AI - From big bang to business outcomes:

PAVING THE WAY FOR ARTIFICIAL INTELLIGENCE'S REAL VALUE

Sensors are everywhere.

In 2012, 4.2 billion sensors were shipped for industrial use; in 2014, that figure skyrocketed to 23.6 billion.¹ The Internet of Things has changed the industrial landscape, promising improved efficiency and production for everyone from shoe makers to milk processors to refineries to power plants.

It’s clear that companies do not need more data, however. In fact, “70% of captured production data goes unused.”² Nor do companies want more data. Instead, they want to extract the trapped value that already exists by leveraging collected data to solve specific problems in much more proactive ways. Analyzing static graphs and charts no longer is sufficient in our ever-changing digital economy.

Here is where artificial intelligence (AI) can make IIoT investments pay off ($105 billion globally in manufacturing operations and another $45 billion in production asset management in 2017³) — by giving companies dynamic tools to make better business decisions. That’s the beauty and magic of AI.

70%

of captured production data goes unused

Transforming industry

AI’s value in the industrial space is undeniable. It has the potential to skyrocket rates of profitability in manufacturing by an average of 39 percent by 2035⁴, for AI brings a fundamentally different approach to decision making that ultimately will produce better results. It uses data to learn patterns that may not be obvious to a labored human eye faced with hundreds of parameters and inputs vs. just a handful of data points in the pre-digital days. As a result, AI presents a great opportunity to augment the essential human expertise of asking the right questions based on the specific needs of the environment and context. This learning is placed into a trained model, which can be deployed as close to the action as possible, transforming both the rate and the accuracy of prediction and decision making.

Rather than relying on rules encoded by humans, AI uses data to learn patterns that may not be obvious to a labored human eye.

But general AI models are not enough; instead, models must be applied specially to realistic applications, including energy management, asset performance, and operational productivity, to make AI worth the financial and human resource investments — not to mention the time. Bear in mind that AI technology is just the foundation. Actionable insights gleaned from continually training and re-training models hold the real key to the data-driven decision-making behind valuable business outcomes.

For example, Rolls-Royce has 13,000 commercial aircraft engines in service around the world. Microsoft enables Rolls-Royce to improve maintenance more proactively and precisely. First, Rolls-Royce collects and aggregates data from disparate and geographically distributed sources at an unprecedented level. Then Microsoft enables Rolls-Royce to analyze that data and perform data modeling at scale to accurately detect operational anomalies. The value here is that customers can plan relevant actions in real-time, using pilot-friendly dashboards to inform on-the-spot decisions and operations.

The science of data science

We want to be clear that AI is a process, a science. As Johns Hopkins University’s biostatistics professor Jeff Leek proclaimed “way back” in 2013, “The key word in data science in not ‘data’; it is ‘science.’ Data science is only useful when the data are used to answer a question.”⁵

“How can I constantly improve efficiency while ensuring uptime?”

In the industrial space, that question often is, “How can I constantly improve efficiency while ensuring uptime?” AI can answer this question with data-based models made to predict outcomes such as “When will this asset fail?” Or “Where and when are we wasting energy?” But therein lies the challenge. As analyst and consultant George Anadiotis points out, “[B]uilding the right thing is fundamental, even more so than building the thing right.”⁶ Critical operations and industries demand accuracy, so investing in experimentation is crucial for building the right models, which always will be as dynamic as the human intelligence they are meant to emulate.

Data scientists don’t always know what any given AI model’s outcome will be, as outcomes depend on how predictive the data are. AI models therefore must start with a certain level of accuracy and improve over time and, in turn, be re-trained, re-versioned, and re-deployed within situational context. MIT’s Michael Schrage says it best: “Empowering algorithms is now as organizationally important as empowering people.”⁷

Schneider Electric and Microsoft both understand the importance of this type of empowerment. Schneider Electric is invested in developing predictive analytics and condition management tools, for example, to enable customers to predict failure long before downtime actually happens. Likewise, Microsoft has invested in a broad AI portfolio of tools and services to build, deploy and manage AI anywhere, including Microsoft Azure Machine Learning, Microsoft Cognitive Services, and others.

Despite this exciting promise of AI applications, there are several roadblocks to realizing the true power and value of AI adoption in the industrial space.

Three challenges of AI adoption

We know that using AI can answer business-critical questions on a daily, even hourly, basis, but companies must leap over three current roadblocks:

01

Data preparation and operationalization

AI, in the form of models, often requires clean data from a diverse set of sources to create the most accurate AI. This is important both at the AI creation stage and through operationalized data pipelines in deployment. Currently, data scientists spend about 60% of their time cleaning up and organizing data⁸ as “data janitors”⁹ before they can even think about analyzing that data. This data wrangling imposes on their time to experiment, re-train, and re-deploy models.

Remember that the quality of the model is only as good as the quality of the inputted data. One new signal can change a model’s course drastically, thereby making experimentation essential to ensure the promise of AI comes to fruition with concrete relevance for the business. In “How to Create a Business Case for Data Quality Improvement,” Gartner estimates that “the average financial cost of poor data quality on organizations is $9.7 million a year.”¹⁰

02

AI lifecycle management

As companies increase their investment in AI, the need for AI lifecycle management tools is rapidly becoming critical. Data acquisition and preparation, experimentation, versioning, dependency management, deployment into production systems, monitoring, security, compliance, and model updates all become key elements when running AI-driven systems at scale.

Industry has the potential to automate the full cycle from the point where data are generated and collected to the point that AI models are built and operationalized. Industry can create systems that are more efficient than many other segments because of the level of coverage that can exist throughout its systems. The average project takes 6-9 months to go from concept to production. Once in production, data can become useful the instant data points are generated and fed into trained models that will use that data to make predictions. Re-training based on new data can happen any time, but it often will follow the cycle that fits best with the business depending on the model and situation.

03

Making AI meaningful across organizations

To realize the full value of AI adoption, companies must expand the discoverability and use of AI from the data scientist to the other roles across the organization. In other words, models need to become much easier for developers, business analysts, and business decision makers to find, understand, and use.

One area we both see as transformative is edge intelligence. Gartner predicts that by 2022, 50% of data will be processed at the edge.¹¹ Together, cloud and connected edge are leading to magnanimous levels of change and acceleration in efficiency and accuracy. When models are deployed to edge devices — that is, closer to the decision making in action — they can be trained to be highly accurate to make decisions more quickly.

A Schneider Electric – Microsoft collaborative case study illustrates this value. Some businesses have remote assets that are not easily cloud-connected. Others may not want to send data outside their own networks. We solved these challenges for the oil and gas industry. Schneider Electric’s Realift^™ rod pump control leverages Microsoft machine learning capabilities to monitor and configure pump settings and operations remotely, sending personnel onsite only when necessary for repair or maintenance when Realift indicates that something has gone wrong. Anomalies in temperature and pressure, for instance, can flag potential problems, even issues brewing a mile below the surface. Intelligence edge devices can run analytics locally without having to tap the cloud — a huge deal for expensive, remote assets such as oil pumps.

The right predictive analytics models, moreover, can weed out false positives, which can be the case 99.99% of the time, thereby saving human and financial resources. For example, Schneider Electric can tell through AI when a solar array really has a problem as opposed to just an accumulation of dust or dirt, helping an operator know when to send a squeegee vs. a repair truck.

01

02

03 AI breakthroughs

Up to 50% reduction in energy costs in first season, as reported by farmers

For Schneider Electric, any AI application that delivers tangible business outcomes is a breakthrough. Our goal is to turn data into actionable insights. Powered by the Microsoft Azure platform and Schneider Electric’s EcoStruxure™ Industry IoT architecture, SCADAfarm is an integrated automation and information management solution developed for WaterForce, an irrigation solutions builder and water management company in New Zealand.

Schneider Electric, in collaboration with AVEVA, and Microsoft

increased visibility of irrigation system performance and status – for both farmers and WaterForce;
more efficient and effective water use;
up to 50% reduction in energy costs in first season, as reported by farmers; and
remote monitoring capabilities that reduce the time farmers have to spend driving to inspect assets.

As a solution builder, WaterForce now can offer additional value-add services such as fault diagnosis and performance benchmarking, driving forward its own digital transformation.

For Microsoft, offering lifecycle management of AI is a true breakthrough as well. Microsoft solutions can significantly cut time wasted on cleaning up and preparing data, enabling data scientists to do what they do best: data science. Security and compliance workflows (e.g., safety and regulatory reviews) can be integrated between experimentation and deployment of models, thereby feeding a most-valuable AI loop.

Up to 50% reduction in energy costs in first season, as reported by farmers

The promise of AI innovation, today

The closer industry can get to delivering on the core value its customers want when they want it, the closer AI’s innovative promise becomes a reality. As more aspects of the end-to-end industrial infrastructure become connected, and AI gets deployed throughout the systems, decision-making becomes faster. Better. And more profitable. Done right, AI will deliver

new business models that are more aligned with the customer, enabling them to shift from feature-centric products to value-centric services (i.e., selling outcomes rather than things);
dramatic efficiency gains as decision making and automation happen earlier before failure occurs, in turn opening the door for customers to create new products and deliver to new markets; and
new experiences through cognitive capabilities such as speech, vision, and language understanding that simplify and streamline customer and employee interaction.

As Schneider and Microsoft continue to work together to remove roadblocks, while embedding cybersecurity safeguards at every step, the future of AI is beyond exciting. It already holds real value as it drives tangible business outcomes, today.

Turn data into action

Connect

Connect everything from the sop floor to top floor.

Collect

Capture critical data at every level, from sensor to cloud.

Analyze

Convert data into meaningful analytics.

Take action

Drive action through real-time information and business logic.

30+ years of Schneider Electric and Microsoft co-innovation

Together, we're helping customers benefit from Schneider Electric's deep domain expertise and Microsoft's trusted, secure cloud.

Discover our co-innovation

Co-authors:

Lance Olson

Partner Director of Program Management,
Cloud AI Platform, Microsoft

Cyril Perducat

Executive Vice President, IoT &
Digital Transformation, Schneider Electric

Share the Story:

References:

¹ Elfrink, Wim. “The Internet of Things: Capturing the Accelerated Opportunity.” Cisco Blog, October 15, 2014. http://blogs.cisco.com/ioe/ the-internet-of-things-capturing-the-accelerated-opportunity. Cited in World Economic Forum, in collaboration with Accenture, “Industrial Internet of Things: Unleashing the Potential of Connected Products and Services,” January 2015. http://www3.weforum.org/docs/WEFUSA_IndustrialInternet_Report2015.pdf

² Technology and Innovation for the Future of Production: Accelerating Value Creation, World Economic Forum, March 2017, http://www3.weforum.org/docs/WEF_White_Paper_Technology_Innovation_Future_of_Production_2017.pdf

³ IDC, “Worldwide IoT spending in 2021,” IDC’s Semiannual Worldwide Internet of Things Spending Guide, 2H16 update, May 2017.

⁴ “How AI Boosts Industry Profits and Innovation,” by Mark Purdy and Paul Daugherty. Accenture Research, June 2017. https://www.accenture.com/t20170620T055506__w__/us-en/_acnmedia/Accenture/next-gen-5/insight-ai-industry-growth/pdf/Accenture-AI-Industry-Growth-Full-Report.pdf?la=en

⁵ Jeff Leek, “The key word in ‘data science’ is not data, it is science,” December 2013. https://simplystatistics.org/2013/12/12/the-key-word-in-data-science-is-not-data-it-is-science/

⁶ George Anadiotis, “Data to analytics to AI: From descriptive to predictive analytics,” ZDNet, November 23, 2016 http://www.zdnet.com/article/data-to-analytics-to-ai-from-descriptive-to-predictive-analytics/

⁷ Michael Schrage, “4 Models for Using AI to Make Decisions,” Harvard Business Review , January 27, 2017. Business Review. https://hbr.org/2017/01/4-models-for-using-ai-to-make-decisions

⁸ CrowdFlower, “2016 Data Science Report,” http://visit.crowdflower.com/rs/416-ZBE-142/images/CrowdFlower_DataScienceReport_2016.pdf

⁹ “Data janitors” used by Josh Wills, as cited in Jessica, Leber, “In a Data Deluge, Companies Seek to Fill a New Role,” MIT Technology Review, May 22, 2013. https://www.technologyreview.com/s/513866/in-a-data-deluge-companies-seek-to-fill-a-new-role/

¹⁰ Susan More, Gartner, “How to Create a Business Case for Data Quality Improvement,” January 9, 2017. http://www.gartner.com/smarterwithgartner/how-to-create-a-business-case-for-data-quality-improvement/

¹¹ Rob van der Meulen, "What Edge Computing Means for Infrastructure and Operations Leaders," Gartner. October 18, 2017. https://www.gartner.com/smarterwithgartner/what-edge-computing-means-for-infrastructure-and-operations-leaders/