Parallax was beginning to explore the use of artificial intelligence opens a new window (AI) or machine learning (ML) to leverage the wealth of data on hand about customer projects, with the goal of improving their resource planning. The company thought it might be possible to create a machine learning model that identifies customer projects at risk, equipping the Customer Success team to make data-driven recommendations on how to head off problems before they occur.

Background

Founded in 2019, Parallax opens a new window helps digital service organizations optimize operations using sophisticated tools that improve capacity planning and resource planning and management. The Minnesota-based company equips small and mid-sized organizations to align people with work, enabling them to make hiring and staff utilization decisions that drive higher growth and profitability.

Context

Parallax was beginning to explore the use of artificial intelligence (AI) or machine learning opens a new window (ML) to leverage the wealth of data on hand about customer projects, with the goal of improving their resource planning. The company thought it might be possible to create a machine learning model that identifies customer projects at risk, equipping the Customer Success team to make data-driven recommendations on how to head off problems before they occur.

But as a SaaS startup, Parallax needs to focus internal capacity on developing solutions that fulfill contractual obligations and enhancing existing services. The company doesn’t have the luxury of tying up staff to explore a proof of concept.

“By working with OmbuLabs, we could quickly experiment and iterate on a new possibility outside our core product, without cannibalizing our team’s time or impacting our product development velocity,” — Jacob Ward - Head of Product at Parallax.

Exploration

The exploratory engagement began with a focus on two questions about the Parallax customer base:

• Is our customer doing a good (or poor) job at resource planning?
• What are the most common problems that derail our customers’ resource plans?

Our goal was to use ML-generated insights derived from these questions to develop a solution that delivered value to customers using its planning tools.

We conducted extensive data analysis and data validation centered around Parallax customers’ typical planning scenarios. Through collaboration and discussions with our client, it became apparent that the original questions weren’t the most relevant or pressing to address. So our focus shifted to helping Parallax identify the actual problem that needed to be solved, and whether machine learning was the right solution.

Together, we determined it would be more valuable to customers if Parallax used ML as a predictive tool to guide resource planning dynamically. The new questions became:

• Can we build a model that predicts the number of hours each role will log on a given project each day?
• How would those predictions improve our customers’ resource planning?

Our Approach for a Successful Engagement

We analyzed a large volume of historical project data across the Parallax customer base to determine the best approach to addressing these questions.

“They worked collaboratively with our engineers to get the data into a format they could model and execute against.” “It felt like they were an extension of our team, helping us explore the concept.” — Ward explained.

The team kept iterating to understand what they could do with the data and how to move forward with the idea in a way that solved a relevant customer challenge.

Through a process of exploring, prototyping, and validating, we built a custom regression model from the ground up. Our team trained the model on a large number of completed projects across the Parallax ecosystem and fine-tuned it to take multiple factors into consideration. Then we developed an application in Python that allows Parallax’s C# platform to interact with the machine learning within its existing infrastructure. The application also applies statistical modeling to the model’s predictions to adjust calculations and improve the confidence intervals.

The Outcome: A Predictive Model

At the conclusion of the engagement, Parallax gained a custom predictive model and a working API hosted in Azure. The solution extracts relevant historical data across all Parallax customer projects, roles, and employees, compares it to the conditions of a current customer project, and predicts what will happen next. The model returns a forecast with confidence intervals bound by upper and lower thresholds.

The model can predict how many hours a particular role will log on a project daily based on planned hours and historical trends, taking into account any deviations specific to the customer’s organization. As the customer adds new roles or employees to the project, the model responds dynamically, accounting for their workload and availability across all their assignments.

For example, the customer’s resource plan might assume that Person A will log 25 hours on the project this week and Person B will log 18 hours. But the model might predict very different activity levels, which could significantly impact staff utilization and profitability. Equipped with this information, a digital service company can take proactive steps early enough to course-correct and prevent costly problems.

Next Steps

Parallax was extremely pleased with both the process and the end result.

“This collaboration represents a successful pivot from exploratory research into a deployable, value-driving tool.” “It laid the foundation for deeper strategic applications within our platform, which could help teams plan more effectively and make informed, autonomous decisions through agents and predictive intelligence,” — Ward noted.

Ward described the process as a valuable learning experience.

“We invested in working with a partner that knew how to guide us down a path of understanding what we could do with our data and how machine learning and AI could make an impact for our customers.”

Parallax especially valued our domain expertise and ability to evaluate how to use data to meet a customer need.

“They consulted us on what our data could do and guided us to a solution that was viable for end users” “It proved our hypothesis: that what would be valuable for customers is actually possible to deliver.” — Ward said.

The resulting model could significantly impact a digital service organization’s top and bottom line.

“Our customers will have revenue leakage and make a lot less profit if they don’t address problems early,” he explained. “This model can provide visibility into where they’re going to end up, so they can make better staffing and resourcing decisions based on better predictability.”

Parallax hasn’t deployed the model within its tools yet, as it’s still a working prototype that would need more training on more data. Eventually it could become a commercialized product, whether built into one of the company’s existing tools or offered as a chargeable add-on. Regardless, the company has a blueprint for the future.

“We know we can do this,” Ward said. “We know it works. And OmbuLabs provided the documentation and knowledge transfer for us to own this. It’s one of multiple paths for our future intelligence strategy.”

Project type:

• Technology consulting engagement
• Machine Learning model development

Built using:

• Language: Python;
• Frameworks: FastAPI, scikit-learn;
• Model Deployment: MLflow;
• Hosting: Azure;

Want to build something amazing with OmbuLabs? Check out our one-week Design Sprint service opens a new window ! We can take you from idea to prototype in 5 days. 🚀