How AIG Moved Toward Evidence-Based Decision Making
New developments in data science offer a tremendous opportunity to improve decision-making. Machine learning, pattern recognition, and other predictive analytics tools can constitute a source of competitive advantage for those companies that adopt them early on; but like any new capability, there is an enormous gulf between awareness, intent and early engagement, and achieving significant business impact.
How can companies better manage the process of converting the potential of data science to real business outcomes? How can companies go beyond merely generating new insights to changing behaviors — not only of their employees, but customers too? We would like to offer some lessons from AIG’s early experiences with deploying new analytical tools to leaders across industries who may be considering embarking on a similar journey.
In January 2012, AIG launched the “Science Team.” One might be surprised to find a Science Team in an insurance company. However, Peter Hancock, President and CEO of the global insurance giant, saw a huge opportunity to apply evidence-based decision making in an industry which was still very reliant on individual expert judgment and in so doing to create not only tactical but also competitive advantage. By early 2014, 130 people from diverse scientific and managerial backgrounds were devoting themselves to realizing the team’s mission: To be a catalyst for evidence-based decision making across AIG.
The Science Team intentionally refrains from using the words “data” or “analytics,” as the team’s capabilities stretch far beyond these two disciplines: behavioral economists, psychologists, engineers, and change management experts work hand-in-hand with data scientists, mathematicians, and statisticians. And for good reason: this multidisciplinary approach is essential to go beyond merely generating new insights from data but also to systematically enhance individual human judgment in real business contexts. Ninety percent of the team was recruited from beyond the insurance industry to enable it to challenge the status quo approach to decision-making. The Science Team not only prepares data and builds models, but also emphasizes the identification of business opportunities and education, change management and implementation—the complete value chain from framing questions through to changing behaviors.
Key factors in the success of the Science Team’s efforts to date include the following:
Start by focusing on questions and problems that matter. A small proportion of worker’s compensation claims account for a large proportion of complexity, contention, delay and losses for AIG: 10% of claims account for almost 60% of costs. Claims severity predictors therefore play a huge role in improving outcomes by enabling earlier and more accurate targeting of intervention measures like physician review and special investigations. This is a good example of the power of fully embedding the technical solution in the business: the result is not only better predictions and lower costs, but also better outcomes for customers.
Ensure that the mandate stretches beyond producing insights — supporting the change and learning process across the organization. AIG not only supports embedding solutions and managing change to realize specific opportunities, but has also launched a company-wide initiative to improve quantitative and decision-making skills using both physical summits and on-demand, modular online learning tools.
Work with early adopters to demonstrate significant wins which are visible to the whole organization. Much of AIG’s business relies on agents and brokers. Relationships are assessed and prioritized based on volume, value, potential, and their overall effectiveness. The decision platform which AIG built is able to accurately predict the retention and “submission” (proposal) efficiency of single brokers — a level of micro-segmentation and prediction which few others in the industry have been able to achieve. Every day, aggregated and deep-dive performance analytics, presented in a user-friendly visual format, are pushed to the fingertips of sales managers to support decisions on how to manage the network of intermediaries.
Don’t make the effort dependent on one or two initiatives: adopt a portfolio approach. In pioneering new approaches to decision-making not every effort can be a success and companies should therefore not bet only on the success of one project. In addition to the examples above, AIG currently has around a dozen decision making related projects at various stages of development.
An iterative, rapid cycle adaptive approach is much more effective than a planned, single step change — much of the learning occurs by taking action. Preventing fraudulent claims is an important area for AIG due to its significant financial impact. AIG has developed proprietary tools and models that identify predictive patterns in claims data using machine learning, predictive modeling, link analysis, pattern analysis and other techniques. After starting from scratch, the second generation of AIG-developed tools already identify almost twice as many cases of fraud than leading vendors’ offerings. First applied to worker’s compensation, the same approaches are being now being rolled-out across multiple businesses. This example illustrates the importance and power of an iterative, learning-based approach to solution development. Ironically, this involves a bias to action rather than planning or analysis — even in the area of analytics!
Plan for impact on multiple time-horizons, combining immediate evidence of value, some medium term big wins as well as a transformational long term perspective. In addition to the short- and medium-term solutions mentioned above, AIG is also contemplating some bolder, longer term initiatives which could potentially change the business model and the scope of the business. For example, it is looking at possibilities like assessing damage claims for auto accidents using image analysis of photographs, or measuring and modulating risk assessments using sensors and telematics.
The constantly evolving tools of data science will both enable and require companies to continue to improve how they make decisions. It’s self-limiting to only improve existing decision-making, however — companies need also to be alert to the opportunity of creating fundamentally new ways of making decisions, and even to reconsider new the business models and the firm’s activity footprint, as a result of the opportunities unleashed.