Smart categories

Business challenge

Sprout Social Listening processes 50,000+ social media posts per second! How is a social media manager to make sense of all of it? The solution was Named Entity Recognition (NER), an ML-driven data science model. NER identifies blocks of text that have names and predicts what category these words fall into. I came in to clearly and effectively present this data to users. We helped customers turn a massive amount of data into business insights that enable them to identify industry gaps and improve their brand health.

Collaborative design process

I worked with Data Science and Product Management to launch a beta, which we called Smart Categories. I interviewed beta testers, who reported that the model provided new value by better highlighting “what’s hot” than simple-text aggregators. They felt the model provided a “better way in” to making sense of data. They asked for more granular control over filtering noise in the data based on the social content. So, I introduced widget-level filtering and increased user control. I also explored ways to group data to enable customers to hone in on the interesting insight and shared these hypotheses with Data Science for model refinements.

Hypothesis: People don’t care about the specifics of what groups are made up of, they care about quickly seeing where the activity is.

Design Approach: Obscure grouping from users. Sprout prioritizes delivering actionable data groups to our users.

Hypothesis: People do care about seeing how Sprout grouped Subjects for them, but they don’t care that much about editing anything.

Design Approach: Display groups as the default, but enable people to understand what’s going on behind the scenes.

Hypothesis: People want to edit their groups and make them “right”, whatever that means to them.

Design Approach: Display groups as the default, but enable people to understand what’s going on behind the scenes and ungroup subjects as they see fit.

Outcomes

Upon launch, Smart Categories surpassed engagement rate KPIs (60% of users engaged with the feature within the first week of launch) and received positive qualitative feedback, indicating its value as a powerful research filter and drill-down starting point.