Making matches that make sense

12th Dec 2019
Personalized services are often driven by inferences that inform powerful features. How can online services drive people’s understanding of these AI generated inferences and give them control over these them?
Product Context

Bae is a dating app that wants to help people find their match while feeling in control of their data. Bae will only use the data collected and submitted during the use of the app and it provides explanations on how data is used to find connections to other people.

The more people engage with the app, the more refined the AI recommendations can become to create better suggestions. Bae’s recommendation engine is setup to be as transparent as possible so that people know why matches have been suggested and can provide feedback on whether this match was good or not.

Bae draws on a complex range of signals and data to suggest potential matches and interest groups to the people who use their service. Inferences about the people who use Bae inform these suggestions - its whole service is about matching people. This includes survey questions as well as their activity on the app.

Problem & Opportunity

Bae wants to offer the people using the service an empowering experience that helps them understand and control how it draws on their activity and others using the service to make inferences and inform its suggestions.

How might we...

...explain which data is used to make suggestions and allow people to influence and control this

Design Features
Refining matches on the go

When a person has a match on Bae they can immediately see the main factors that influenced this match. For instance, a shared interest in a topic like extreme sports or visits to a similar location. They can refine the information by indicating whether the match is good or not with a simple ‘x-out’, and rank how important the different pieces of information are for them by moving the elements around. A person can help the AI learn and make better suggestions in the future and remove any unwanted inferences.

Asset 4
Design Features
Constantly yet effortlessly building the profile

To further strengthen the matchmaking profile, Bae prompts people to answer a small number of questions at different times while using the app. For example, at signup Bae only requests a small number of questions to be answered - minimizing data collection - rather than a full personal questionnaire. After a match has been accepted or rejected, similar lightweight questions are asked to refine the profile in areas where information is needed to make better matches over the course of the entire Bae experience. The value exchange between data input and results feels more conversational and natural than adding all data upfront.

Next steps

Bae wants to help people to step outside of their own ‘dating filter bubble’. Perhaps in the area of romance it would be interesting to consider how AI might reveal people to a Bae user that may not be obvious at the outset. Bae wants to enable people to get a recommendation that is outside of their typical profile and it to feel natural and welcomed.

How might we build on this concept to...

  • enable people to find meaningful connections, even those that aren’t that obviously similar to their own profiles.