HubSpot powers GenAI, saves $100,000 with Fivetran

The world’s leading sales and marketing software maker improves the accuracy of its recruiting forecasting from 70% to more than 90% percent while saving hundreds of engineering hours.
“In our first year with Fivetran, we built 40+ different pipelines for nearly 700 tables, which handle nearly 40 million records every month. It took fewer than 40 hours to build it all, saving us nearly 1,000 hours and $100,000 across multiple departments, leading to a 150% ROI.” 
— Sandro Frattura, Analytics Engineering Manager, HubSpot People Operations

Key results

  • Improved workforce productivity and planning by utilizing HR data with generative AI
  • Improved predictive metrics accuracy by 90 percent to within 3-5 percent of actual results
  • Enabled accurate forecasting and delivered meaningful insights by providing access to hiring and recruitment data 
  • Reduced pipeline development from 6-10 weeks to under an hour
  • Saved $100,000 in data engineering labor in the first year

Data stack

  • Pipeline: Fivetran
  • Destination: Snowflake
  • Transformations: dbt
  • Data sources: 13 sources, including Greenhouse, Workday, Survey Monkey, Google Sheets
  • Business intelligence: Looker

HubSpot is a globally recognized enterprise with more than 205,000 companies using their all-in-one CRM platform to grow their businesses. The organization has grown to employ more than 7,600 HubSpotters since it was founded in 2006. Their mature and proactive use of data over the years has allowed them to successfully ride out the unpredictable ups and downs of market conditions.

Coming out of the pandemic in 2021, HubSpot’s leadership team wanted to take a more strategic approach to hiring, resulting in a deeper desire for visibility over their processes. In early 2022, they formed the Workforce Planning Department with the mission to lead and execute strategic headcount planning and people data reporting. It was this mission that would help them manage through post-pandemic market downturns and come out stronger than ever.

Data discovery and modeling, one source at a time

Understanding and managing team growth came down to one thing: data. HubSpot was already collecting all the data it needed for important recruiting insights. However, the data wasn’t easily accessible in a centralized location where it could be easily reported and fed into dashboards.

Sandro Frattura, Analytics Engineering Manager, and Eliza Geeslin, Data Analytics Manager, were pivotal leaders for the Workforce Planning data strategy and implementation. The first step was locating and understanding the most critical recruitment data. Geeslin recalls, “Everything we needed was scattered in various locations, including financial planning and analysis (FP&A) spreadsheets. If the CEO asked where we’d be in January for headcount, we couldn’t quickly answer.”

Frattura tried everything he could to ingest and centralize the data needed, including: 

  • Building workflow automation for spreadsheets
  • Custom data pipelines from Greenhouse and Workday
  • Partnering with the internal BI team
  • Manually importing CSV files from its Human Resource Information System (HRIS) 

What he ultimately designed was a monthly ingestion process that allowed Finance to get the answers they needed in spreadsheets. It was enough to build data models and early dashboards but was fragile and unsustainable. The data models and daily spreadsheet syncs couldn’t seamlessly handle column changes or data adjustments. The business needed a more robust and automated solution.

Replacing DIY pipelines 

Frattura turned to Fivetran for its reliable and scalable data integration platform. It didn’t take long for him to build new pipelines that ingested the FP&A spreadsheet data into Snowflake and immediately handled data shifts. “Every time a spreadsheet ingestion broke with our old method, it would take three to five hours to fix,” Frattura notes. “When I used Fivetran, it took less than five minutes to set up and had no issues with data changes.

Motivated by his early success, he moved all spreadsheet ingestions to Fivetran and then focused on two major data sources: Workday and Greenhouse. To his surprise, he could ingest much more data from Greenhouse and Workday via Fivetran than the existing DIY pipelines. The Fivetran pipelines gave him expanded access to valuable vacation and leave data, which he could leverage for predictive analytics around capacity and productivity.

Armed with Fivetran, Frattura’s team could now create their own data ingestions instead of “taking a number” with the centralized corporate data ingestion team, who are often focused on building ingestions for other departments, such Product, Engineering, Marketing and Sales. Within two weeks, he had built a complete proof of concept with Fivetran, which included replacing all of the spreadsheet workflow automation and custom data pipelines to Workday, Greenhouse and Survey Monkey. This gave him real-world insights into the long-term cost savings of this new data architecture.

“Two things are great about Fivetran’s flexibility and cost savings. The initial load is free and every connector is free to use for 14 days. We can experiment with data, make business decisions and empirically forecast costs, allowing us to focus on high-value analysis tasks at very low risk.”
 — Sandro Frattura, Analytics Engineering Manager, HubSpot People Operations

Building an accurate forecast for hiring demand

Before Fivetran, Geeslin found it very difficult to predict recruiter productivity. “We didn’t have access to who was on vacation,” Geeslin notes. “We relied on recruiting managers and they’d tell us after the fact that someone was out.”

With Fivetran providing access to leave data from Workday, Geeslin can proactively forecast how many hires will be made in the upcoming 18 months. Before the data provided by Fivetran, forecasts were about 70% accurate. Now, forecasts are 90-95% accurate to what happens. This directly improves HubSpot’s bottom line and business strategy, as more accurate forecasts allow for more accurate budgeting around hiring and recruitment. The benefits extend beyond the hiring of applicants; the data could even predict a new employee’s chances of success in the HubSpot culture using behavioral characteristics measured in the recruitment processes. 

“Thanks to Fivetran, our focus has shifted entirely to where it truly matters: diving deep into the data and uncovering valuable insights that help us better forecast labor supply and demand, analyze skill gaps, predict future needs, and ultimately help us understand how strategic HR affects different areas of our business.”
 — Sandro Frattura, Analytics Engineering Manager, HubSpot People Operations

In addition, eNPS has been a major area of investment at HubSpot to improve overall performance and employee satisfaction. The People Analytics team built new data models to improve HubSpot’s management culture with eNPS data, which is collected quarterly and stored in Survey Monkey. Frattura notes the increasing hunger for insights. “As soon as the new eNPS analysis packets are ready, dozens of managers request it within minutes,” Frattura says. “Within two days, 90 percent of managers have pulled their personalized eNPS insights.”

Generative AI for strategic workforce insights

As a leader in AI-powered software, HubSpot sees a bright future for delivering business-impacting insights in intelligent dashboards and text-powered chatbots. They’re able to use all of their pre- and post-hire employee performance data — including text-based information that was unusable just a year ago. And they’re only getting started.

“Fivetran has revolutionized our approach to data, enabling AI/ML and GenAI initiatives on employee performance and providing managers with faster insights. These advancements were unimaginable just two years ago.”
— Sandro Frattura, Analytics Engineering Manager, HubSpot People Operations

HubSpot has a lot of exciting projects planned for 2024, including: 

  • Candidate pool: Is the top of the candidate funnel diverse enough?  
  • Performance reviews: Are some departments or managers “grading” too leniently? Too strictly? Are performance reviews equitable? Is there unintentional coded/exclusionary language being used? 
  • Remote/hybrid connectedness: What can we learn from other data sets (Slack, Zoom, ticketing systems, etc) that will help us make HubSpot employees feel more connected?
  • Interviewing: Are we interviewing fairly? How good are we at predicting employee success when making offers?    

All of these use cases are made possible by the data they can ingest because of Fivetran.

Download this IDC report to learn how Fivetran drives millions in financial impact and enables new business initiatives for enterprises.

Start for free

Join the thousands of companies using Fivetran to centralize and transform their data.

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.