Skip to content

Machine Learning Enables BigChange Customers To Analyze Field Service Performance

Executive Summary

BigChange wanted to apply machine learning capabilities to analyze free-text data created by customers in the BigChange field service management platform that runs on AWS. This would give customers visibility into industry performance statistics on various job types. To solve the challenge, BigChange turned to Mission, which designed a machine learning algorithm that leverages natural language processing. The algorithm moves data into Amazon SageMaker and pulls a clustered model to upload analysis outputs to Amazon EC2 servers. Mission also implemented autoscaling capabilities so the solution can handle spikes when BigChange does customer analysis activity. Because the solution classifies job types numerically and consolidates similar phrases into a single semantic meaning, BigChange can generate custom analytics on the fly for any job type. The accurate search results give BigChange deep insights into how their customers' job performance statistics compare to other companies in their industry.

Key Highlights

  • Determined appropriate algorithms, job type data, and natural language processing model parameters. 
  • Consolidated wide range of job types to reduce variability of customer input data. 
  • Training model includes historical data access, feature engineering, data transformations, and artifact storage. 
  • Test and evaluation model incorporates historical data access and data model updates. 
  • Implementation processes comprise access to suitable historical data and deployment of the classification model. 
“Our immediate need for this project was machine learning, but we were also impressed with Mission’s capabilities across all AWS service offerings. They understand cloud best-practices, which can help us going forward for both the performance and the cost of our AWS environment as well as the security of our infrastructure.” 
Johann Levy 
Chief Innovation officer

The Challenge

Companies use the BigChange field services management platform to input job types along with associated job attributes to analyze, for example, how long jobs take to complete, project pricing, and how far contractors travel. Each customer develops their own job types so the language of the data that’s keyed in varies greatly across the population of BigChange customers. 

A new capability BigChange wanted to provide to the platform was visibility into industry-wide data. This would allow customers to compare how well their field service teams perform in completing various jobs with other companies in their respective industries. But due to the free-text data that customers use, there were many different names for similar job functions. For example, one boiler service company might refer to a job as <maintenance> while another might refer to a similar job as <repair>. 

BigChange customers also operate all over the world so there’s a mix of languages, including English, French, and Spanish. In some cases, the power range for an appliance to be fixed is injected into job types. The use of different job attributes, naming conventions, and languages makes the analysis less accurate when customers compare their performance against each other. 

“The varying job attribute jargon further complicates customer analysis,” adds Johann Levy, Chief Innovation officer at BigChange. “We needed an accurate classification of job categories to provide additional value to our customers as they analyze their performance.” 

To solve this challenge, BigChange needed to implement machine learning capabilities to group similar job attribute data and classify free-text job categories. This would enable the field service platform to report on and provide analysis comparisons across all the job attribute categories. As a starting point, BigChange decided to first focus on developing a solution for its plumbing and heating customers.

Why AWS

BigChange utilizes the cloud for hosting its IT infrastructure and relies on a combination of Amazon Web Services (AWS) and Microsoft Azure. For the customer field services platform, BigChange hosts the entire solution on AWS for its ability to scale compute resources quickly to handle spikes in customer activity. 

Another key feature BigChange values is how AWS provides a wider range of mature services so new technologies, such as machine learning, can be added more easily. In addition, AWS offered an incentive for BigChange to adopt machine learning capabilities through funding from one of its various funding programs. 

Why Mission

BigChange first turned to AWS for help in solving the challenge of implementing machine learning capabilities into the field services platform. AWS referred BigChange to Mission, an experienced AWS partner with machine learning expertise. BigChange and Mission also share the same private equity firm, Great Hill Partners. Besides coding machine learning capabilities for the BigChange platform, Mission discussed AWS cost optimization opportunities. 

Strategy and Solution

The Mission discovery process included reviewing the BigChange platform’s existing data as well as features, and attributes. Mission also examined data cleansing and transformation operations and the impact of data quality issues. Mission then reviewed heuristics, patterns, manual and ad hoc techniques, and the current machine learning models. 

From there, Mission gathered the project requirements, including the desired segmentation performance, evaluation criteria, and performance goals. The planning and design process included determining the appropriate algorithms and data, selecting the appropriate natural language processing model, and identifying suitable historical training data. 

Focusing on improving field service job analytics for BigChange customers in the plumbing and heating sectors, Mission devised a small number of job types to consolidate the variability across all customer input data. Mission identified the data that was important to capture and the meaningful attributes of each job type. This enabled the machine learning algorithm to identify a group of jobs that are basically the same. 

The architecture for model training includes historical data access, feature engineering, data transformations, and trained model artifact storage. For test and evaluation, the design incorporates historical data access and data model updates. At the implementation stage, processes comprise access to suitable historical data and deployment of the classification model. Mission also evaluated model performance and tuned each parameter of the model. 

“As Mission packaged the model, they did a great job at cleaning up the code to ensure reproducibility for our customers. Mission also ran a smooth knowledge-transfer session with us that included working sessions along with a recorded video and written documentation.” 
Johann Levy 
Chief Innovation officer

Results and Benefits

  • Represents job types numerically to enable data model algorithm to pull appropriate data.
  • Enables BigChange to generate custom analytics on the fly for any job type by consolidating similar phrases into a single semantic meaning.
  • Pulls clustering model from Amazon SageMaker to enable autoscaling of compute resources.
  • Provides BigChange with accurate search results they can share with Customers, giving them deeper competitive insights into their industry. 

Mission completed the machine learning model in one month by reducing the amount of data that needed to be analyzed, and core to the project’s success was the way Mission represented job types numerically in the machine learning model. This allows the data model algorithm to use natural language processing to automatically cluster and categorize job types, which simplifies search results.

BigChange can generate custom analytics on the fly for any job type, and gain insight every time a customer inputs a job type, the Mission solution pulls clustered data from Amazon SageMaker and passes the job types through the algorithm model. The solution then passes the analysis output to Amazon EC2 servers. 

“Using SageMaker in the process is key because it’s more scalable and easier to manage than a population of EC2 servers,” says Johann Levy. “This makes it possible for our analysis tool to tap into additional compute resources to handle spikes in customer activity.” 

The machine learning model consolidates similar phrases into a single semantic meaning, providing customers with the ability to generate custom analytics for any job type. As a result, BigChange has additional knowledge around its plumbing and heating customers which allows them to give their customers deeper competitive insights into their industry. 

With the new analytical capabilities of the Big Change field services management platform, customers can't see the specifics of what other companies are doing. But they will be able to see, for example, if it's taking five hours to complete a particular service task for which competitors require only three hours. BigChange can then investigate and create reports for their customers allowing them to determine how to streamline their workforce operating procedures. 

 

“The machine learning capabilities implemented by Mission are ideal for an environment like ours where we don’t have a discrete list of all the different possibilities for job type data. For every customer, there are slight variations, and the Mission model consolidates all of those into a single semantic meaning—transforming an unscalable software problem into a practical statistical problem.” 
Johann Levy 
Chief Innovation officer

Next Steps

The machine learning model developed by Mission gives BigChange insights into their plumbing and heating customers. BigChange also has a framework for which the analysis model can be expanded to customers in other industries. 

“With improved data categorization and machine learning capabilities, we now have the option to take advantage of a modernized data warehouse,” says Johann Levy. “We can offer new types of services to customers, including high-level reporting that gives performance advice and predicts job outcomes.” 

AWS Services
  • Machine Learning Services on AWS
  • Amazon SageMaker
  • Amazon Elastic Compute Cloud (EC2)
  • Amazon Simple Storage Service (S3)
Third-Party Integrations
  • Python
  • Streamlit
  • SBERT