Big international assessment organisations like Cambridge Assessment have long held considerable amounts of data. When text is produced digitally, we can do more with it. Surprisingly to many, there have been examples of automatic scoring of extended writing for around 20 years, though what works well in one context may not be applicable in all others. High-stakes tests of writing usually restrict automatic marking to providing a "second opinion" for comparison with human markers. There is no well ­defined dividing line between big data and ordinary data, but big data is often considered to have three characteristics. In addition to "volume," it has "variety" - encompassing text, video, images, log files of, for example, the key strokes and mouse clicks which students make as they engage with a computer-­based test, and of the time spent focused on each task - as well as structured data, like marks and grades. This data might be streamed for analysis in almost real time at high "velocity," which is the third characteristic of big data. Machine learning has had some notable successes recently. Essentially, machine learning is statistical modelling rebranded and applied to automation. As we collect data from more frequent testing, we will accumulate rich longitudinal data, reflecting each student's learning trajectory. Big data, the convergence of teaching, learning and assessment, and the increasingly sophisticated operationalisation of machine learning and of data science more generally, are creating real opportunities for improving our understanding and practice of education.