From 1 Record to 1,500: Fixing the Month-End Bottleneck

When Month-End Processing Stopped Keeping Pace

For a national fleet management provider, every month-end used to mean trouble. A MuleSoft Batch Integration connecting Salesforce with the provider’s internal application would slow to a crawl. It processed roughly one record every 2 to 3 minutes during the periods when volume was highest. Messages would get stuck mid-batch, sometimes trapped after a mid-process restart. Clearing them meant shutting down the runtime and manually deleting files just to start again. 

The provider’s Salesforce support and infrastructure teams spent those nights on extended calls. They worked through each incident by hand because there was no automated way to recover. Every failed run chipped away at trust in the platform and raised real questions about whether it could keep up as the business grew.

Finding the Bottleneck Hiding Inside the Code

Our MuleSoft team joined the investigation in the middle of a live outage. We worked alongside the client’s Salesforce engineers to trace the failure directly from the production Mule Runtime as it was happening. Reading through the stack trace in real time pointed to the root problem. Database Lookup calls were built into the processing loop itself. Every record pulled a fresh set of database connections instead of reusing what was already available. Multiplied across every record processed, that pattern was starving the runtime of the database connection pool it needed to keep up.

With the root cause identified, we rebuilt the integration architecture rather than patching around the symptoms. The new version restructured how the integration accessed the database, removed duplicated logic, and cut the number of calls needed to process each record. None of the solution added complexity. The goal was fewer, smarter calls, not more moving parts.

Rather than replace the integration in one step, we rolled out the new version in controlled phases. We checked performance at each stage before moving to full production. That intentional approach let the client validate the fix under real conditions and confirm it held up before betting month-end on it.

A Process That Finally Works the Way It Should

The results showed up immediately. What used to take 2 to 3 minutes per record now processes 1,535 records in 3 minutes and 25 seconds. Since the new integration went live, the client has completed month-end closes without a single production issue. This marked an unprecedented zero-incident streak for the company.

For the provider’s Salesforce and infrastructure teams, that means the end of the recurring war rooms. Month end is no longer something to brace for. It runs the way it was supposed to all along, freeing the team to spend that time on the business instead of the infrastructure behind it.