Co-Lending Platform — Souporno Mukherjee

Entry 02, The brief

The starting point

The brief was a spreadsheet.
Actually, five of them.

There was no PRD. No user flows, no problem statement. The preliminary requirements were vague, and the business team walked in with an Excel sheet and a sentence that still makes designers flinch:

The actual brief

“We have designed the UI. Just replicate it in Figma.”

Business team · handing over the spreadsheet below

The “UI” we were handed

	Axis		CBI		RBL		ICICI
	Leads	Amount	Leads	Amount	Leads	Amount	Leads	Amount
Axis	Eligible w/ Partner 1 only	Eligible w/ Partner 1 only			Eligible w/ 1 & 3	Eligible w/ 1 & 3	Eligible w/ 1 & 4	Eligible w/ 1 & 4
CBI	Eligible w/ 1 & 2	Eligible w/ 1 & 2	Eligible w/ Partner 2 only	Eligible w/ Partner 2 only	Eligible w/ 2 & 3	Eligible w/ 2 & 3	Eligible w/ 2 & 4	Eligible w/ 2 & 4
RBL	Eligible w/ 1 & 3	Eligible w/ 1 & 3	Eligible w/ 3 & 2	Eligible w/ 3 & 2	Eligible w/ Partner 3 only	Eligible w/ Partner 3 only	Eligible w/ 3 & 4	Eligible w/ 3 & 4
ICICI	Eligible w/ 1 & 4	Eligible w/ 1 & 4	Eligible w/ 4 & 2	Eligible w/ 4 & 2	Eligible w/ 4 & 3	Eligible w/ 4 & 3	Eligible w/ Partner 4 only	Eligible w/ Partner 4 only
Total	Total of P1 (unique)	Total of P1 (unique)	Total of P2 (unique)	Total of P2 (unique)	Total of P3 (unique)	Total of P3 (unique)	Total of P4 (unique)	Total of P4 (unique)
Eligible	Being sent to P1	Being sent to P1	Being sent to P2	Being sent to P2	Being sent to P3	Being sent to P3	Being sent to P4	Being sent to P4

Recreated faithfully from the original handover file← this was meant to be the interface

Hard to parse? It was hard for everyone, not just me. So instead of replicating it, I read between the lines. What the matrix actually encoded wasn’t a screen; it was a set of decisions the ops team was making every day about partner overlap, exclusivity, and residual pools.

Co-lending Docs.xlsxSheet0 · 10,415 rows

Lead Id	Product Id	Overall Status	Assigned Partner	Lead Processing Status	Outstanding Principal	Rate of Interest
SCUBL0001A65	UBL	RECEIVED	YUBI	FAILURE	5,13,706	17.49
SCUBL0001A66	UBL	RECEIVED	,	PENDING	8,20,450	16.75
HLSA0001B12	HL	RECEIVED	,	PENDING	42,10,000	9.15
HLSA0001B13	HL	RECEIVED	,	PENDING	38,75,200	9.40
LAPC0034F09	LAP	RECEIVED	,	PENDING	1,10,00,000	11.20

…10,410 more rows. The other four files looked like this. Someone was deciding these one by one.

Entry 03, The users

Research

I set the “Excel UI” aside
and called the people behind it

The handover said replicate the spreadsheet. Instead, I scheduled calls with the two people who actually lived inside it, the Credit Strategy team in Kurla. What they described wasn’t a UI requirement. It was a weekly ritual held together by meetings, favours, and very large spreadsheets.

User 01

Sujata Nachare

Credit Strategy · Kurla

Business Ops · Decision Owner

“By the time the meeting ends, I know exactly how the loans should be split. The next three days are spent making the spreadsheet agree with me.”

Context

Partakes in key business decisions for Piramal Finance’s co-lending book
Anchors the weekly allocation meeting, which loans, how many, to which banks
Balances current capacity, guidelines from business heads, and available partner banks

Goals

Translate the meeting’s decisions into allocations, same day, not same week
Honour partner commitments and business-head guidelines exactly
Be able to show why any loan went to any bank, months later

Frustrations

Decisions leave the meeting as intent, nothing captures them as executable rules
Depends on developers or PMs to run the BRE; their queue, her deadline

Comfort zoneExcel, email, meetings

Needs from designSpeak business, not SQL

User 02

Deepa Patil

Credit Strategy · Kurla

Business Ops · Execution Lead

“There is no platform. I mail the sheet to whoever can run the engine, wait, get a sheet back, and then the real work starts, row by row.”

Context

Partakes in key business decisions for Piramal Finance’s co-lending book
Owns the follow-through: uploads the Excel sheets to the BRE and works its output
Sits with the returned sheets and assigns loans to banks, manually, one by one

Goals

Run the BRE herself, no developers, no product managers, no waiting
Bulk-assign the obvious cases; spend her judgment only on the exceptions
Trust that exclusions and edge cases won’t slip through 10,000 rows

Frustrations

Output arrives as yet another Excel; assignment is still fully manual
One mis-assigned row is invisible until a partner bank flags it weeks later

Comfort zoneExcel power user

Needs from designBulk first, exceptions second

As-is journey

One decision, five hand-offs

What Sujata and Deepa described on our calls, laid end to end. The decision took an hour. Everything after it took days.

Step 01 · Weekly

The allocation meeting

Team decides which loans, how many, and to which banks, based on current capacity, business-head guidelines, and available partners.

Step 02

Decisions become… memory

The meeting ends with a shared understanding of how loans should be assigned. It lives in heads and meeting notes.

nothing executable leaves the room

Step 03

Beg a BRE run

Excel sheets are sent to developers or product managers to run through the Business Rules Engine. There is no platform.

blocked on other teams’ bandwidth

Step 04

Excel returns as Excel

The BRE output lands back as spreadsheets, eligibility verdicts across thousands of rows.

10,415 rows, no summary view

Step 05 · Days

Assign, row by row

Sujata and Deepa sit with the sheets and start assigning loans to banks manually, reconciling against what the meeting decided.

slow, error-prone, unauditable

Design target: keep step 01, it’s good judgment. Collapse steps 02-05 into one morning.

The Brainstorm(s)

Entry 04, Process log

Process

Thinking on whiteboards, tables,
and whatever else was nearby

This project was drawn before it was designed. The information architecture went up on office glass; the core idea was literally scribbled on a desk with a marker.

Whiteboard sketch of platform information architecture

Log 01

Mapping the territory

Products (HL, LAP, UBL) × Partners (Axis, SBI…) × Programs. The IA had to hold a matrix that business could grow without redesign, new partner banks were signing every quarter. This was sketched while discussing the initial brief with a product manager.

Sketch: product tabs for Home Loan, LAP, UBL

Log 02

Structure before pixels

Product-first navigation. Every allocation session starts by answering “which book am I working today?”, Home Loan, LAP, or UBL.

Sketch: loan table with per-loan partner checkboxes

Log 03

First instinct: a smarter table

Loans with eligible partners as checkboxes, “assign all to Axis / SBI” shortcuts. Better than Excel, but still asking ops to make 10,000 micro-decisions. It felt like faster manual labour, not a different job.

Log 04, Brainstorming exhibits

Three options went to the table

The first two options came out of brainstorming sessions, competent, expected, table-centric. They’re preserved in the Figma file’s “Presentation” page as the roads not taken.

Option 1, Smart table

Filterable loan grid with inline partner assignment. Faster Excel, same mental model.

Option 2, Guided batch flow

Upload → BRE verdicts → stepped bulk assignment. Better, still form-driven.

Sketch over UI: upload file, run BRE, execution summary

Log 05

Let the engine do the reading

Upload the business file, run the BRE (business rules engine) on every row, and return a verdict: eligible, rejected, hold. 10,415 rows executed in one pass. The human should judge, not scan.

The allocation rule sentence written in marker on an office desk

Log 06

The desk moment

Option 3 didn’t arrive in a session. It struck mid-conversation, while talking the problem through with colleagues. The idea was fast, and so it wouldn’t escape, I jotted it straight onto the table with a marker: “I want to allocate 1,000 loans to Axis Bank, except LAN ID ___.” The sentence was the interface. The seed was sown on office furniture!

Log 07

Designing the sentence

“Good morning, Soup. Let’s set the allocation rules for the day.” The rule became a fill-in-the-blanks sentence with structured inputs, readable by a human, executable by a machine. No query builder. No training manual.

Annotated near-final rule builder screen with partner coverage bars

Log 08

Pressure-testing with ink

Iterating on near-final screens by hand: showing what a rule will do before it runs (“activating this rule will allocate 70% of cases”), visualising partner overlap between Axis, CBI and SBI, and keeping every activated rule as replayable history. The data visualisation part was key to making the users understand the current allocation status.

Entry 05, The design

The design

Ops doesn’t fill filters anymore.
They write the rule.

The shipped Rule Builder, recreated here. This is the actual interaction, try it.

ColendingSouporno ▾

☼ Good Morning, Souporno — let’s set the allocation rules for the day.AXIS · 40/500 ~ ₹10,000 CR READY

I want to allocate loans in stage of total amount except LAN ID HLSA0001, HLSA00012 to .

Activating this rule will allocate ~68% of eligible cases

Rule history, every rule is auditable & replayable

One sentence. The BRE validates every case behind it, exclusions are respected, partner overlaps are resolved by priority, and rejected cases re-route to the next eligible bank, Axis first, SBI if Axis declines.

Early sketch of the allocation rule builder

⇔

SketchShipped

Drag, from first sketch to shipped screen.

Entry 06, Movement

User flow

Deepa’s Tuesday, redrawn

The flow the platform shipped, one person, one morning, no hand-offs. The weekly meeting still happens; it’s good judgment. Everything after it now takes minutes.

Phase 1, Set up the day

Good Morning, Deepa

→

Select product

HL · LAP · UBL

→

Upload loan file

the same Excel business already makes

→

BRE runs, all rows

10,415 rows in one pass

→

Run summary

eligible / rejected / hold

Does a rule cover it?

Yes, the bulk (rule-based)

Write the rule

“I want to allocate 40 loans… except LAN ID…”

→

Preview impact

“will allocate ~70% of cases”

→

Activaterule saved to history — auditable, replayable

No, the exceptions (manual)

Open un-allocated pool

overlap bands show eligible banks per loan

→

Assign by judgment

the human decides, but only here

Phase 2, The system settles it

Allocate & resolve overlaps

priority order · exclusions respected

→

Partner declines? Re-route

Axis rejects → next eligible bank (SBI)

→

Allocated view

day-wise batches per bank

→

Pool to partners

250+ loans · ₹55 Cr+ via bulk upload

→

Download report

the paper trail, without the paper

↺ next batch? write another rule

Before, five hand-offs

meeting → memory → beg a BRE run (devs/PMs) → Excel returns as Excel → assign row by row

Days

After, zero hand-offs

meeting → login → upload → rule → activate → report

One morning (−80% TAT)

Entry 07, Structure

Information architecture

A structure that absorbs new banks
without a redesign

The IA mirrors how Credit Strategy actually thinks: pick the book, see the state of it, act on it with rules, and account for every action. Partner banks and programs are configuration, not architecture, a new bank signs, the business team adds it, nothing gets redesigned.

Co-Lending Platform, sitemapledger of screens · v. final
0Accesslogin · role-based entry for business ops
1✓Login & accountwho ran what, tied to a name
0Product selection“which book am I working today?”
1✓Home Loan / LAP / UBLevery session starts here
0Dashboardthe state of the book, at a glance
1✓Allocated loansday-wise batches · filter by product, partner, period
2✓Un-allocated poolpartner-overlap bands, where the next rule will bite
3✓Table & detail viewsdrill from batch to single loan
0Allocationthe working core of the platform
1★Upload & BRE runself-serve, no developers, no queue
1.1Run summaryeligible / rejected / hold, verdicts, not rows
2★Rule Builderthe sentence, with impact preview & exclusions
2.1Rule historyevery rule auditable & replayable
3✓Manual allocationjudgment, reserved for exceptions
0Poolswhat gets sent to partner banks
1✓Pool creation & detailper-pool view, e.g. UBLA901119
0Configurationgrowth without redesign
1✓Partner & program setupnew bank signs → business team adds it here
0Reportsthe paper trail, on demand
1✓Select · filter · downloadreport type, date range, loan & partner filters
✓ mapped in Figma master file★ the two moves that changed the job

Entry 09, Impact

Impact

Presented by the Design Director
as one of the best designs of the year

At the April ’25 company townhall, Co-Lending was showcased first among the “Top 3 Designs Delivered”, running the BRE on the whole database and co-lending every case in a few clicks. Turnaround time fell by 80%; 12,500+ leads processed; 250+ loans worth ₹55 Cr+ pushed to partners via bulk upload.

Exhibit A

Townhall slide: State of OKRs, UX Design, Top 3 Designs Delivered, Co-lending

The actual townhall slide, “State of OKRs · UX Design · Apr ’25”

“Souporno, surprised us with a very innovative approach which helped us reduce our efforts by a huge margin. He actually understood how we work and devised a solution that fit our everyday workflow.”

Aniya Paul · Business Stakeholder · Co-lending

“Soup, kept showing various approaches during the entire project. He didn’t stop until he felt the problem was actually solved. He went to the root of the problem and solved it, which makes him stand out.”

Chintan Jat · Design Manager

First, what is co-lending?

The brief was a spreadsheet.
Actually, five of them.