Case Study — FinTech · AI · Enterprise
BICICI Smart Data
Workspace
I designed an AI-powered platform for BICICI BNP Paribas Bank that merges 3 separate databases into a single unified view — with intelligent product recommendations ranked by score — enabling commercial agents to recommend the right product to the right client.
01 — The Business Problem
Three databases,
zero unified view
BICICI Bank faced a critical operational challenge: client data was scattered across three independent databases — Clients, Operations, and Deposits — with no unified view. Commercial agents wasted hours manually matching information across systems, leading to low product adoption, missed cross-selling opportunities, and a poor understanding of each client's full financial profile.
3 Separate Data Sources
No unified client profile
Clients database, Operations database, and Deposits database — each maintained independently. Agents had to manually cross-reference all three to understand a single client.
The Consequences
Lost revenue & wasted time
No intelligent product recommendation system. Low equipment rate (clients owned too few bank products). Commercial agents wasted hours on manual data matching instead of selling.
02 — The Business Need
Increase equipment rate by
recommending the right product
The objective was clear: increase the client equipment rate (taux d'équipement) by recommending the right banking product to the right client using AI. The bank needed a single platform that automatically merges the 3 databases and generates product recommendations ranked by score — turning fragmented data into actionable intelligence for every commercial agent.
03 — My Role
Requirements to
high-fidelity in 1 month
Phase 1
Requirements Gathering
Collected business needs directly from BICICI stakeholders. Understood the 3-database structure, the manual matching process, and what commercial agents needed to see.
Phase 2
Wireframing
Created low-fidelity wireframes for the complete flow — login, 3-step file upload, AI processing, and results table. Iterated with client feedback at each stage.
Phase 3
High-Fidelity Design
Designed the final UI — login page, solution presentation, upload interface (all 3 steps), dashboard with merged output table, and distribution workflow to agency heads and sales reps.
04 — Solution Architecture
Upload, process,
recommend, distribute
The platform follows a linear flow: bank employees log in, upload 3 database files sequentially (Clients → Operations → Deposits), the AI processing engine merges all files into one unified table, classifies clients by segment, scores products per client, and recommends the best product. The output can be downloaded as Excel/CSV or sent directly to agency heads and sales representatives.
Step 1–3
Upload 3 Databases
Clients → Operations → Deposits
AI Engine
Processing & Scoring
Merge → Classify by segment → Score products → Recommend
Output
Unified Table
Client + Segment + Products ranked by score
Distribution
Agency Head
Distribution
Sales Representative
Export
Download Excel/CSV
The final output table contains: Client name, Current sub-segment, Recommended sub-segment, Agency, Entity, Account manager, Current products (ranked by score), Recommended products (ranked by score), Current equipment rate, Target equipment rate, Products sold, Client contact, Contacted status, and Comments.
User flow for the system
05 — Key Screens
From login to
AI-powered recommendations
Login — employee authentication for the Smart Data Workspace platform
File Upload — 3-step sequential upload with drag-and-drop, progress indicators, and file validation
Dashboard — unified output table with AI-scored product recommendations, filterable by agency and segment
06 — Results & Impact
Before and after —
measurable improvement
| Metric | Before | After |
|---|---|---|
| Data processing time | Several days (manual) | Minutes (automated) |
| Data sources | 3 separate files | 1 unified table |
| Product recommendation | No system | AI-powered scoring |
| Commercial agent productivity | Low | High (single view of client) |
| Equipment rate | Baseline | Increased (target now visible per client) |
07 — Reflection
Designing AI for
non-technical users
BICICI was a masterclass in designing complex AI logic with a simple interface. The underlying system is sophisticated — merging 3 databases, classifying clients into segments, scoring every product per client — but the user interface needed to feel like uploading 3 files and getting a table. If a commercial agent couldn't do it in under 5 minutes without training, the design had failed.
The 3-step upload flow was a deliberate design decision. I could have designed a single upload page for all 3 files, but testing showed that sequential steps with clear progress indicators reduced errors and increased confidence. Users knew exactly where they were, which file they needed next, and what would happen at the end.
The 1-month timeline forced extreme focus. No feature creep, no nice-to-haves, no optional modules. Every screen had to earn its existence. That constraint produced a cleaner product than a longer timeline would have — because there was no room for complexity that didn't directly serve the business need.