An architecture review that prevented a six-month rewrite for a Riyadh fintech
Two weeks of senior consulting instead of a rebuild the board had already priced.
- Product
- Payments platform
- Technologies
- Java, Kafka, PostgreSQL, AWS
- Timeframe
- 2026
A payments core was blamed for outages the architecture could actually survive — once the real bottleneck was found.
The challenge
Latency spikes during settlement windows were read as proof the monolith had to go. A rewrite would have frozen the roadmap for two quarters right before a licensing milestone.
No one on the team had built a system at that transaction volume before, so every opinion was a guess.
The decision had a deadline attached: a licensing milestone one quarter away meant the platform could not be frozen for a rebuild, and no one internally had run a system at that transaction volume.
In a nutshell, when the company came to us it was grappling with:
- Latency spikes blamed on the architecture.
- A rewrite already priced by the board.
- No in-house experience at that volume.
- A licensing milestone one quarter away.
Key solutions
An architecture review for a payments platform
Our consultant reviewed the data path, traced the spikes to lock contention and a missing partition strategy, and produced a costed remediation plan with three phases the team could ship incrementally.
We paired with their engineers through phase one, then handed over the runbook and the load-test harness.
Solutions we provided:
Evidence before opinion
We instrumented the data path and measured, so the rewrite decision was made on numbers rather than on intuition.
A phased remediation plan
Three phases the team could ship incrementally, each with a cost, an owner and a measurable target.
Handover, not dependency
Runbook, load-test harness and a pairing period, so the client's engineers own the fix from day one.
Team on the project
One consultant, then a pairing team of three
- Principal architect
- Data engineer
- SRE
The engagement started with one principal architect and grew to three specialists for the remediation phase, then wound down as planned once the runbook was handed over.
The client's CTO owned the decisions; our architect presented findings to the board and joined the regulator's technical review.
Two on-site weeks in Riyadh framed the engagement — one for discovery, one for the handover — with the rest remote.
The engagement ran on site for the moments that mattered:
- Discovery week in Riyadhwith the platform and data teams
- Board presentationwhere the rewrite decision was reversed
- The regulator's technical review
- Handover weekrunbook, load tests and a pairing period
- Saudi ArabiaSA
- UAEAE
- GermanyDE
The remediation plan is now the client's own roadmap — their engineers ship each phase, with our architect available for review rather than execution.
They measured before they recommended. That is the difference between advice and an opinion.
Results
- -82%
- p99 settlement latency
- 2 weeks
- engagement length
- 6 months
- rewrite avoided
- 3
- phases handed over
Two weeks of the right person's time replaced a project we had already budgeted. That is the cheapest decision we made this year.
Project completed:
A rewrite avoided, a platform that holds
Two weeks of senior consulting turned a planned six-month rebuild into a three-phase remediation the client's own engineers shipped incrementally.
p99 settlement latency fell by 82%, the licensing milestone was met on the original platform, and the client kept the roadmap it had already committed to.
We measured before we recommended: instrumented the data path, found lock contention and a missing partition strategy, and handed over a costed plan with a runbook and a load-test harness.
The best consulting outcome is the project you do not have to run. Two weeks replaced two quarters.