Salute is the world's first family of virtual assistants for Sberbank that was lounched in . Three characters. One goal —provide a new way of interaction with bank services and create effective dialogue scenarios for quick problem-solving without waiting for a response from the call center.

As part of this project, we developed 3 information scenarios for the virtual voice assistant: consumer credit, card expiration, and double debit.

My role: VUI Researcher and Designer (in-house voice team)
Credits: Project directing by Natalia Hutorskaya, senior product manager
Platform: SberBank mobile app (Actions on Apple and Android plateform)

The project methodology was based on the Conversation design framework, which included the following stages:

• User research
  
• Ideation
  
• Prototyping
  
• VUI design - Sample dialogues, User flows, Storyboarding
  
• Testing and iteration

NOTE: Some images are low-res or blurred and cannot be expanded for confidentiality purposes.

The foundation of the Convenience Design approach is to create a natural dialogue between man and machine. To do this, we need to create a personas of our virtual assistant and clients.

Persona-models take into account characteristics such as context, individual personality traits, and standard vocabulary. The bot persona is based on marketing research of the bank's customer preferences, communication policy provisions, and business goals.

Persona is a collective image of the user segment, which represents a vivid need. We created four personas based on marketing analysis and research. We created a family, each member of which represents a specific age segment, a need and a set of social and psychological characteristics.

A dialogue design template is needed to understand the context, scenario development, and tasks of both the client and the virtual assistant. Based on this template, we prepare for a kind of theatrical production, where the dialogue between a human and a machine will be played out.
The artifact includes a description of the context and the current situation, a description of the environment, the needs of the client and the bot. It is also important here to describe the bot's limitations and the information it needs to successfully complete the task.

We broke into pairs and assigned the roles of client and bot to each other. By familiarizing ourselves with the staff models and the dialogue template, we better adapted to our roles and tasks.
This is one of the most interesting and important stages. It allows us to play out the dialogue and make it emotional and natural.
We sit down with our backs to each other and simply start a dialogue between the client who has a task and the bot who has to help. An audio recording of the dialogue is made. At this stage, it was very important to work out the blue sky scenario with the successful execution of the task. We will work through the problematic branches later.

We made audio prototypes of all the scenarios, transcribed them, and turned them into diagrams of dialog scenarios. Then we refined these diagrams by introducing additional situations and bot responses in case of problems and errors.
Now we to the test of the wizard of OZ...

We wrote off audio fragments of the script for the bot and prepared a set of phrases for subsequent tessellation. During the testing phase we invited 6 external respondents. We asked respondents to perform tasks within the scripts and communicate with the bot. Our operator simulated the bot's actions by snapping contextual phrases from his laptop. All tests were conducted remotely and recorded on video.
After the tests were finished, we got feedback from the customers, marked problematic points in the scripts and improved them.

The main topics of the questions were:
1. Values and expectations of the clients when using the Virtual Assistant
2. Multimodal interaction (the use of visual images, infographics, web-forms)
3. Switching of the client to other channels - cases, scenarios, problems and limitations
4. Cognitive features of perception and work with information in virtual assistant

Based on the finalized audio prototypes, I created a working prototype of the scripts using Voiceflow for later testing on a wide audience and debugging the script logic. After testing, the scenarios with recommendations for refinement were handed over to the developers.

The Conversation Design approach allowed our team to create 4 information service scenarios in a relatively short period of time and test them on real users. At the moment, all 4 scenarios are part of the skills of family of virtual assistants Salut from Sber.