Saveo Healthcare · UX Designer · 2021 — 2023
Money moved every day. The record of it lived in a spreadsheet.
Delivery executives collected ₹10–30 lakh in invoice payments daily across four payment modes. Reconciliation happened one invoice at a time, in Google Sheets, with three teams keeping three different versions of the same status. I redesigned the financial backbone underneath it — and made settlement deliberately slower.
What I found
Three teams. Three versions of the same invoice.
Collections were logged in a sheet first and entered into the system later. By the time an entry reached finance it had been retyped at least once, and its status meant something different to whoever was reading it.
Illustrative reconstruction of the collections sheet, built from observed workflows. Not a production screenshot.
- The status column was an opinion, not a state. Cash counter, finance and credit each read the same invoice differently — pending, unverified, missing — with no shared definition to arbitrate between them.
- Reconciliation was one invoice at a time. No batching, no bulk settlement, no way to close a delivery executive's session as a unit.
- Errors surfaced days later, untraceable. A discrepancy could be found, but not attributed — nothing recorded who changed what, or when.
Before designing
I mapped the machine before I drew a screen.
Three weeks sitting with the teams who actually handled the money — watching collection sessions end to end, tracing where a number changed hands, and noting every point where what was said and what was entered came apart.
Cash counter team
- Daily collection sessions observed end to end
- Handoffs happened verbally, then got written down
- What was reported and what was entered diverged routinely
Delivery executive reporting
- No standard format — verbal handoffs only
- Re-entry into the system after the sheet
- Errors compounded at every transfer
Bank statements & cheques
- Manual uploads, no repository, no versioning
- Cheque bounces had no tracked state at all
- Finance found discrepancies too late to act on them
Speed versus financial correctness. Operations had optimised for the first. Finance needed the second.
The core tension — and the thing every subsequent decision had to answer to.The shift
Build the backbone first.
The brief was a screen redesign. The finding was that no screen would survive a broken record. So I proposed sequencing it: get operations and finance onto one foundation, then layer capability on top of something that holds.
Foundation
Phase 01- Centralised bank statement repository
- Invoice-level reconciliation against bank credits
- Structured state definitions shared across every team
- Clear ownership assigned per invoice state
Scale
Phase 02- Bulk settlement tooling
- Cash session tracking and batching
- Cheque lifecycle management, bounce to resolution
- Unified invoice landing across teams
- GTM bulk deposit handling
Sequencing was the design decision. Everything in phase two would have been feature sprawl on a broken base.
System design
Four states became the shared language.
Not a status badge. A contract — each state defines what has been proven about the money, who owns the invoice at that moment, and what has to happen before it can move. Settled became unreachable without a bank match.
Initiated
Invoice created, collection logged
Unverified
Awaiting bank confirmation
Partial
Part-payment mapped
Settled
Bank mapped, finance signed off
Illustrative reconstruction of the redesigned state system. Not a production screenshot.
The hard call
Settlement got slower. That was the point.
Every validation gate I added cost the cash counter time. Two minutes an invoice became four to five. I took that to finance and product before building it, because a slower workflow you can't defend is just a worse workflow.
Settlement time
~2 minutes
Per invoice — fast, and unvalidated
Duplicate entries
High
No validation layer; duplicates slipped through regularly
Discrepancy visibility
None
Gaps discovered at end of day, or later
Finance confidence
Low
Closed invoices frequently reopened for correction
Settlement time
4–5 minutes
More steps, structured validation — intentionally slower
Duplicate entries
Reduced
Validation gates block duplicates at entry
Discrepancy visibility
Real time
Flags surface as invoices move between states
Finance confidence
High
Every state change logged with a timestamp and an owner
We chose integrity over speed — deliberately, with finance and product aligned before a line was built.
Figures reflect operational behaviour observed during discovery and after rollout. Not formally instrumented.Where it landed
From reactive correction to proactive clarity.
The operations teams never had to stop working to adopt it — the states mapped onto what they were already doing informally, so the change was in what the system recorded, not in how the day ran.
Comparison drawn from qualitative observation and stakeholder feedback during and after the redesign. Not A/B tested.
How I work
Design as infrastructure.
This project starts where I prefer to start — in operational complexity, with the machine mapped before a single screen is drawn. The interface was the last decision, and the easiest one, because everything underneath it had already been settled.
More of the same thinking, at platform scale.
Nineteen product areas of a derivatives platform, held together by one token system.