Azure · Cosmos DB · Performance
A SaaS Azure bill forecast down by 75% in six days
A SaaS reporting layer was throwing throttling errors and driving up the Azure bill. In six days we fixed the root causes and the client's own forecast fell from about $6,000 a month to about $1,500.
- 75%
- Lower monthly Azure forecast
- 5.0
- Client rating
Client's Azure Cost Management forecast, confirmed in their Upwork review
Upwork review by Hudson G., August 2026
The problem
A SaaS company's reporting layer was reading directly from the same Cosmos DB containers their application was writing to.
Under load that produced 429 throttling errors and sharp spikes in request unit consumption. Two problems at once: the platform was slowing down for customers, and the bill was climbing with no obvious cause. Their Azure spend was around $6,000 a month and rising.
What we built
The expensive default fix is to raise provisioned throughput until the errors stop. We did not do that.
- Diagnosed the throttling down to its root causes in query patterns and partition strategy.
- Fixed the queries and partitioning that were driving request unit consumption.
- Separated the reporting workload from the transactional path, so reports no longer competed with the live product.
- Handed over a re-architecture plan for the reporting layer, so the problem would not come back as they grew.
The result
The throttling stopped and reporting got faster. The client's own Azure billing forecast fell from roughly $6,000 a month to roughly $1,500, about $54,000 a year projected, on six days of work.
These are forecast figures from Azure Cost Management, not twelve months of realised invoices, and we describe them that way.
In the client's words: "Furqan is a top-tier Azure engineer. He jumped into our stack and immediately resolved our Cosmos DB throttling (429 errors) and RU spikes on our reporting layer." Hudson G., 5.0 rating on Upwork.