Kuda achieves predictable pricing and faster data refresh with Airbyte

Company Size

700-800

Region

Africa

Industry

Digital Banking (Neobank)

Sources

Microsoft SQL Server, File-Based, Various External Feeds

Destination

Google BigQuery

Tech Stack

• Source(s): Microsoft SQL Server (Azure), File-Based, Google Ads, Intercom, Iterable, LaunchDarkly, Snapchat
• Destination(s): Google BigQuery
• ELT: Airbyte (migrated from Fivetran)
• Transformation: Dataform

Key Results

• Achieved predictable, transparent pricing, eliminating unexpected cost overruns from credit-based billing
• Reduced data refresh frequency from 15-minute minimum to faster intervals, enabling near real-time analytics
• Completed migration in under 3 weeks, including full data re-sync of billions of banking records
• Successfully managing CDC data replication from SQL Server to BigQuery at scale for a rapidly growing neobank

About Kuda

Kuda is a fintech company making financial services accessible, affordable, and rewarding for Africans.

Founded in Nigeria in 2019 by Babs Ogundeyi and Musty Mustapha, the company serves over 7 million customers in Nigeria through its mobile and web apps, offering services including money transfers, savings, business banking and instant access to credit. Kuda has raised over $100 million from investors, including Valar Ventures and Target Global, and has a team of over 700 people across Africa, the UK, Canada, and Europe.

Seeking more predictable pricing and faster refresh rates

Mondor La Grange, Head of BI and Data Engineering at Kuda, oversees all data feeds and movement within the company. With responsibility for ensuring data flows reliably from Kuda's SQL Server-based line of business systems to BigQuery for analytics, Mondor's team manages everything from transactional banking data to marketing feeds and call center information.

When Mondor joined Kuda nearly four years ago, the company was using Fivetran for data replication. While the platform functioned well, Kuda's growing data volumes created challenges for the credit-based pricing model.

"The challenge with Fivetran was not the absolute cost; it was the lack of predictability. The credit-based model lacked sufficient transparency to support effective planning. There was no clear or consistent mapping between credits and tangible usage metrics such as data volume or row counts, which made it difficult to forecast consumption as the platform scaled. As a result, credits were frequently exhausted mid-cycle, often without warning, forcing reactive purchases at premium rates. This usage-driven pricing structure limited our ability to plan costs proactively and introduced ongoing financial uncertainty as data volumes grew," Mondor explains.

For a fast-growing neobank handling billions of records, this made financial planning difficult. Additionally, Fivetran's 15-minute minimum refresh interval didn't align with Kuda's objectives. "In our game, it's actually about having more real-time information," Mondor notes.

As Kuda's data needs evolved, Mondor began exploring alternatives that could offer more predictable costs and faster refresh capabilities.

Thorough evaluation leads to Airbyte selection

Mondor's team evaluated several alternatives, including Meltano and Dagster, and considered building their own solution using Apache Airflow. The key requirement was straightforward: reliable Microsoft SQL Server CDC support. Many open-source options supported various databases, but SQL Server connectivity remained a gap.

At the time of evaluation, Airbyte's SQL Server connector was in beta, but the Airbyte team was able to push a mature production-grade connector live for Kuda's onboarding. More importantly, Airbyte offered exactly what Fivetran lacked: predictable, transparent pricing.

"We got in touch with Airbyte at the time, and it offered us predictability based on the conversations that we had," Mondor says.

The evaluation process was in-depth. Kuda started with Airbyte's cloud trial, then engaged directly with the Airbyte team. To validate performance, they ran both Fivetran and Airbyte in parallel, comparing functionality, speed, and reliability. The systems performed similarly, but Airbyte impressed with low-latency refreshes under fifteen minutes.

Seamless migration completed in under three weeks

Once Kuda committed to Airbyte, the technical migration proceeded smoothly. The actual transition took less than three weeks, with most time spent on initial data syncs rather than technical configuration.

"From a technical perspective, the cutover itself was relatively straightforward, as it primarily involved re-pointing existing data sources and consolidating them within the new environment," Mondor explains. "The majority of the effort was not in the cutover mechanics, but in executing the necessary data synchronization and backfill processes required to ensure historical consistency and data integrity."

The scale of the data estate made the migration timeline particularly noteworthy. As a bank, Kuda operates on datasets comprising billions of transactional records, resulting in substantial data volumes that require careful handling. During the initial data take-on phase, certain high-volume tables required between 48 and 72 hours to complete a full synchronization, driven by the need to ensure completeness, ordering, and consistency across source and target systems. As Kuda's customer base continues to grow, these synchronization windows naturally expand, underscoring the importance of an efficient, resilient migration approach that maintains data integrity at scale.

Predictable costs and faster refresh rates deliver operational confidence

For Kuda, the primary benefit of Airbyte has been cost predictability. Unlike the credit-based system that created budget uncertainty, Airbyte's pricing model allows Kuda to forecast expenses accurately and avoid surprise bills.

Beyond cost control, Airbyte enabled faster data refresh cycles. Where Fivetran imposed a 15-minute minimum interval, Airbyte supports more frequent syncs. This capability aligns with Kuda's need for near-real-time analytics in a fast-paced data-driven environment.

The primary use case remains CDC replication from SQL Server to BigQuery, with the core value centered on the reliable, predictable movement of banking transaction data at scale, supporting analytics, reconciliation, and regulatory reporting.

Simple, straightforward data movement at scale

When asked about Airbyte's strengths, Mondor emphasizes simplicity and reliability.

"I think it's an easy, simple, straightforward system. That's what makes it a great tool," he notes.

As Kuda continues to grow and handle increasing data volumes, the stability and predictability of Airbyte's platform provide the foundation the data team needs to focus on delivering insights rather than managing infrastructure surprises.