Big Recipes VUI
An appliance company wanted an app that would allow users to upload recipes and access them over a visual and voice user interface. I created conversational flows, built the bot in our proprietary software, and collaborated with marketing to run a beta test.
In the beginning…
We were tasked with creating a bot that could deliver one of multiple step-by-step recipes. The company provided us with a series of 20 recipes in different formats. We decided to create a web-based as well as voice-based bot as an Alexa skill.
Preparation and Research
I discussed the vision with the customer. The customer made it clear that they wanted the product to be more akin to an interactive cookbook than a cooking class. From a conversational standpoint, this was unfortunate because it meant I wouldn’t be able to rely as heavily on user’s natural understanding of conversations and instead would have to give them explicit instructions on how to navigate a cookbook by voice.
I researched a number of different cookbooks, pulling their tables of contents to determine the most popular categories. I would then use the most popular categories as options to help a user locate the recipe they wanted.
I also looked into other recipe skills available for Alexa and read user reviews to learn about pain points.
GUI vs. VUI flow
There were two parts to the recipe bot, the user selects the recipe (green in the flow) and the user interacts with the recipe they select (yellow).
When designing the chat flow for the GUI, I was able to put a lot of information in each output because all of the text would stay on the screen so the user could reference it over and over again. For instance, when selecting a recipe, I could give the user a list of 8 options to choose from. But, when designing for the VUI, I had to limit it to just a few options at a time so as not to tax the user’s working memory.
In the GUI, once a user selected a recipe, I gave them the option to view the recipe in its entirety. This option was removed from the VUI.
We added the ability to navigate back and forth between steps in the VUI. This wasn’t really necessary for the GUI since a user would be able to see previous steps displayed in the chat history.
Error handling in the VUI needed options to be repeated, but could be very simple lines of text in the GUI as the user would still be able to see options that had been presented in the chat history.
One of the pain points noted in another recipe app was having to verbally navigate through every single ingredient. So I made my ingredient list be listed in full and created an option for users to ask how much of a specific ingredient was needed if they wanted to hear one item in particular.
Note- although not pictured in the flow, any step could be repeated if a user requested it.
Other Challenges
Uploading different recipes took a lot of time and the company commissioning the bot wanted to greatly expand the number of recipes. Initially I created a packet of documentation that described step-by-step instructions of how to create a recipe within our system. This was passed on to remote contractors who uploaded more recipes. Then, the customer clarified that they wanted a bot where a user could upload their own recipe and then access it over voice. Even with my documentation, it would take a person a couple of hours to input a recipe into the system. Since the company wanted it automated, this task was handed over to our engineering team which created a chrome extension that could extract a recipe from a webpage and automatically upload it to our system.
When creating a skill in Alexa, Alexa assumes you will be building the skill directly into Alexa. We wanted to connect Alexa to our conversation design platform, but because of how Alexa handles user utterances, it would only send us truncated versions of user input. I was eventually able to find a work around to get Alexa to send us the full text of all user utterances.
User Beta Testing
After multiple rounds of internal testing, I collaborated with our marketing team to run the beta test. Testers were drawn from the employees of the company that was commissioning the recipe bot.
Beta Test Onboarding Bot
I provided feedback and edits to a bot that one of my coworkers created to onboard people to a beta test. I proposed changes which prevented the bot from dead-ending without signing a user up in ⅔ of paths. When it went live, 80% of users who accessed the bot signed up for the beta test.
Emails
I drafted emails with instructions on how to sign up for the various parts of the beta test as well as how to upload and access the recipes a user uploaded. I sent out emails with user login credentials and monitored the email account so I could respond to user queries and requests for password resets. I also drafted several themed monthly emails that gave users ideas about recipes they could try and reminded them that they could reach out for help if needed.
Alexa Beta Test
I initiated and managed the Alexa side of the beta test, sending follow up emails to people who didn’t download the Alexa skill after a period of time.
User Surveys
After the beta test, I sent a short survey to the beta testers to collect feedback about their experiences.
Results
The company that commissioned the bot decided to discontinue it after the beta test. They had anticipated more beta testers from their company, but low interest meant low adoption throughout the test.
Had they chosen to continue, next steps would be using the user feedback to improve the skill.