This project was an exploratory case to address a personal itch - learn voice assisstance channel. Better yet, while producing a tailored solution for other avid, tech savvy readers.
In a sprint of 2 weeks, together with skilled product team, I produced and tested an MVP - a voice skill for book recommendations.
Amazon users (readers) require intuitive ways to get tailored book recommendations.
Exploring voice assisstant technology to produce voice based application. A skill that empowers Amazon power users to get recommendations and complete the purchase by using their voice.
Amazon Echo Alexa Skill MVP that would allow to further validate hypotheses.
Project plan and responsibilities overview
Consisted of: one UX designer, one developer and two researchers.
UX, product management and prototype development.
LEAN UX. A two week sprint, including prototype development time.
3 days - discovery & ideation followed by initial hypotheses validation; 1 week - prototyping; 2 days - final validation.
Using personal/team SME knowledge on voice technology and evolving voice assistant market, we came up with several hypotheses.
We started to validate initial ideas by using online survey tools target Amazon users and using filters to capture those who: purchase books online and have or consider buying a voice assistant. The results painted the market and allowed us to define user profile (proto-persona) to be used in documenting insights, pain points and opportunities for feature analysis to follow.
After validating the findings with actual users (3 individual interviews) we validated initial findings and started brainstorming on ideas to enhance the opportunity map.
Some of the ideas included fully functional voice based purchases, book recommendations based on past purchases etc. However, for MVP purpose we've decided to prioritise the list based on user value and other dependencies (e.g. limited sprint timeframe).
Using the findings team came up with a basic prototype containing 2 MVP features: skill introduction and recommendation engine.
Using hardcoded book titles, we've split the values into categories (i.e. business and entrepreneurship, marketing, leadership, fiction, sci-fi and lastly fitness, health and wellness) and created corresponding invocation list.
In the end MVP consisted of 1000+ titles and the app to support it. Now, users would be able to get a recommendation by using a simple invocation, such as:
"Alexa, recommend me a book in business"
Final MVP validation was performed with 5 users on individual basis.
The users were asked to respond to inaudible instructions - a slide with 3 simple images (see below). As we shaped the product based on best practices (a set of heuristics from top performing Alexa skills) it was clear that testing voice experience by simply asking users to use certain invocations wouldn't yield sufficient feedback.
The responses and initial confusion was captured in order to reshape skill invocations and linguistic structure.
Key actionable feedback can be put into the following categories:
Three users reported invocations being too confusing. They also suggested that it would be helpful to have a help feature or instruction of what they have to say and when.
Every user reported the conversation with Alexa: "way too robotic", "limited", "unnatural", "taking too long".
At a time (late 2016) Amazon Alexa had no continuation capabilities (e.g. it would not recall what was asked before or connect it with other invocations). Users pointed this out as unnatural and deal breaker when making a decision if they'd consider using device/skill often.
Based on findings from user validation sessions, we've captured actionable feedback to update MVP that would allow us to publish it to the Amazon Skill store. The updates included:
Users are introduced to all invocation options when skill is launched. Furthermore, help/tutorial section has been added.
We created and validated a text based script with the users. As a result, the interaction now is much more fluid and doesn't take as long to get required results.
Custom continuation support has been added to allow users to ask for more recommendations without having to restart the skill.