How Pondera Services Got Its Call Center Data Out of Spreadsheets and Into a Data Warehouse

Key Metrics

The Operation Was Solid. The Data Infrastructure Was Not.

Pondera Services LLC runs inbound call center operations for nonprofits. With a decade in business and an agent team based in the Philippines, the organization handles significant call volume daily. Independent contractors choose their own schedules, get compensated by the minute for talk time, and are monitored by quality analysts who score calls through structured evaluation forms.

The operational model worked. The data model behind it did not. Three disconnected systems and a manually reconciled compensation process led Pondera to hire Data-Sleek, applying the same data warehouse best practices that anchor every engagement, to build a foundation the spreadsheets never could.

The Challenge: Three Platforms, No Shared Foundation, and Compensation Built on Spreadsheets

Pondera’s data came from three places that never spoke to each other.

Five9 served as the call center platform, capturing agent login and logout activity and detailed call records for every interaction. Separately, quality assurance analysts scored calls using Google Forms, evaluating agents against a structured set of criteria. And a third platform, Bubble, managed contractor scheduling: agents selected their own shifts, and that data needed to be matched against actual call activity to calculate accurate compensation.

Each minute of agent talk time had a dollar value attached to it. That meant the compensation report, run daily, had to reconcile three independent data sources with precision. It was being done manually in spreadsheets.

The errors compounded. Reconciling across platforms by hand introduced inconsistency. There was no single source of truth for agent performance, no reliable audit trail, and no foundation for analysis. Leadership could not answer basic operational questions without digging through spreadsheets that different people maintained differently.

The Solution: A Five9-to-Snowflake Pipeline with Daily Automated Loads

Data-Sleek designed and built a data pipeline that automated extraction from Five9, structured it for analysis, and landed it in a Snowflake data warehouse on a daily schedule.

Two AWS Lambda functions handle Five9 extraction. One pulls the MRD Call Log, which contains per-call detail including agent ID, campaign, talk time, disposition, and time-of-day breakdowns. The second pulls Login/Logout activity, capturing agent availability, on-call time, ready time, not-ready time, and after-call work time. Both functions use Five9’s SOAP API, convert the report output to CSV, and write to dedicated S3 paths. AWS EventBridge triggers both on a 24-hour schedule. Credentials are stored in AWS Secrets Manager, not in code.

Bubble scheduling data arrives in S3 separately. A third Lambda function, triggered by S3 PUT events, filters the Bubble report file from a daily drop folder and routes it to its own S3 path for ingestion.

Fivetran S3 connectors sync all three data streams into Snowflake: MRD call logs, login activity, and Bubble scheduling records each land in their own staging tables. The dbt transformation layer moves data from the raw staging database into the PONDERA_DW analytical environment.

The result is a single warehouse where call data, agent time data, and scheduling data can be joined and queried together for the first time.

The Results: What the Infrastructure Made Possible

Before this project, answering a basic question about agent performance required pulling from three systems manually and reconciling the results in a spreadsheet. The margin for error was built into the process itself.

That reconciliation now happens in the warehouse. Agent talk time from Five9, shift selection from Bubble, and login activity land in Snowflake on a daily automated cycle, and the compensation report runs off that joined data with an auditable load history behind it, no spreadsheet required.

The bigger shift is what became possible beyond the daily report. Agent utilization, schedule adherence, and call disposition patterns can now be examined together instead of in isolation, turning what used to take hours of manual spreadsheet work into a query.