Microsoft Helps Tacoma Public Schools Use Data Analytics to Predict At-Risk Students

For the last few years, Microsoft has partnered with the Tacoma Public School (TPS) district in Washington to ask a fundamental question: Is it possible for schools to use predict analytics to identify students that are likely to disengage and ultimately drop out? Microsoft and TPS think so. Microsoft created Azure, which equipped teachers and administrators within TPS with data that empowered them to help more students succeed. As a result of using these Microsoft cloud technologies, the district has already seen an improvement in graduation rates from 55% in 2010 to 82.6% in 2016 [1].


tacome-public-school-logoThe Tacoma Public School System

Just a few years ago, the reputation of public education in Tacoma was anything but inspiring. A 2007 national study declared the district’s five high schools, which educate nearly 30,000 students, “dropout factories.” The district’s 55 percent on-time graduation rate was well below the national average of 81 percent [2]. Needless to say, TPS was not delivering on its promise to educate “every student, everyday” [3]. Azure machine learning was just the tool to get TPS out of this rut.

What can predictive analytics tell educators?

Initially, TPS questioned if predictive analytics were attainable in an educational setting. To prove their concept, the district used 5 years worth of data for students from grades 6 through 12, and focused on pulling a variety of information: demographics, health records, and student performance information such as grades and attendance. This data was then compared against the district’s key benchmarks, which include math and reading standards, graduation rates, school environment, and readiness for life after high school. Although the district had concerns about what could be gleaned from mapping this data, they ultimately felt confident in the strength of the numbers. “As we progressed and used more historical data, the model proved to be almost 90 percent accurate,” said Baidoo-Essien, a TPS Intelligence Analyst [2].

Educators now have 72 different formats to view the data, and can even see metrics at the classroom level. “This tool was very good at helping us build some analytics to see student performance from a near-time historical perspective,” says Shaun Taylor, the district’s CIO [2].

What has TPS done with this new information?students-and-teacher

 Because teachers and school administrators now have up-to-date information on how students are performing across a spectrum of standards, educators can more effectively think about how to help students advance. Almost instantaneously, teachers can identify if a student is falling off track and implement a multitude of targeted intervention strategies. A personalized message, a one-on-one or group tutoring session, peer-to-peer mentoring, and meetings with a teacher or adviser are just a few action steps teachers can take [4]. In sum, these points of intervention are what push at-risk students across the graduation finish line prepared for success in their post-high school educational path and beyond.

To share the success of the model within the district benchmarks and actions plans are discussed at school board meetings. Furthermore, takeaways from these meetings are shared with community members as well as broadcasted across the district [2].

The Risks of Using Predictive Analytics in Schools

Despite the clear benefits of using predictive analytics in TPS, educators should not be unconscious consumers of the data. Why? Because we have seen computer-generated predictions go very wrong. For example, a risk assessment in Broward County, Florida wrongly labeled black people as future criminals nearly twice as often as it wrongly labeled whites, according to ProPublica, an investigative journalism organization [5].

These errors lead to real problems for students and citizens and work against the very goal of using machine learning. As such, educators should remember that although data can provide a snapshot of information about student outcomes, each student is different and he or she should therefore not be blindly bucketed into categories.

Going Forward

The success of Microsoft Azure in TPS has encouraged school districts in Virginia and Wisconsin to adopt this cloud technology [1]. But there is still more to do and more question to answer:

  • TPS is just one of many “drop-out factories” around the country. If the cloud tools continue to have promising results, Microsoft and school districts should look into implementing Azure in the worst-off districts. Because early intervention is critical, perhaps elementary and middle schools could begin using this technology, too.
  • Teachers use a plethora of strategies to help students overcome the identified barriers to success. What are those strategies and can similar data analytics help us understand which strategies are most effective?

(762 words)

[1] Education News “Tacoma among U.S school districts recognized for smart data use” The Seattle Times, September 30, 2015,, accessed November 2016.

[2] Microsoft Customer Stories, “Predicting student dropout risks, increasing graduation rates with cloud analytics,”, accessed November 2016.

[3] Tacoma Public Schools, “Strategic Plan: Measuring the whole child,”, accessed November 2016.

[4] “Higher Ed’s Moneyball? ”, Eric Westervelt, All Things Considered, National Public Radio, October 14, 2015,, accessed November 2016.

[5] Data Analytics, “The power of learning,” The Economist, August 20, 2016,, accessed November 2016.


Making it rain…Chinese Red Packets


Groupon: A Good Deal for Anyone?

Student comments on Microsoft Helps Tacoma Public Schools Use Data Analytics to Predict At-Risk Students

  1. This is really inspiring! I love how Microsoft is leveraging technology to help educate the future generations in a scalable away. By empowering the teachers to provide much more tailored and personalized teaching curriculums and interventions, Microsoft Azure is significantly expanding a single teacher’s capacity and ability to make a difference. What’s also incredibly promising besides identifying students that are at-risk, by working with a school systems like TPS, they can collect longitudinal data on a student to see how their learning habits and potentials change over time. I think that is incredibly valuable because an intervention strategy of say peer-to-peer mentoring may work extremely well in 7th grade math but 11th grade english may require one-on-one tutoring sessions with the teacher.

    I wonder how integrating technologies such as laptops and tablets into the classroom will influence these predictive algorithms. The possibilities of data to be collected is endless. For instance, you can track how long a certain question takes to answer or use eye-tracking technology to see how long it took a student to read a certain paragraph. But when does too much data become a bad thing that just creates noise in the analytics?

  2. Really heartening to see data and predictive analytics being used for good rather than to track a consumer’s next purchase! Thanks for sharing!
    It does make you think though since a large part of a great teacher’s job is, in fact, to implicitly provide these predictive analytics and then use that data internally to customize lesson plans to each students’ learning strengths. One could look at technology like this and purport that it is just another way that workers will be replaced by technology. For example, if such predictive analyses can get good enough, one could envision online courses modifying themselves based on inputs from students to better customize the learning experience for each student. I like to take a more optimistic view though. I feel like such technology will only serve to augment teacher’s current intuitions allowing them to focus on students that need more attention and help.

  3. * obligatory joke about TPS reports *

    Thank you, really, for writing this up. I hadn’t heard about this initiative, and as tdubs and Varun mentioned, it is truly inspiring.

    The improvement in graduation rates you cited are incredible. What struck me most is the ability to use this data to assess and react to this data almost instantaneously (pull the andon cord!); academic issues, home life issues, self-esteem issues, and the like can snowball very quickly. The plural of anecdote is not data, but both of my parents are middle school teachers, and it’s sobering to hear stories about star students of theirs who quickly fell behind because of issues at home, issues with peers, and other instigating factors.

    I am genuinely excited about future applications of this technology, and I just sent this post to my parents – perhaps they can advocate for its use in Los Angeles Unified School District, which is in sore need of innovative ways to improve graduation rates. Thanks again.

Leave a comment