Tweaked Onboarding Sequence Improves HR Chatbot Performance

As ubisend's client base continues to grow so does our commitment to a variety of non-disclosure agreements. Though we cannot disclose our client's name, we hope this case study still inspires you.

Our client is a global information technology organisation based in France with over 120,000 employees across 73 countries. As a company which delivers IT solutions, they need to remain on the cusp of technological developments to maintain their reputation and stay ahead of competitors. However, with a hand in IT services ranging from data management to e-commerce to cyber security, their product knowledge base is understandably large.

Enter ubisend

They first made contact with us in 2018, 18 months into a digital transformation journey implementing internal robotic process automation (RPA).

The second phase of their journey was planned to deliver an HR chatbot emphasising user experience and reducing HR enquiries as an out-of-hours service. These functions are important to our client because many of their contractors are client-facing during working hours, meaning they aren’t able to enquire while the HR team are in the office. As a result, HR received hundreds of email and phone enquiries a month which they failed to respond to in a timely manner.

We weren’t just handed the contract though. A global IT company wouldn’t have made it to the top without due diligence. A board-level research team thoroughly investigated nine options, including ubisend, before ultimately choosing us. Our client needed the utmost confidence that we could deliver a viable long-term solution.

How chatbots factor into the client’s digital transformation project

What makes this project interesting is that the client had initially tried and failed to develop their own chatbot internally. Failure, in this instance, didn’t constitute a lack of technical ability on their behalf, but rather the approach and implementation knowledge which comes with specialising in chatbot technology.

Our expertise must have shone through during the research stage as the clients were fully aware of the knowledge required following their initial efforts. Work quickly began on the new chatbot, starting with the process of determining exactly what information the chatbot would help deliver. Over 1,000 FAQ answers were compiled for the initial launch. Although this is a lot more than most companies would typically have, it doesn’t pose a problem at a base level as chatbots are capable of absorbing more information than any human could possibly dream of.

Complex questions hinder chatbot training

Chatbots work by using natural language processing to identify message intent and deliver what it believes is the most relevant response. They can initially be trained by humans reviewing low confidence responses to deepen the chatbot’s understanding of how similar enquiries can be worded differently.

However, the client’s workforce started asking multi-faceted questions which require context and nuance that only a human can decipher. Complex questions cause issues because the chatbot struggles to identify which parts of the message are important, reducing its confidence in supplying the correct response. In turn, this makes it more difficult to train the chatbot and ensure that it provides the right information more often than not. The problem was compounded by the volume of FAQs covering similar topics (learn more about FAQ chatbots). You can’t train a chatbot to pick between two responses when both are relevant.

Making sure that our client’s employees understood that they needed to keep things simple was a vital next step in order to continue training the chatbot. They were given a handy note to try asking quick questions rather than make lengthy enquiries. A carousel was also added to provide users with information about how to use a chatbot. Sometimes it’s easier to take users through a fixed conversation rather than hoping that the chatbot will be able to understand their intent. Buttons which provide a conversation tree lets users find answers hassle-free, directing the flow of the conversation with clear signposts.

Continued development of the solution

With an established HR portal already in place, integrating the chatbot into it was a top priority since day one. We’re continuing to support this and, as a result, the client has experienced a diverse level of communication with our team. Their ubisend account manager and pre-sales engineer helped get the ball rolling, supplying their board with the confidence they wanted before the project began. It then began under the oversight of a dedicated project manager and a lead developer to ensure that the chatbot works seamlessly.

Broadcasting to the workforce

Though the chatbot is mainly designed to provide employees with an access point to explore HR information when they need it, our client also wanted the ability to communicate with their employees through the chatbot. A broadcast message gives the client a method of delivering information to their employees in a more direct way than a company-wide email, which may otherwise sit unread in their mailbox. It’s currently capable of sending a message to every 'active' user at a certain time. Soon it will be further developed to show a 'pop up' message between certain times, allowing more people to see the message.

Start improving your employee experience today. Talk to chatbot experts.