casevoice assistant application.
productGoogle Assistant .
Voice is the new chatbot making customer interactions even more human. By simply speaking, it makes it more easy to interact with their favorite brands or get the information they need. We’ve created a voice application for Expert. With Voice they hope to give their shoppers the ultimate customer experience.
Expert won the award of best electronic store chain in the Netherlands three years in a row. To keep their crown, they need the best services available. The intelligent and voice controlled Google Assistant that can service their customers is the solution, and that is exactly what we've created for Expert.
The key challenge we face with the Google Assistant is to take into account the range of answers that a user can give. Answers such as "no idea" or "I don't care" to very specific answers such as "I'm looking for a 49 inch television". Another challenge is to give the assistant a human persona, and reduce the robotic feeling inherent to these solutions.
In addition we had to ask the right questions, in a natural way, so that certain filters can be applied. Finally, we also had to take into account that a users doesn’t always know what their looking for, or may not have a preference. Alternative paths must therefore also be created. In short: the assistant must have an appropriate answer to everything.
The Google Assistant of Expert.
The Google Assistant will be released in multiple stages. This is the first release and contains the following futures.
This feature allows customers to find their nearest stores, with a direct link to Google Maps to easily access directions.
Easy overview of the latest offers from Expert, with convenient links to the relevant store pages.
The assistant will ask the user questions, catered by Experts expert sales team, to discern which TVs on offer best suit their needs and desires.
Customers can also use the assistant to directly filter specifics and find the TV they’re after, or perhaps something better which meets their criteria.
Creating the conversation flow
Before we start writing a conversation flow, it’s important to take into account that what should always come first is that the conversation between the Google Assistant and the user should be as human as possible. The more human the conversation feels, the less users have to learn about how the application works. Before a flow is established a few goals must also be described along with which steps must be taken to achieve these goals. i.e. for televisions, this goal is to provide advice to the user and determine which steps are required for this. When the script is complete, we visualize the flow. We made use of Miro to connect all the different sub-streams into a holistic view of the flow.
Implementing and testing the conversation flow
The next step is to implement the conversation flow. We use DialogFlow, which allowed us to build and test the conversation flow. Once completed, testing can begin. We play a role playing game, wherein which one person is the user and another the assistant. In this way we check whether the client's image has been created correctly. After the first test phase, the conversation flow is adjusted and tested again. Rinse and repeat until the desired result is achieved.
Building the application
Once we have our flow, the application can be built. DialogFlow tries to discern what the user is after through using machine learning to match key phrases to a range of things users might be saying. This is referred to as the "intent". Every "intent" has a so-called "fulfillment". This adds to the word, as it were, and collects the information from the internet. Through "webhooks" data is sent to a "custom endpoint". Once here we talk to Expert’s API in order to collect the matching information from their systems. In the case of Expert, the API is based off a Magento2 platform. With the help of this data we can find the right match such as a television based on your preferences. Consider the image quality, format and brand of a television. As soon as the correct match is found, it is sent directly to the user.
We don't design for you or your organization. We design for your end users!