American Water (NYSE:AWK) is the largest American publicly traded water utilities company with a market capitalisation of $16.2b (as of Nov 9, 2018) and annual revenue of $3.4b (FY2017). The company processes over 1 billion gallons of water daily and offers water and wastewater services to approximately 15m people in 1,600 communities in 16 states. AWK operates thousands of water infrastructure assets, including but not limited to water treatment and storage plants, wells, dams as well as a network of over 50’000 miles of water pipes .
Being the largest American water utility AWK is uniquely positioned to be at the forefront of technological change within the water utilities industry. In its 2017 annual report AWK’s management states that the company has a three-year technology and innovation plan to “leverage secure artificial intelligence to better serve customers” .
The company does not describe in detail which aspects of serving customers it specifically intends to target with AI applications, yet the potential use cases for a utility services provider could be as diverse as automating customer service and predicting business effects of global warming [3,4]. Most importantly machine learning/AI could be a key performance lever for improving infrastructure management practices through better monitoring and replacement work scheduling .
The management of infrastructure is an issue that sits at the forefront of every utility business. The investments are massive – AWK plans to spend up to $7.2b in infrastructure related capital expenditures from 2018 to 2022 . Anything that could better predict the infrastructure replacement needs and potential risks within the water delivery system could therefore prove to be transformational from a business cost perspective.
AWK management has come up with a strategy and a special focus group within the company that addresses digital transformation. However, it appears that efforts are still at their nascent stage . The quarterly earnings call in August, 2018 revealed that the company has introduced drone technology to monitor the exterior for some of its infrastructure. The COO Walter Lynch remarked that such visual inspection is not only safer for the workers but takes only one eight of the time manual inspections used to .
Using drones instead of manual inspection is simply the beginning of what the company could do short term to improve surveillance of its open air infrastructure. Machine learning algorithms have already been applied on drone image data to generate maps of normality and automate screening for changes in physical assets (e.g. ). Such automation, if implemented at AWK, could allow for increasing the frequency and accuracy of infrastructure surveillance. In the medium term applying machine learning algorithms on satellite data would enable a fully automated, daily asset inspection that could detect damage or deterioration in near real time (see e.g.  – application of the technique for oil spill detection).
Furthermore, the company could use machine learning to improve monitoring of its ageing network of pipes. Currently piping replacement schedules are dictated by age which is an unreliable measure of true durability. The thickness of a pipe, the pressure, the soil in which it rests and climate are all factors that affect durability and the true wear down is not straightforward to predict . Leaks and breakages that occur due to aged pipes not being replaced can lead to severe loss of pressure and interruptions in water service for the network parts affected as well as flooding in individual homes and even neighbourhoods.
Most recently, there have been successful attempts at predicting water breaks by using past breaks data and machine learning techniques (e.g. [5, 10]). From the outset it appears that such approaches are general enough to be quickly deployed also on AWK’s network.
In the long term the company could benefit from installation of IoT sensors or crawling robots that coupled with machine learning algorithms would measure the true condition of pipes. Such combination could detect and eventually predict leaks with high precision, rank the network components according to their need for replacement and eventually unlock features that appear to predict pipe durability for best future replacement practices [3,4,5].
These are just two examples of potential machine learning use cases within AWK that demonstrate that there are significant productivity gains to be unlocked by applying machine learning to current AWK infrastructure management practices. The machine learning technology is becoming mature, but the business transformation within traditionally non-digital operations will require growing significant expertise in applied machine learning and related hardware technology .
While AWK has displayed some commitment of becoming a digitally enabled business, whether the company will unlock its full potential remains to be seen. I’d be curious to hear my classmates’ thoughts on what key barriers will AWK and other utility businesses face when introducing machine learning into their operations.
 CapitalIQ NYSE:AWK company summary. Retrieved 11 Nov, 2018.
 American Water (NYSE:AWK) 10K Filing for FY2017. Retrieved 11 Nov, 2018.
 McKinsey’s Digital Utility Compendium (2018). The Digital Utility: New challenges, capabilities, and opportunities.
 Cognizant 20-20 Insights (2017). Digital Disruption in the Water Utility Value Chain.
 Slavko Velickov (2017). Machine Learning in Utilities and Water Industry: Making Digital Ripples. https://www.linkedin.com/pulse/machine-learning-utilities-water-industry-making-digital-velickov/
 American Water October, 2018 Investor Presentation. Retrieved 11 Nov, 2018
 American Water 2018Q2 Earnings call transcript. Retrieved 11 Nov, 2018.
 Narayanan, P., Borel-Donohue, C., Lee, H., Kwon, H., & Rao, R. (2018, April). A real-time object detection framework for aerial imagery using deep neural networks and synthetic training images. In Signal Processing, Sensor/Information Fusion, and Target Recognition XXVII (Vol. 10646, p. 1064614). International Society for Optics and Photonics.
 Singha, S., Bellerby, T. J., & Trieschmann, O. (2013). Satellite oil spill detection using artificial neural networks. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 6(6), 2355-2363.
 Kumar, A., Rizvi, S. A. A., Brooks, B., Vanderveld, R. A., Wilson, K. H., Kenney, C., … & Ghani, R. (2018). Using Machine Learning to Assess the Risk of and Prevent Water Main Breaks. arXiv preprint arXiv:1805.03597.