Apply OKRs to your Data Science project

Andrew Ng, the famous AI-pioneer, teaches in his Deep Learning Specialization that every Data Science project should start with defining a measurable metric. The goal of the project is to fulfill this metric. This goal steers the actions of your project.

Metrics like accuracy, precision and recall, or F1-score are common Data Science metrics. Each metrics offers distinct advantages and disadvantages depending on your business case. Accuracy might not be the best metric when you have a highly imbalanced dataset. Would you compare your algorithm to Human-level performance? How important is detection speed?

Pick the most meaningful metric to define whether you’ve solved the business case or not. Analyzing the pros and cons of each metric is beyond the scope of this article, but you can read more about it here. Defining the key metric will prevent you from doing research while you should create a prototype or continuing to solve an issue when the metric has been already been reached.

After defining the key metric, we are closer to defining our Data Science OKRs. In many projects, it makes sense that the key metric is a part of the Objective. It shows where we want to go. Next, we need to define Key Results to display how to reach the goal.

Let’s say you’re working in the automotive industry. You identify the need to detect pedestrians in an urban surrounding to warn truck drivers. You decide to develop a driver-assistance function for trucks to identify pedestrians accurately. The truck chassis shakes heavily, hence general pedestrian detection models don’t work well enough. The team agrees that a 98% detection rate is a proper stretch-goal for the first quarter.

Next, you decide that you need a data set with at least 10.000 labeled images. You will need time to do research and to implement a first prototype. Lastly, you need time and resources to iterate until you reach the goal. Let’s transform the information into OKRs.

These OKRs guide your work on the pedestrian detection project in the next quarter. You review your OKRs at the end. Did you manage to achieve your key results? If you missed them, how so?

Applying the OKR method to your Data Science project will keep you on track for successful project delivery.