India's largest Bank · Enterprise-scale · High complexity · Massive data

Self-serve analytics at Branch scale, under 3 seconds load times

The bank ran on custom-coded reports with one hard rule: everything should load under 3 seconds. I worked as the Analytics Solution Engineer who helped implement the Tableau dashboards meeting all requirements.

  • 3,000+peak concurrent users at go-live
  • 500–800Mrows per table, refreshed daily
  • < 3sload time, the bar that never moved

scroll the story

01 — The problem

Reports you had to code, not use

Custom backend legacy code meant every small change had Developer dependency. Inefficient data architecture led to slower dashboard load times.

A hand-coded SQL tabular report next to the self-serve dashboard that replaced itA hand-coded SQL tabular report next to the self-serve dashboard that replaced it

02 — My role

I supported the whole arc, from RFP to Adoption

I wore multiple hats - responding to RFP technicalities, guiding Server installation, Designing and building dashboards and pushed adoption.

5 stages

RFP → demo → install → build → adoption

Timeline: RFP technical response, demo and training, infra sizing and Linux install, use-case discovery and build, adoption and best practicesTimeline: RFP technical response, demo and training, infra sizing and Linux install, use-case discovery and build, adoption and best practices

03 — The craft

Fast at scale was the hard part

Optimized the backend data models to hold under 3 seconds load times with Row level security.

  • 500–800Mrows per table, multiple tables
  • ~7M / dayrows added per table
Performance dashboard: 3,042 peak users, 1.8 second average load, 6 server nodes, all under the 3 second targetPerformance dashboard: 3,042 peak users, 1.8 second average load, 6 server nodes, all under the 3 second target

04 — The outcome

Analytics reached the Branch Managers, then spread.

Analytics left the developer's queue for the branch floor - and then more use cases pulled in.

  • Branch-levelself-serve, not the dev queue
  • 3 use casesdeveloped since then
One core platform connected to many branch-level use casesOne core platform connected to many branch-level use cases

05 — What I'd carry forward

The install and 3 seconds SLA was solvable but the hard part was earning trust of people who had lived in Excel for years.

  • Designing & optimizing for performance from day one
  • Keeping right business stakeholders in loop for right feedback to align the design
  • Conducted constant trainings & education workshops to make the people comfortable with the tool