This is Airbyte’s page, Kuda, published 1 February 2026. Read it as a vendor case study. Mondor La Grange, Head of BI and Data Engineering at Kuda, is the person they quote. The outcome lines are the vendor’s.
Kuda is a Nigerian neobank, founded in 2019 by Babs Ogundeyi and Musty Mustapha. The page says it serves over 7 million customers, has raised over $100 million, and has a team of over 700 people across Africa, the UK, Canada, and Europe. A facts table on the same page says company size 700 to 800. The 7 million is a retail customer count. It is not a row count. “Billions of records” is how they describe the banking data.
Mondor had been at Kuda for nearly four years when this was written. He owns the feeds from SQL Server line-of-business systems into BigQuery, plus marketing feeds and call center data. Fivetran was already there when he joined. The Fivetran → Airbyte playbook is connectors, cursors, and a backfill you can reset. This page is why a bank got tired of credits, and how long the historical sync took.
The bill they could not forecast
He says the problem was not the absolute cost. Fivetran credits had no clear map onto volume or row counts, so the team could not forecast the next cycle. Credits ran out mid-cycle, often without warning, and the top-up was at premium rates. A growing neobank made that worse, not better.
Refresh was the second miss. Fivetran’s minimum interval on their setup was 15 minutes. He wanted something closer to real time for a bank. The page does not say which dashboards were late, or what a 15-minute lag cost them in operations.
What else they looked at
They evaluated Meltano and Dagster, and they talked about building the pipelines on Apache Airflow. The requirement he names is narrow. Microsoft SQL Server CDC had to be real, not a connector that existed for Postgres and stopped at SQL Server. A lot of open-source options failed that test.
Airbyte’s SQL Server connector was in beta during the evaluation. The page says the Airbyte team shipped a production connector for Kuda’s onboarding. Pricing was the other half. After conversations with Airbyte, Mondor says they were offered predictability. There is still no price on the page, and no Fivetran invoice to compare it with.
They started on Airbyte’s cloud trial, then worked with the Airbyte team directly. Fivetran and Airbyte ran in parallel so they could compare behavior, speed, and reliability. He says the systems performed similarly, and Airbyte’s refreshes landed under fifteen minutes.
Under three weeks, most of it backfill
Once they committed, he calls the cutover straightforward. Existing sources were re-pointed and consolidated in the new environment. The time went into sync and backfill so history matched.
The page says the transition took less than three weeks. Some high-volume tables needed 48 to 72 hours for a full synchronization, because they cared about completeness, ordering, and consistency. He also says those windows grow as the customer base grows. That is a warning about the next backfill, not a measured curve.
Transformation on the facts table is Dataform. Sources listed are Microsoft SQL Server, file-based inputs, and other external feeds. Destination is BigQuery. The use case they center is CDC of banking transactions for analytics, reconciliation, and regulatory reporting.
What you do not get from a vendor page like this is the failure log. No table that had to be resynced, no schema they refused to move, no hour of user-facing downtime, because this move is the replication layer under an analytics warehouse, not the core banking system itself. The trade they describe is a shorter refresh and a bill they can plan, paid for by a multi-day historical load on the biggest tables and by taking a connector that was beta until the vendor finished it for them.