Professional Services Company Leverages AI NLP to Streamline Construction Proposal Process

This North American professional services company offers data, software, and expertise to successfully guide contractors and facility owners through every phase of the building lifecycle. From construction planning and building to facility operations, this organization empowers its customers to overcome business challenges by delivering critical data, innovative technology, and extraordinary services


This professional services organization supplies software to contractors and site owners to expedite and standardize the process of creating contracts for construction work. Contractors and site owners work together on building a written scope of work for the project, and contractors use this scope to build a proposal of all the tasks that will be required to complete construction.

The process for building out these proposals is manual and timeconsuming; contractors must select each construction task associated with a project from a giant project catalog. On average, these catalogs list 150,000 tasks – a daunting amount to sift through. Using this system, contractors were typically taking about 27 days to build an initial proposal.

Once the proposal is built, the contractor and owner exchange edits until it is finalized. This iterative process tacks on an additional 31 days to the process on average. This organization understood that contractors and owners are bogged down by this manual back-and-forth and that proposals are less likely to gain approval the longer they sit in this pipeline. They needed a faster, smarter solution to cut the overall process from about 60 days to 20 – 30 days. This would not only improve the customer experience but would also mean more projects could be approved in less time, bringing in more revenue.

This organization reached out to Microsoft for a solution, and they recommended the data science experts at CCG.


After a joint two-day Architecture Design Session (ADS) , the organization enlisted CCG to deliver a Minimum Viable Product (MVP) solution to accelerate the proposal development process. CCG quickly built an internal proof of concept to demonstrate the integration of an AI agent with their software.

The AI program leverages state-of-the-art natural language processing capabilities to read through a project’s work scope and compare the language in that work scope to all the other work scopes that have ever been submitted, looking for similarities.

Once similar work scopes are identified, the tool automatically generates a recommended list of associated tasks for the new proposal. While contractors might have to add or delete some items from this newly generated list, the ability to intelligently and automatically narrow down the task selections is an enormous timesaver.

CCG utilized Google’s BERT model for language understanding for this application. BERT is an innovative open-source neural network first introduced in October 2018 that was trained to understand English by reading every article on Wikipedia multiple times and attempting to perform a few simple tasks like word replacement. CCG fine-tuned the default BERT model to learn construction-specific terminology by having it read through every work order a few times and perform the same word-replacement tasks.

The customized BERT model was then built into a broader machine learning pipeline and exposed via a REST API to integrate the solution with the organization's existing application.

Microsoft Azure was chosen to host the solution to take advantage of Azure DevOps and Azure’s ability to seamlessly and dynamically scale GPU hardware as needed on Data Science Virtual Machines, which reduces hardware costs and increases performance.

Results & Implications

This professional services company and CCG have started on the path to move this task recommendation engine to production, after obtaining overwhelmingly positive customer feedback. The MVP approach to creating this solution has helped drive continued support for the project and other innovative initiatives in the organization.

The use of transfer learning is still new in its application to natural language problems, and there are extraordinary implications and efficiencies to be gained from adopting a model trained on massive amounts of data to a custom task, as in this case. Other industries with their own vocabulary and shorthand can benefit from fine-tuning general AI models just as easily.

CCG’s appetite for learning and growing sets us apart in our ability to quickly master the latest, most innovative data science technologies and put them to work for our clients. If you’re interested in how we can solve your most mission-critical challenges, contact us.

Quick Facts

  • Industry - Professional Services/ Manufacturing
  • Solution - Data Science / Artificial Intelligence
  • Technology - AI, BERT, Flask, Microsoft Azure


Previous Windows server got a major update from it's last update in 2012.

Join the hundreds of CCG customers.
Talk to Intelligent Enterprise Guide.