Best Time Recommender

The Challenge

A client we worked with previously to build a location recommender engine approached us again to help them build a recommender system to suggest the best times for a given social event to take place. This was an interesting problem to solve because it required a combination of machine learning and efficient data algorithm work to create an effective solution. This was an interesting problem because the solution needed to not only be able to take in relevant event details and user preferences to it’s time suggestions, but also suggest in real-time to be useful.

The Solution

An intelligent system for this problem must to account for many users, each often having many different calendars that an automated system would need to consider. It must also understand the context of the event. It wouldn’t be appropriate for brunch to be at 9pm on a Tuesday, or Movie Night to happen at 6am. For this challenge, leveraging artificial intelligence became crucial. It was necessary to employ Natural Language Processing (NLP) techniques in order to both classify events, suggest sensical times, and optimize for event attendance simultaneously. Not to mention that many additional details like “when does the place close” or “reservation required” complicates complete automation.

We created an all encompassing system that was not only able to find the most appropriate blocks of time for an event utilizing user’s calendars and preferences, but also understand the event details via NLP embeddings, account for everyone’s travel time from their homes, and respect all of the aforementioned place details by integrating with a multitude of different location info providers such as Google Places, Foursquare, Opentable, TripAdvisor, and more.

The Outcome

Our system far exceeded our client’s expectations. What resulted from our work was an AI system that could factor in dozens of people’s calendar availabilities, understand the context around a social event, and each user’s preferences and details to deliver highly relevant and accurate time suggestions. All in real-time.

After integrating our time recommender, our client saw a 73% increase in suggested times usage and perhaps more importantly were able to unlock a new system that they could leverage into building more relevant suggestions and ultimately are looking to leverage into their ad network to improve CPM for their advertisers.

Average Classification Time

1.27s

Across 100 Attempts

Average Classification Time

1.27s

Across 100 Attempts

Suggest Time Usage

73%

User Increase

Suggest Time Usage

73%

User Increase

Ready to get started?

Let's Talk

Ready to get started?

Let's Talk

Ready to get started?

Let's Talk