A leading US construction company partnered with us to develop machine learning models that predict project costs and durations with maximum accuracy. Through advanced data science techniques and survival analysis research, we helped them replace manual forecasting processes that took up dozens of hours per project with automated predictions.
The Story of our Partner
Our partner is one of the largest construction firms in the United States, with over $4 billion in annual earnings. They employ more than 3,500 people.
In the modern digital age, our partner has made remarkable progress in transforming itself into a data-driven organisation, recognising that leveraging data effectively would be crucial for maintaining their competitive edge.
The Opportunity
In the construction market, winning projects often comes down to who can price most accurately while maintaining healthy margins. Our partner needed to predict future costs for their construction projects to improve financial planning and win new business.
The company had assembled a data engineering team and a small data science team, but they sought more specialised expertise for complex predictive modelling to complement their team. Manual forecasting processes were taking up to 65+ hours per project and consuming valuable resources that could be better allocated elsewhere.
Beyond that, they realised that improved forecasting accuracy was impacting their ability to price competitively, and their sales teams lacked the reliable data they needed to win more projects.
Our partner recognised an opportunity to optimise their forecasting methods and improve their profit margins as a result.
The Dreamix Solution
We started with a comprehensive review of the existing data pipeline, identifying areas for improvement. Working closely with business stakeholders, we continuously refined business filters to align with the company's evolving priorities.
We developed an automated system that delivers consistent, reliable predictions. Our solution tackles the problem in two phases: first predicting how long projects will take, then forecasting costs over the following 12 months.
We built a completely new forecasting engine using advanced machine learning techniques. We developed over 100 features that help predict costs more accurately than human experts, while processing information from multiple internal systems to ensure predictions reflect real-world conditions.
A key innovation was creating standardised cost templates that serve dual purposes. We did that by adapting state-of-the-art academic research to meet specific business needs. The templates provide a foundation for accurate machine learning predictions while giving sales teams reliable starting points for new project proposals. This means sales teams can respond to opportunities faster, and maintain pricing discipline.
We also addressed the significant data quality challenges typical of large-scale construction companies. By implementing smart data processing and working closely with business stakeholders, we ensured the system produces reliable results despite inconsistencies in historical data.
The Results
Our work together has transformed how our partner approaches project planning and sales. Projects that currently require up to 66 hours of manual forecasting would receive reasonable predictions instantly, freeing up expert time for strategic activities.
The new system consistently matches or outperforms human forecasting accuracy while eliminating the variability that comes with manual processes. This gives leadership confidence in their project planning and pricing decisions.
It also gives our partner’s sales teams access to reliable cost baselines within minutes of receiving project specifications, enabling them to respond to opportunities faster and with greater confidence.
The system processes thousands of data points across multiple projects simultaneously, and provides insights that would be impossible to generate manually. Ultimately, our collaboration enables our partner to scale their operations even further while maintaining their foresting accuracy.
