Delivering new insights via artificial intelligence

Fairhair was the name of a legendary King who united Norway (where our company was founded). It was also the code name for Meltwater’s new platform intended to unite our various back ends. It then became the official name of our developer platform. My team was tasked with supporting the platform through the design of the website, several proof of concept applications and, ultimately the user experience for a series of AI-driven insights based on a knowledge graph of entities, events, people and places.

How do we show AI-driven content?

A key UX issue was how to integrate this new insight offering into our existing product experience while also enabling it to stand out. It needed to display in a feed similar to our main content feed but communicate that it was not a news article or social media post.

Proof of Concept

We focused on a series of pills that described the event type (e.g. an acquisition, a product launch). The event type is the foundation of the insight. The event is then described in natural language. Finally, the card contains some supporting metadata, most typically the entities involved in the event.

A proof of concept application was designed on the developer platform. This was tested internally at first and then with prospective customers.

After the proof of concept, a decision was made to start building the service in our customer-facing product. The product went through a number of rounds of testing with beta customers. Some of the key learnings included,

• While enterprise customers indicated they would be willing to help train a machine learning model, normal users said they would not. As a result, we dropped the thumbs up/down feature which helped train our models.

• Customers need to be able to see the articles that drove the signal to verify the event. This led to us adding the ability to click an event and see a sample articles in the feed (see screenshot below).

• Users are only interested in specific event types that relate closely to their job function or company use case.

We initially thought customers would want to see all signals associated with their company or a competitor so the setup process focused solely on finding and selecting organizations. But in testing, we found that if a team using our product is not involved in financial management, they don’t want to be notified about financial results. So the setup flow now includes the ability to select only the event types of interest.

Release to Customers

The final product was renamed Alerts as part of a larger initiative. The alerts can be delivered into the main app on a side panel, via emails or notifications on your phone.