"We had the data, but it was locked inside a loan management system that was never designed for analytics. We needed a way to turn raw transactions into portfolio intelligence without asking the team to build spreadsheets by hand."
Balloon Ventures is an SME lender operating across East Africa, providing loans of $20,000 to $500,000 to small and medium businesses in sectors like agriculture, education, healthcare, and construction. With over 50 staff working from more than 10 branches across East Africa, the organisation has disbursed millions in loans and supported thousands of entrepreneurs. Every loan includes six months of business support, and the lending model is built around digitising cashflow data to make smarter credit decisions in markets where traditional banks will not lend.
The challenge
As the portfolio grew, so did the gap between the data Balloon Ventures was collecting and the insights the team could actually extract from it. Loan records, borrower profiles, repayment transactions, and schedules all lived inside their loan management platform, designed for origination and servicing rather than analytics. Getting a clear picture of portfolio health meant manually exporting data, reconciling it across spreadsheets, and computing risk metrics by hand.
The team needed to answer questions that the loan management system was not built to answer: which loans are falling behind schedule, what is the true arrears position across the portfolio, how does repayment behaviour vary by season or sector, and where should the next round of follow-up support be directed. These questions required computed measures that did not exist in the source system, including FIFO-based instalment allocation, forward-looking risk classifications, and rolling financial ratios.
With operations spanning two countries, each with its own branch structure, currency considerations, and materiality thresholds, the challenge was not just analytical but also operational. Any solution needed to run reliably, update frequently, and deliver results in a format the portfolio team could use without technical assistance.
What we built
We designed and delivered an end-to-end data pipeline that extracts raw data from the loan management system, transforms it through a multi-stage analytics layer, and publishes curated portfolio views to an internal portfolio workspace, where the lending and operations teams consume them daily.
Data ingestion
The pipeline connects to the loan management system’s API for each country branch, pulling borrower records, loan details, original schedules, and repayment transactions. Ingestion runs on an hourly cycle during business hours, keeping the warehouse current without overloading the source system. Records are normalised and upserted into a MySQL data warehouse, maintaining referential integrity across entities.
Analytics layer
The core of the system is a chain of over 24 stored procedures that compute derived measures from raw transactional data. This runs in four stages. First, a calendar framework maps each loan’s lifecycle against time. Second, repayment allocation logic distributes actual payments across scheduled instalments using FIFO ordering, producing an audit trail of how each payment was applied. Third, arrears and risk measures are calculated, including days past due, outstanding balances, and risk classifications with materiality thresholds calibrated for each market. Fourth, aggregation procedures compute portfolio-level summaries: repayment patterns, financial ratios, and seasonality signals, which help the team anticipate cyclical patterns in borrower performance.
Portfolio delivery
The final stage publishes computed measures to country-specific portfolio workspaces. Loan measures update hourly, while repayment summaries and instalment status snapshots update daily. The sync process compares database records against existing workspace content, upserts changes, and removes stale entries, ensuring the portfolio team always sees a clean, current view without duplicates or orphaned records.
Operational reliability
The entire pipeline is containerised and hosted on Google Cloud Platform. Cloud Run handles the API service, Cloud Scheduler triggers each stage of the pipeline on the appropriate cadence, and Cloud SQL provides the managed database layer. Deployments run through GitHub Actions with separate staging and production environments, and the system includes built-in rate limiting, retry logic with exponential backoff, and detailed logging for troubleshooting.
The result
The portfolio team now has a single, continuously updated view of loan performance across both markets. Measures that previously required hours of manual spreadsheet work are computed automatically and available in the portfolio workspace within minutes of new data arriving from the source system. Risk signals surface early, giving the team time to intervene with support before a loan reaches a critical state.
The analytical depth of the system, from individual instalment allocation through to portfolio-wide seasonality analysis, gives Balloon Ventures a level of insight into borrower behaviour that’s unusual in SME lending. That insight feeds directly into better lending decisions and more targeted business support, which is ultimately what allows the organisation to keep lending where others will not.