Synaptiq.ai

Predicting Project Profitability

Predicting Project Profitability

architecture-blue-build-162557.jpg

 

The Problem

Our client is one of the largest, most diversified construction companies in the United States. This company has over 6,000 employees and provides nearly every type of civil, commercial, and industrial construction service. In addition to having many subsidiary operating companies with a wide variety of expertise, this construction group is the originator of many major state highway interchanges.

For several decades, the firm relied on its highly experienced project managers to identify and map project risks to profitability. This project-related data was collected in various disconnected systems. But these experienced project managers’ collective “tribal wisdom” had never been assembled or codified in a way that allowed business development to make efficient connections and analyze future growth. Furthermore, as project managers retire out of the company, and with them their project knowledge, business planning risk goes up .

The CIO and Project Management team identified an opportunity to apply innovative thinking to project management and business planning.

The Solution

The Synaptiq team began the engagement by interviewing the firm’s experienced project managers to establish a baseline of understanding regarding project risk projections. We learned that the project managers had formed rules to identify project risks. These risks, called “leading indicators”, essentially project milestones related to anticipated delays, help to predict project risk and associated losses or cost to the company.

Collectively, the project managers articulated five leading indicators for the company’s projects over all. We used these indicators to create a model based on past project data. The model revealed the most promising leading indicators and helped validate assumptions made by project managers. When diving into past project data, we discovered that tasks and overall budget were tracked in separate tables. We needed a bridge table to connect the data, which led to a substantial data engineering effort.

We provided a spreadsheet for the CIO and Project Manager’s team to join data rows from the separate tables manually. Once we obtained connected data, we performed statistical analysis to understand the distribution of values for each key field and how many had null or erroneous values. We used this information to clean the data in preparation for building a predictive model.

In addition to building the foundations for the model, we also learned that the leading indicators’ names were entered into project plans using different variations per project. This afforded us another opportunity for improvement by unifying these variations into canonical names for downstream processing.

The Results

We built a machine learning model to predict project profitability from the values in the leading indicators, i.e., delays in key project milestones. A key element was finding the right profitability threshold to define what profits are acceptable and unacceptable. We worked with the CIO and Project Management team to incorporate domain knowledge into visualizations of past profits. The final result was a classifier that predicted whether project profit would be “acceptable” or “unacceptable” with 86% accuracy.