The calculation and notice engine behind a £500bn loan book.

Our client administers loans on behalf of lenders across many facilities and jurisdictions. The numbers they produce sit at the centre of that work. Interest, payment-in-kind schedules, fees and payment dates all have to be right, because every figure ends up in a notice to a borrower or a lender, and a mistake is expensive in both money and trust.

Some of the existing calculations needed putting right, and the firm could not simply switch them off while that happened. They needed the figures corrected, proven against known results, and rolled out in a way that let the business move across at its own pace, without any loss of confidence in what was already running.

We built the calculation and document API on Azure Functions in Python, covering interest for lenders and facilities, payment-in-kind schedules, risk-free-rate calculations, and the full set of payment dates the business runs on. Where calculations needed remediating, we delivered corrected versions alongside the originals rather than in place of them, so the firm could verify the new results and migrate when it was ready.

Correctness is the whole point of a system like this, so we pinned it down with tests. Close to 1,900 automated tests cover the calculations and date rules, which means a change can be made and its effect understood before it reaches anyone. Notices to borrowers and lenders are generated as branded PDFs through an Azure Service Bus pipeline, with each facility agent's own logo applied. Every release moves through five environments, from Dev to Production, gated by tests, a route-registration check and a mandatory changelog entry.

The firm has a calculation and notice platform it can trust with a £500bn book. The remediated figures match their expected results and are held in place by the test suite, so the business can change what it needs to without putting at risk what already works. Releases are routine: a tagged version moves from one environment to the next, each step tested, checked and recorded, with a clear way back if it is ever needed.