“We want to free up 50 % our support time.” Our client had a problem – their support specialists were spending most of their time helping the end users browse the users’ manual.
This is propably not the only case in the world where users don’t refer to the manual but instead contact support as their first problem-solving action. The support is happy to help of course, but advising the users solve the cases already solved takes up valuable time they could be using where they’re really needed – helping to solve users’ unique problems that have no reference anywhere yet.
Empathizing with the end user was key in designing the conversation flow and tone-of-voice.
The idea was raised: maybe an AI assistant could take care of the most obvious helpdesk cases, and human support could concentrate on the more advanced problems, where their expertise is in the most valuable?
Framing the case
After some necessary and insightful business dialogue Teamit started to work on the case. We had already been doing a lot of studying and been developing our own RAG AI application, so we had quite a bit of ideas to start with and merge with the client’s wishes and requirements.
In the briefing meeting there was a great conversation already going on. We loved the open atmosphere where lots of ideas were being thrown in and discussed. We were able to find the common ground to base the development on and sketched a rough timeline for the project phase one.
Designing the AI helper
Our client was extremely willing to share all possible information and we took this opportunity. We paid a visit to their premises and saw the system onsite. A client expert ran us through the setup and demonstrated the system setup, how it works and what the user interface looks like.
Seeing the expert at work helped us understand the system benefits and understand the language. Expert’s views on system use and optimization started to give ideas for the tone-of-voice of the AI entity as well.
We also had really informative meetings with several other client experts from their product development and support teams. We went through some support cases and talked about what kind of help the users usually ask for.
Our notebooks full of notes and scribblings we started the design work. We took all material to Miro and started the synthesis. Our high-level idea became clear – we had started a path of creating a skilled helper bot who would evolve by time. It’s first task was to help users at the system application, offering its help when it was needed, where it was needed.
User journey mapping showcasing the necessary
We mapped the user flow and possible dialogues based on the insights and learnings from the client experts. This helped greatly defining the bot identity, tone-of-voice and overall functionality. One important decision was to make the bot “just a bot”. Thanks to this, we we able to settle on generic helper identity.
To further narrow down our MVP, we moved on to bit more spesific mapping of the dialogue. The client system has some very clear workflows but also some unexpected ones thanks to variance in the production environments.
In client meetings we got valuable feedback about the proceedings. Having visualised our thinking helped the discussion stay on the selected features and we got great informative comments on the process.
The core – conversation flow
After the visual design, we once again noticed how quickly the discussion moved back to UX. It might not be a shocking revelation that conversational user interface design must focus on the actual conversation. In our case the bot is an integral part of the next software version and thus was built on existing UI elements.
The flow of the actual conversation is far more important. Empathizing with the end user was key in designing the conversation flow and tone-of-voice. We wanted the bot to be empathetic, positive and always able to guide the user forward using the knowledge from the manual and support experts’ knowledge base.
We paid extra attention on how the feedback is asked and how often.
Last super important component in the dialogue was feedback loop. We paid extra attention on how the feedback is asked and how often. Building functional AI apps is all about learning about the users and bringing the learnings back to them.
Next steps in the life of an AI bot
The first generation of the bot is rolling out and we’re curious to learn about the users’ reception. The feedback loop hopefully keeps us updated about the state of mind of the users at their everyday tasks.
Meanwhile, the next iterations also engage more thorough methods of service and UX design: workshops, interviews and more mapping of various use cases. The insight being gathered at phase one will guide us further down on the development path.
Interested?
We hope this case study helps you to understand the phases of great AI bot design and development. Please contact us to hear more!
