A major commercial airline receives tens of thousands of safety reports each year, coming from nearly 100,000 employees, in various formats. It’s an analyst’s job to sort through all of it, assess the reports, and recommend appropriate action. Utilizing extensive data collected in their QMS, the airline has developed several predictive models based on machine learning to improve the speed and accuracy of the assessments, and the relevance of actions. The result is a significant improvement in operational efficiency.
In all types of industries, machine learning (ML) tools are finding the needle in the haystack of data, augmenting quality and safety professionals with a new kind of intelligence that can unlock hidden data patterns that are impossible for the human mind or eye to absorb.
As a subset of AI, machine learning automates the building of analytical models. With machine learning, the algorithm learns from the data it is trained on, to identify patterns and make decisions with minimal human intervention (although human reasoning will always need to be part of the equation).
Traditional data analytics integrated into a QMS provide all of the building blocks for effective use of data-driven decision making. Modern data analytics solutions for quality management systems contain a data lake that houses all critical data from the QMS as well as critical data from integrated systems such as a CRM, MES or ERP. It also offers dashboards to visualize the data and gain insight on trends related to critical issues such as non-conformance and corrective actions, root causes, safety management and injury analysis.
Yet, a quality management system (QMS) that leverages machine learning brings data analysis of quality and safety metrics to a whole new level of accuracy and insights. For example, today it’s fairly easy to use data analytics to identify issues, such as equipment that isn’t properly maintained, and take corrective action to fix those issues. It’s much more difficult to uncover systemic issues that fly under the radar but that may cause a significant risk, such as the fact that updated training was never conducted for the staff using the equipment. ML can do this because it makes relationships between assorted data, in multiple formats, such as rich text, video, images and others, which humans cannot do. Analytics can look for exact matches, but ML can make correlations between things that may never occur to humans.
It has been said that data analytics helps humans find the things they expect to look for, yet true machine learning can take it a step further, using intelligence to show them what they should be looking for.
One area where machine learning could provide immediate value in quality and safety management is for intelligent trending, or predictive analysis. Leveraging extensive historical data-sets, professionals can move beyond simple extrapolation of historic counts, and, using hundreds of related data points, actually predict what will happen in the future, with a cone of confidence (as opposed to the cone of uncertainty) that conveys the margin of error over time. It takes a wide-range outlook of future outcomes driven by data instead of hunches.
While a recent S&P Global report found that 95 percent of businesses consider AI to be important to their digital transformation goals, integration of machine learning into quality management processes is still nascent. The reality is that it will take time for it to see widespread adoption in quality management because it requires a systemic change in how companies operate, as well as skilled personnel to implement and optimize it.
Yet companies are smart to set out on their machine learning journey today, with the understanding that by taking small careful steps, many more miles can be successfully traversed in order to safely reach their destination: AI-driven quality management. Consider the following best practices:
1. Begin with a problem. Before deploying ML-driven quality management (or any ML project for that matter), identify what it is you are hoping to accomplish. Is it to predict the likelihood of product defects with great statistical accuracy, or to identify manufacturing floor behaviors that are creating corporate risk? Once you identify your goals you can work backwards to the solution. In some instances, you may realize that ML is not even the answer.
2. Get your data house in order. Good data – and lots of it – is required for effective data analytics, as well as machine learning. As a first step on your journey, it’s important to take a data audit, identifying where data, including structured and unstructured data, resides across the company, and making sure it’s cleaned and classified for use as training data for the algorithm.
3. Democratize the data and remove the siloes. Companies already are leveraging advanced analytics tools within their QMS to consolidate data into data warehouses and data lakes. Thanks to robust platforms, more and more data should be gathered across the enterprise and the supply chain and centrally shared to inform insights across functions.
4. Start with a pragmatic approach. Quality software vendors are touting advanced new AI capabilities – from ML to natural language processing (NLP), but the reality is that while technology is advancing, AI-driven quality is a very new concept. Companies aren’t fully set up yet to leverage it to its full potential, nor do they have the data science expertise in house to manage it. It’s wise to take a pragmatic approach and start small. By deploying intelligent trending, for example, which doesn’t require extensive training, they can add a ML layer to the data lake, and forecast future outcomes with very little risk. When quality leaders become comfortable with how ML works at a basic level, they can pursue more advanced applications.
5. Secure the talent. While ML is being integrated into quality management systems and will eventually become a standard feature, it will always require the expertise of data scientists for proper management and continuous data maintenance. This is new territory for quality professionals; collaboration and coordination with data science experts will be integral to effective deployment.
Good quality management will always require the insights and expertise of humans in order to achieve success, since risk-based decisions require a level of thinking and abstraction, which is highly subjective. Yet, by augmenting human-centric intelligence with machine learning, quality management can become more data-driven, combining human insight with the objectivity of AI – a winning combination for a whole new paradigm in quality.