Disrupting the Construction Industry With Data Science


Anton working at GA's Hong Kong Campus.

Anson working at GA’s Hong Kong Campus.

By moving beyond analysis into prediction through data science, General Assembly Hong Kong’s DAT graduate Anson Au has brought unparalleled performance and efficiency to traditional practices in the construction industry.

Before coming to GA, Anson was already an avid learner, having completed both an MSc and MBA at HKUST. In his current role as head of IT projects at Alliance Construction Materials, he sought to use data and technology to improve the performance in the traditional construction materials industry.

The problem:

Anson’s company supplies concrete to construction contractors and sites. Concrete in HK is in very high demand, and often contractors will over-order or cancel last minute, which creates significant inefficiency and losses. Planning ahead is very difficult for these types of over (or under) ordering. The company had years of customer data and order history to draw on for analysis.

The company has a lot of experienced professionals in the concrete industry and Anson had conversations with the ‘old hands’ who often relied on gut feelings and hunches to predict the clients or orders that would be cancelled at the last minute. From a statistics point of view and as a data scientist, these are heuristics models that could be replicated through data analysis and techniques from data science. The key was getting access to the data.

Anson worked with the CEO and talked to him about taking the course and developing a solution to the business problem. The CEO was very experienced and always on the lookout to improve existing processes. He wanted to dissect, and decode the underlying logic of these given tricks and existing ‘rules of thumb’.

The solution:

Anson enrolled in the Data Science course and designed the algorithm with the first set of major statistical models covered, the Ridge regression. “The original algorithm developed in the first few weeks was frankly, crap.”

As he learned more about Scikit Learn and the other libraries and tools available through Python and the Anaconda package, a whole spectrum of choices was opened for him to test and develop. The algorithm takes into account internal customer data and order history, as well as external factors such as weather and climate.

Through the weeks, the algorithm was recalibrated and improved upon. “One of the great things that Mart, the instructor, did was show us how to approach a set of data and how to attack and analyze it,” Anson said. A large data set can be very intimidating but if one knows how to approach it and quickly get the basic facts from it, it makes the course of analyzing it a lot easier.

With the introduction of decision trees and the inclusion of random forest and grid search techniques, Anson struck upon a winning combination. “By cross validating the data 5 times, and with the regular daily updates of the data, we have trained the model to have an R-squared of 0.93 for predicting next day orders.” This is a very high correlation and the model has successfully helped save the company from hundreds of thousands of dollars from potential cancellations and losses.

What’s next?

Through data science techniques, Anson is currently optimizing the company’s raw materials usage by predicting the targeted materials required based on the concrete mix requirements and plant performance. With this project, he is still conceptualizing and writing code to test out theories and predictions.

He’s not quite done with GA courses yet either, as he is currently halfway through GA’s User Experience Design course and loving it. He and seven of his classmates recently won first place at Startup Weekend Hong Kong with a doctor-patient symptom communication application.

“We made use of techniques and strategies from class,” explains Anson, “and were able to show the judges how user research informed our business decisions.”

This was ultimately what pushed the team’s winning project ahead of the rest, he believes. The team are applying for startup accelerator programs now, and hope to bring this initial prototype into reality.

Armed with the potent combination both user experience and data science expertise, Anson is poised to lead both his work and startup projects to great success.

Explore Data Science at GA