Cut Pintu’s infrastructure cost by 20–30% overall across Pintu Pro, User and Payment Gateway, by merging services, switching off what nobody used and load-testing every change before it went out.
Role
Senior Engineering Manager, then Head of Engineering
Team
Engineering teams for Pintu Pro, User and Payment Gateway
Timeline
Jun 2025 – Apr 2026, in two phases
Domain
Infrastructure · FinOps
One example from Pintu Pro: matching engines went from one per token to three, with tokens grouped. Token pairs shown are illustrative.
The challenge
Spending less without slowing the teams down
Pintu had a cost-saving target, and I already reviewed infrastructure cost on a regular basis. Cost monitoring was in place, so we knew where the money went. The hard part was deciding what to cut.
Cost work is easy to start and hard to keep going. The goal was to bring spend down in a way that didn’t put reliability at risk or slow down product teams that were still shipping.
The cheapest system is no good if it falls over during a market spike.
01Reliability. Trading traffic is spiky, so capacity can’t simply be cut to the average. Every change had to keep the system’s capacity where it was.
02Priorities. Cost work takes engineering time away from the roadmap, so it had to be weighed against features.
03Staying down. Savings only count if costs don’t creep back up a few months later.
My role
Planning it and seeing it through
I ran the work in two phases. As Senior Engineering Manager, I led cost saving on Pintu Pro from June 2025 to January 2026. As Head of Engineering, I did it again for User and Payment Gateway from January to April 2026. The work stayed within engineering.
01Plan. Used cost monitoring to find where the money went, and chose changes that lowered cost without lowering capacity.
02Alignment. Some P2 features had to wait. The trade-off was calculated carefully, and all of senior management agreed to it.
03Keeping it down. Reviewed cost with each team in our weekly 1:1s.
Approach
How we did it
1. Merge services that do the same job
On Pintu Pro, matching engines went from one per token to three, with tokens grouped across them. This is the cost-saving roadmap from the Pintu Pro case study.
2. Remove what other services can cover
Small services whose job another engine could already do were shut down.
3. Turn off what nobody uses
Staging was switched off at night and on weekends, when no one used it, which cut its cost by 30–40%. Unnecessary logging was removed.
4. Load-test every change
We benchmarked and load-tested before and after each change, so we knew the capacity on both sides. If the difference was large, the change didn’t go ahead.
5. Make it a habit
Cost became a standing topic in my weekly 1:1 with each team, so it stayed down after each phase ended.
Key decisions
Trade-offs we made on purpose
Decision 01
Capacity is the line we don’t cross
Every change was benchmarked and load-tested before and after. A change that lowered capacity too far didn’t ship, however much it saved.
Trade-off: Some savings were left on the table because the capacity after the change was too far from before.
Decision 02
Cost work ahead of some P2 features
Engineering time went to cost saving, and some lower-priority features waited. The impact was worked out in detail first, and all of senior management agreed.
Trade-off: Some P2 features were delayed.
Outcome
What changed
20–30%
Lower infrastructure cost overall; the drop varied by team
3
Matching engines on Pintu Pro, down from one per token
2
Phases: Pintu Pro, then User and Payment Gateway
Every change was load-tested before it went out, so the savings came without giving up capacity, and the weekly 1:1 reviews kept costs from creeping back.
What I learned
What I’d keep, and what I’d change
Switch environments off when nobody is using them. Turning them off outside working hours cut their cost by 30–40%.