Project Risk
Prev Next

Project Risk

Risk emerges from the combination of uncertainty and the utilitarian effect of the event's outcome. In most cases it has both a downside (higher cost) and an upside (lower cost).

Risk effects are asymmetrical. Avoiding risk can be good for managers but bad for shareholders. For a project manager both sides of risk are seen as marks of incompetence (downside because of the supplemental cost, upside because the unneeded resources were taken away from other projects). However for a company to completely eliminate risk is generally too expensive to be worth it.

Due to risk's nature a manager has only three levers to tackle it by lowering the vulnerability, the threat, or the consequences.

In Data Analytics we analyze two generic categories of risk:

Implementation Risk

To estimate the Implementation Risk one can make an educated guess or (if possible) analyze the history of forecasts to describe the outcome's probability (risk profile). In the example from the figure below the projected cost of $4M is at 45% the most likely outcome. Here the right side of the graph represents the downside (cost higher than prediction) and the left side the upside (lower cost).

Projected Risk Profile

From the forecasts' history risk can be estimated using the formula:

Risk Formula

Where xi represent the outcomes and pi the associated probabilities. In our case the risk was valued at $862,612.

Usage Risk

An analytic system takes in the primary data and generates another set of data representing an insight derived from raw data. The insight can be an explanation of past events (Descriptive Analytics), a concise description of the most relevant facts (Data Visualization), or a prediction of events outside the range of existing primary data (Predictive Analytics). In all of these cases the insight may not always be perfect thus laying the foundation of the Usage Risk.

Because the utilitarian value of insight has a very complex structure (non-linear and multi-dimensional) a comprehensive risk measurement formula as the one presented above is generally not feasible. The most common approach is to divide the range of possible values and process the data from the quasi linear domain using formulas similar to the one above. The remaining range could be analyzed using sophisticated mathematics or other techniques as "what if analysis", game theory, chaos theory, etc.