if their question is an IT issue or a new request, and from there they can navigate most of KITT’s content using buttons if they wish. Buttons help guide users to the right response faster, not only because selecting a button is often quicker than typing, but also because by presenting users with a small number of valid options and explaining the difference between these options, users are more likely to stay on the correct conversation path. Buttons and guided conversation paths can be really valuable in use cases where the terminology used by the company or team, might not be the same terminology used by the user, or when the user might not understand which process they need to follow or what they need to ask the chatbot for.
Working with the Colleague Care team we identified common points of confusion or potential misunderstandings with the current process for raising requests. At these points, KITT asks clarifying questions to make sure the user is following the correct path and if not sends them to the correct path. For example, when requesting new laptops for a new starter this is often raised as a new hardware request, but it should be a new starter request. Therefore, at the start of the new hardware flow, KITT asks if the hardware is for an existing or new colleague, if they say existing the user proceeds in this flow, but if they say a new colleague, they are sent to the new starter flow.
Buttons are really useful guides, however, it is still important that users can and know they can ask their questions using free text. KITT uses Google Cloud’s Dialogflow as its Natural Language Understanding provider. We trained KITT to understand the language, jargon and phrases NGN employees use to express IT issues. When resolving IT issues, KITT identifies which system or hardware the user is having trouble with and what the particular issue is. We used entities to teach KITT to recognise the different synonyms for these systems or hardware. For example, KITT understands when a user refers to their ‘pass card’ they mean a ‘door access card’ or when they say ‘maps’ they mean ‘mymaps’ and so on. We trained intents for each of the common issues that employees could have with each of these systems or hardware. Once the system or hardware is identified and the right issue intent is triggered, KITT will guide the user through a ‘resolution flow’. KITT will briefly explain the most common cause of the issue and provide clear step-by-step instructions on how the employee could resolve the issue themselves. If the issue still persists, KITT provides another common solution and if the issue still persists, KITT will offer to raise a support request for them or tell them to call Colleague Care directly (depending on the issue’s urgency).
Rather than just linking users to the right online form where they could raise a request themselves, we developed an integration with NGN’s services management system, TOPdesk to allow KITT to submit a new request on the user’s behalf with all the details it has collected from the conversation. When raising new requests, KITT asks clarifying questions to determine what type of request is needed. KITT then opens a form view within in chat window, the form has drop downs, date selections and text boxes for longer descriptions, which make it easier for the user to input, check and edit the required information for that particular request. We made sure that the conversation flows for new requests and for IT issues, where dynamic and that if the user provides all or some of the required information in their previous messages, KITT would capture this and not ask for it again. For example, if the user has already said they want to request a new laptop, KITT will even pre-populate the relevant field with this information in the form view. Once KITT has the required information, it will send it to TOPdeck and respond to the user with a confirmation that their request has been created and prove them with the request ID number. With this integration, users are able to create the correct request needed to resolve their query and after trying to troubleshoot the issue themselves, submit the details of their IT issue to Colleague Care without having to channel shift.
The are many befits of chatbots like KITT for the user, such as self-service and instant 24/7 responses. A significant benefit of chatbots like KITT to the business is that guided conversation flows and checks make sure employees are understanding and following the right internal processes, so support teams spend less time picking up requests that employees could have resolved quicker themselves, or requests that are raised incorrectly or missing key information.
Since launching in, KITT has handled thousands of conversations and answered 96% of messages correctly. The Colleague Care team regularly use our bot management platform, EBM to monitor KITT and improve it by retraining and adding new content based on real user interactions.
This initial project did not only proved the value of chatbot technology, it established the best ways of working together and a chatbot delivery process that could be replicated across multiple departments and use cases. We have since successfully delivered NGN’s second chatbot user case in HR and are currently working on their third chatbot use case in customer service. We will continue to support NGN as they design and launch many more chatbots in the future.