Want to learn more about the technology and science that powers our Termination Compensation Calculator and especially, how machine learning can help to predict how judges determine ‘reasonable notice’ in wrongful termination cases? See our paper in The McGill Law Journal by Professors Samuel Dahan and Jonathan Touboul, alongside graduate students Dan Sfedj and Jason Lam.
The Current Landscape
In Canada, an employer has the right to terminate an employment contract without cause at any time so long as they provide their employee with reasonable notice of termination. In practice, this often results in the employer paying the employee the amount they would have earned had they continued to work after the notice date rather than having the employee continuing to work (known as “payment in lieu of notice”). While the Employment Standards Act (“the ESA”) outlines the minimum notice period that terminated employees are entitled to, employees are often entitled to a notice period longer than the minimum. Determining how long this reasonable notice period is, or how large this payment should be, is a highly contextual and complex undertaking which requires consideration of numerous factors.
The Conflict Analytics Lab
Rapid advances in data analysis techniques, particularly predictive algorithms, have opened radically new perspectives for legal practice and access to justice. Professor Samuel Dahan and Professor Jonathan Touboul have begun utilizing these new techniques to tackle issues within the Canadian legal system. Professor Dahan has established the Conflict Analytics Lab (“the Lab”), a joint-venture of the Queen’s University Faculty of Law and the Smith Business School, with the aim of using data science and machine learning to aid in dispute resolution. The Lab involves collaborations across multiple disciplines and is advised by various leaders in industry and academia. The Conflict Analytics Lab seeks to release these technologies to the public through a free open-source platform to improve access to justice in Canada.
Reasonable Notice Period Prediction Algorithm
One of the Lab’s main projects utilizes Artificial Intelligence (“AI”) methods to predict reasonable notice periods. This will help individuals navigate through a complex legal system and improve their bargaining position with employers. This tool would also help small employers seeking to understand their legal obligations and potential liabilities.
A determination of reasonable notice requires a contextual assessment of certain factors. These factors include, but are not limited to:
1) The employee’s length of service,
2) the age of the employee,
3) the character of employment, and
4) the availability of similar employment.
The Lab applied statistical methods to better understand how important each factor is in determining the length of the reasonable notice period. The research found that the reasonable notice period is strongly correlated to the length of service of the employee: there was an approximately 77% positive correlation. The Lab also found that each of the other enumerated factors also has a statistically significant impact on notice period determinations.
With this information, the Lab proceeded to use AI technology and statistical methods to create algorithms that could try to predict the length of notice periods. The Lab tested multiple techniques, but ultimately determined that the “decision tree” method was most successful. This method yielded an average error of 1.89 months when predicting reasonable notice period length. Further, 25% of cases were accurately predicted to within 0.88 months, and 75% of cases were accurately predicted to within 3.90 months.
The Lab further analyzed cases in which its AI system’s predictions were significantly different from the court’s decision. This investigation led to the conclusion that the inaccurate results were almost always caused by a set of circumstances that led the judge to determine the reasonable notice period without consideration of the factors listed above. These cases accounted for approximately 8% of the dataset analyzed by the Lab. The relative inaccuracy of the predictive algorithms for these cases can be explained by the power of judges to look beyond the above factors to make a more just determination of what the reasonable notice period should be.
Although the technology is unable to predict reasonable notice lengths with absolute statistical certainty, it still provides important insight into how these notice periods are determined. The slight uncertainty is indicative of a system that strikes a good balance between predictability for employees and employers, and remaining sufficiently flexible to address the inherent limitations of the usual rules.