In the fast-paced world of fintech, efficiency isn't just a goal, it's a necessity. For one London-based fintech, their skyrocketing Azure bill had become a silent drain: £100,000 every month. That’s more than some startups raise in seed funding. And for a company their size, it was a significant expense.

When Koritsu AI stepped in, we didn’t just see a cloud bill, we saw a puzzle waiting to be solved. Our first discovery? Half of that cost, £50,000 per month, was coming from a single Cosmos DB database. Now, £50k for a database isn’t unusual if you’re processing terabytes of data for AI training or global transactions. But this? This was different.

While looking at the infrastructure only, the £50,000/month is justified-after all, the DTU (CPU/memory) consumption was maxing out and causing the high costs. But that's where Koritsu's differential comes in. Looking at infrastructure only is like looking at a slow truck and wondering why it can't go faster, when the real problem is that it's carrying too heavy a load.

At Koritsu, we understand that the application is what determines the load, so sometimes the problem exists at the intersection of infrastructure and application. Our deep dive revealed the culprit: a single, misconfigured query, running hundreds of times per second.

SELECT c FROM c WHERE c.data = 'JoihnDoe2018-02-01'

At first glance, it looked harmless. But in Cosmos DB, a NoSQL database built for horizontal scaling, this query was a financial time bomb. Unlike traditional databases that scale vertically (increasing the size of a single machine), Cosmos DB scales horizontally by spreading data across partitions.

And here’s the catch: if a query doesn’t specify which partition to search, it scans every single one, like checking every room in a skyscraper to find a single key.This was an exponential problem. As data grew, so did the cost, dramatically. By the time the fintech’s team was ready to act, their bill had ballooned to £150,000/month.

The fix? Surprisingly simple. By restructuring the schema so that the query targets the right partition, we slashed the cost back to £50,000 instantly. That’s £100,000 saved every month, just by understanding the intersection of infrastructure and application logic.

At Koritsu AI, we don’t just optimize infrastructure, we optimize what runs on it. Because the cloud isn’t just a "set it and forget it" tool. It’s a dynamic system where small inefficiencies can cost millions. And in this story, the ending was clear: a smarter query, a leaner bill, and a fintech free to focus on growth, not waste.