A snapshot of the work — the problem we were called in on, what we found, and what changed. Each case is written from the engineering side — the symptom, the diagnosis, the change, and the verified result. While there are a very large number of our engagements which only deals with trivial performance troubleshooting and tuning, these case studies are some special cases involving complex troubleshooting. Names and figures are anonymised where required.
Tier-one retail bank
The IBM Datacap workflow was experiencing multi-second delays on every action rendering the tool unusable and the branches had to switch to manual operations. After identifying and resolving the root cause, request times improved drastically from seconds to milliseconds.
Tier-one retail bank
Customers were experiencing intermittent login failures. Some could log in on the first attempt, while others required two to five attempts before succeeding. The issue was widespread and not isolated to specific users or accounts.
Tier-one retail bank
After a hardware refresh for a statement generation application, pre-production testing showed the batch job taking more than twice as long as BAU on the older hardware. Resolving the issue reduced the cycle time from over 5 hours to under 2 hours.
Tier-one retail bank
Following the launch of a new online banking platform at a top-tier bank, a subset of customers were unable to access their accounts. Although login appeared to succeed, the system immediately logged them out, with the same behaviour observed across both web and mobile channels.
ISV Application
A reporting batch that previously took over 11 hours was reduced to under 3 hours. Within the same batch, one task that previously took about 50 minutes was tuned to complete in less than a minute.
Tier-one retail bank
A code change caused the regression and slowed down all the channels including ATMs.
Tier-one retail bank
Bank Islam nonce.
GLC
Customer-facing API was timing out under evening load. A combination of connection-pool tuning and a single index change brought p99 latency from 1.8s to 800ms with no application rewrite.
Tier-one retail bank
Year-end filing portal was crashing under predictable seasonal load. We re-engineered the queueing layer and absorbed a 4× traffic spike on existing infrastructure.