By Elena Belov and Marc Wagner
What do the Volkswagen emission scandal and the subprime mortgage crisis have in common? They both relied on deceptive algorithms. In the case of Volkswagen, software embedded in the engine evaluated if the car was being tested. If so, the software changed a complex system of engine settings to reduce emissions. During regular driving, however, the settings were automatically changed back to become more fuel-efficient—increasing emissions. In the subprime mortgage crisis, algorithms were applied that led to better credit ratings for securitized assets than were economically justified. In both cases, an independent expert review of the analytical methodologies and their application may have alerted senior management to the misuse.
The challenge for executives is to stay on top of all the analytics used in a company, their implicit and explicit assumptions, and how exactly insights are derived.
Effective use of analytics has the potential to make significant positive contributions to business, such as Amazon tools that analyze a buyer’s previous interests, product ratings, and purchasing behavior to issue recommendations. But, executives often struggle to understand the exact mechanism of how data is translated into decisions by their organizations. This lack of understanding can lead to a false sense of confidence in decisions, or decisions that are sub-optimal for the company.
The challenge for executives is to stay on top of all the analytics used in a company, their implicit and explicit assumptions, and how exactly insights are derived. This is particularly difficult if insights are being derived from multiple models and analyses with different approaches and independent assumptions.
Learning from the financial services industry
Other industries can learn a lot from the financial services industry and its efforts to govern the use of algorithms and models. After the subprime mortgage crisis, banks, rating agencies, and investors started improving the quality of their analytics governance, and their management of complex models and model interdependencies, including model use and implementation.
Governance: Banks significantly invested in model governance structures. Applying a “two pairs of eyes” principle with independent review, financial firms track who develops algorithms, how testing and implementation is performed, and who oversees these activities. Developers are required to write sound documentation that lays out approaches, assumptions, and data used in terms a layman can understand. Dedicated independent parties then review these models and documentation. A feedback loop compares model outputs against actual results, and any differences need to be explained and addressed.
If the analytical application is critical for the business—as in investment advice—it is also vital to establish strong controls, limiting access to the analytics engine and protecting it from changes.
Complex models and interdependencies: Another lesson learned from banks is the need to understand a company’s ecosystem around analytics. This includes both internally and externally developed analytics, the data being used, and how it is derived. If “bugs” and glitches in data and analytical models from third parties have the potential to ripple through company models, the impact needs to be understood.
Model use: Lastly, senior management needs to understand and guide how the outcome from analytics is transformed into decisions. Decisions need to be made consistently across users of analytical outputs, solely based on the specific business context in which analytics are applied.
Choosing the right approach for your company
Companies in all industries need the right governance, understanding of the model ecosystem, and model-use processes to capture the most value from analytics and avoid inaccurate predictions or bad decisions. The actual approach varies, depending on the business model and environment. Some companies approach analytics with a “fail fast and improve quickly” mindset. Companies in which analytics are critical to business sustainability cannot afford to choose this approach, and will need to invest in more rigorous understanding of analytics used–just as financial institutions do now.