A conversational style assistant that can provide personalized dressing tips and occasion-based suggestions
4 months ago, Xiaonan and I initiated a side project to enhance our daily life. As the typical life of CMU students, our schedules are crammed with meetings, courses, workshops and assignments. Time/Attention is so limited and precious. Meanwhile, we, young women, really want to look our best everyday rather than grabbing a sport wear randomly and rush to school in a hurry at the last minute. We designed Mia to help young women tackle with the problem.
Picking the outfit that can satisfy young women is not just a problem about time. Even when we have plenty of time, facing the wardrobe full of clothes, we still feel nothing to wear. Our tastes change, our confidence levels peak and trough. We need to consider the weather, the temperature, the event, the people we are going to meet, the impression we will make on others, etc.
Dressing is not just wearing something on our bodies, but a way to please ourselves, to present ourselves and to impress others. Self identity and social function both matter.
For young women at 20s, clearly knowing the personal suitable styles, picking up appropriate and compatible items that not only please theirselves but also meet certain social requirements/dress code can be challenges and time-consuming. Many interviewees showed they lacked convenient way to learn such knowledge. (link to our research findings)
Who is Mia?
That why we introduce Mia, a conversational style assistant that can provide personalized dressing tips and occasion-based suggestions to help young women look their best.
Our final product system includes a mobile application and a smart mirror.
Our goal for this product was reduce the cost of learning dressing knowledge and make young women become more confident
Our high-level goals were to:
- Help young women understand the language of garments and develop their unique style.
- Behavior Change: Help them streamline their closet and better manage their life.
- Mindset resetting: Help them understand dressing can be way to do self-reflection and to reflect on relationships with others.
The 10-week process was not quite smooth but full of twists and turns.
At the outset of the project, we didn’t have a clear mission or specific goals for the dressing problem. Without pre-existing insights, we interviewed 13 women aged 22–29 to explore what painpoints they have.
After the interview, we found 3 design opportunities:
- Personalized & Need-based
Personal style trumps the so-called fashion. And they need occasion-based suggestions, like for dating, for job hunting. Current fashion apps that focus on fashion tips and merchandise promotion.
- Personal Identity & Social Image
Dressing is a crucial introduction to ourselves. It’s also has important social functions. People may need to wear in different styles for occasions to show different aspects of themselves.
- Efficient, Intuitive & Try on clothes
People care about their styles and curious about dressing behavior, but manual input block their ways to track dressing behavior. People appreciate efficient way to try on clothes and intuitive ways to track their behavior.
Brainstorming: 6 scenarios
Then we synthesized 6 problems and brainstormed 6 sets of solutions. We carried several speed datings to prioritize the most important/fundamental problem and to test how users think about system components. We finally landed on 4 problems and 4 components:
- Smart mirror: tell people feedback and remind people weather and schedule, read their emotion and give suggestions, real-time display keywords of outfits, virtual fitting effect, record daily outfit: people love this idea — very innovative, intuitive;
- Mobile Application: integrate physical closet with digital closet, provide report of daily dressing behavior;
- Tracking tags: help them to find clothes;
- Closet modules: help them organize and optimize closet.
Our first problem statement:
Help young female users at their 20s understand the language of garments and form their unique styles.
Iteration 1: A fundamental problem
Is it too much to solve 4 problems? Will so many components increase the learning cost and make the situation more complex? Anyway, the product is not the purpose, but a tool that can help users reach the ideal life situation.
We go back to our research findings and asking “Why” helped us figure out the fundamental problem we want to solve:
“Why do people forget items? Because they have too much things. Why do they have too much things? Because they keep buying things. Why they keep buying things? Because they have new expectation of themselves.”
People have new expectations for themselves, especially when they cope with different occasions. For example, they may want to look professional and smart in a job fair, look chic and energetic in a cocktail party, look elegant but not too eye-catching in attending other’s wedding ceremony, etc.
This was the beginnings of a working north star.
After a deeper analysis of affinity diagram, we found there are 2 gaps in their decision making:
- the lack of knowledge of what is appropriate for their personality and body shape.
- the lack of knowledge to match desired appearance with appropriate clothes.
We then decided to narrow our design down to the second gap, since the first one is more about self-recognition and will significantly broaden the scope. We hope that through helping matching the desired appearance and certain personality traits with garment language and visualizing their current dressing habits, users will gradually develop a better dress sense that accommodates to their evolving self-identity.
We renewed our problem statement and goals as:
A style assistant to help young women pick the most suitable outfit to look their best in different occasions.
The system includes 2 components: mobile application and mirror.
- Home: Remind users of weather, schedule and recommend outfits based on weather, schedule and user’s own closet.
- Closet: Users can build a 3D model in the app and virtually try their clothes.
- Explore: Explore new outfit matching methods based on their own closet, explore new style and get shopping recommendations for specific occasions.
- Look book: User can save their favorite outfits, daily outfits and inspirations.
After a Mid-Fi prototype, we cut off some other thoughts because currently we want to keep this product light and focused, but we love to share some of them and to further discuss the value:
1. User can also plan ahead and save looks in their calendar.
2. Every week or month, users can receive a report that contains the insights of their dressing behaviors, as well as prompts for streamlining closet or explore new style.
3. A recycle service can help people streamline their style and contribute to a good environment. Fast fashion always do a lot of pollution and waste.
4. Users may want to have a closet sharing community, on the one hand they can share inspirations and learn from each other, on the other hand, this satisfies people’s curiosity about private space and belongings (closet).
Mia smart mirror:
The Mia mirror also have 4 functions and provide a seamless experience with the application. User can virtually try the outfit and get keyword analysis immediately which saves users’ time and labor. The mirror gives an intuitive way to save their look and input item data.
We also expect the mirror not only can reply reactively, but also read people’s emotion and give suggestions proactively, which relies on the development of image recognition technology and machine learning.
Iteration 2: Application Hi-Fi Mockup and Workflow
Iteration 2: Mirror Prototype and Hi-Fi Mockup
Iteration 3：How to reduce the cognitive load for users in mobile app?
In our user testing, a main problem identified was that the cognitive load in using the app was too heavy. As an app used on a daily basis, users expect it to be super simple and intuitive. Based on the feedbacks, we iterated several versions to reduce unnecessary information as much as possible.
In the version 1, we let users choose word tags of styles to get recommendation.
In the version 2 (view2+3), users will see a greeting, weather and calendar at the fir sight; when the user scroll up, 4 outfit pictures will show up.
In the version 1, we provide 5 outfits for users and put keywords on the bottom. We provide 3 actions include “remove this one(I don’t like this)””compare mode” and “save to lookbook”. Later we think 5 was too much and wanted to make the interface lighter, so we reduced to 4 outfits, put keywords onto 3D model, and reduced action options.
Since it’s easy to compare outfits by swiping screens, we removed the compare mode. Later inspired by music app, we change “like” and “don’t like” into heart and trash bin icons, which not only reduces the cognitive load but also make the brings more fun.
Explore is function where users can get some merchandise recommendations. Users can get outfit first and then enter an item view to see details. In the version 1, keywords searched by users are on the top, in the middle there is a card of item and 2 actions, on the bottom there is a back button. To help users focus more on the item, we introduced a modal view and made the item pop up. In the version 3, we simply put a heart icon for user to collect the item and for system to learn user’s taste — we saved more space for item information, meanwhile, out item’s name, price and description together to make the content coherent and and user’s reading flow smooth.
- Process is not a smooth double diamond, but full of twists and turns.
- Do prototype and user testing as early as possible.
- Try to focus one problem and doing small things very well, rather than solving all problem with 1 solution.
- People’s expectation of themselves and Mia will evolve as Mia become more integral to their life.
- Onboarding: since Mia includes 2 touch points and 4 functions, the onboarding experience could also be a challenge. How to make it light and engaging? That’s our next step.
Other Mia stories:
- Mia research findings: how do young women manage their closet? (TBD)
- Like, Save, Collection — what does it mean when people like something on app? (TBD)
- How to reduce user’s cognitive load on Mia app? (TBD)
Ranjitha Kumar, Kristen Vaccaro. 2017. An Experimentation Engine for Data-Driven Fashion Systems. Proceedings of AAAI Symposium on UX of ML ’17.
Kristen Vaccaro, Sunaya Shivakumar, Ziqiao Ding, Karrie Karahalios, and Ranjitha Kumar. 2016. The Elements of Fashion Style. Proceedings of UIST ’16.