Building data services to bring education to millions, Part II


An embedded approach to in-platform analytics

This post was co-authored with Juhi Singh and is Part II of the series (Part I here).

In this installment, we cover Coursera’s In-Platform Dashboards, which serve descriptive and advanced analytics in easy-to-read visualizations directly on the site. At our partner institutions, these are consumed by instructors all the way up to program leaders; and among enterprise customers, from individual managers to heads of Learning and Development. The common thread? The need for aggregated insights for decision-making — and lightweight access to those insights, ideally right on the platform.

At Coursera, we’ve provided in-platform analytics for years. Our initial dashboards were for university partners, and we built them end-to-end in-house. At first, home-grown dashboards had the benefit of being fully flexible and limited cost expenditure. However, as our data needs grew — both for our partners, and then also for enterprise administrators in our newly-launched Coursera for Business channel — we realized that leveraging external options could help us respond more nimbly to the varying data needs of stakeholders across our community.

As a result, a little over a year ago, we started exploring third-party solutions to facilitate in-platform analytics, and we made the leap to pilot Looker (our then internal reporting tool) as an embedded solution in the platform.

With little more than a quarter’s work by a single data engineer, we went from nothing to a full suite of high-quality dashboards for our Enterprise customers. These included descriptive data — for example on enrollments, course progress, and learner feedback — as well as advanced analytics on areas like skill development and ROI. The dashboards were a hit from the start, frequently cited by our customers at conferences and round-tables as one of their favorite admin features.

Example data visualization that helps our enterprise customers understand new skills acquired using Coursera.

Given the quality and efficiency of the embedded solution, we doubled down, pursuing a similar approach for our university and industry partners, and with similarly impactful results. The screenshot below shows one example: the course-progress funnel provides instructors with insights into how learners are progressing through a given course and where they are falling off.

Example viz from our partner center providing insights into how learners are progressing through courses

One concern our design team had when we first pursued this embedded approach was the limitations in the flexibility of the visualization that reporting tools like Looker and Tableau would support. Fortunately, this has not been a blocker. The tools allow us to build flexibly using d3 or other charting libraries; dynamic filtering has been a great asset for us, allowing us to customize visualizations across individuals; and some of the most desirable features, such as the ability to download PDFs and raw data further enhance the analytical experience.

Operationally, we take a service-oriented architecture approach. Product engineering is responsible only for embedding the data visualization in the front end using i-frames; all other efforts — from building the backend data pipelines to building downloadable visualizations — resides with the data engineering team.

For Coursera and our external stakeholders, embedded analytics have been a win-win. Our partners and enterprise customers love that we are getting high-quality data and visualizations in front of them more quickly. Our product engineers appreciate that they can focus on Coursera’s core teaching and learning experience. And our data engineers love that they can efficiently design, build, and ship innovative new on-platform analytics straight to end users.

Check back soon for Part III: Piloting Self-Serve Analytics as a Service.

Interested in Data Engineering @ Coursera? We’re hiring!



Source link