AI Document Extraction for CRM
Overview
In enterprise banking workflows, Relationship Managers (RMs) spend a significant portion of their day manually extracting data from financial documents, bank statements, income proofs, identity documents, and compliance files just to complete form-filling and lead processing. This process was slow, error-prone, and mentally exhausting.
ROLE
Product Designer
RESPONSIBILITIES
I worked on designing an AI-powered Document Extraction feature within the BusinessNext CRM that automatically reads, understands, and extracts critical data from structured and unstructured financial documents with accuracy, allowing RMs to focus on decision-making instead of data entry.
COLLABRATION
I collaborated with a cross-functional team that included the VP, Product Manager, Data teams and Engineers. My role spanned research, workflow design, interaction patterns, prototype, validation states, and handoff to development, ensuring the feature was scalable and compliant across BFSI use cases.
IMPACT
Designed a single document ingestion flow for bank statements, ID proofs, and financial reports
Used confidence tagging to show when AI was certain vs. when review was recommended
Enabled auto-fill with manual override, balancing speed and control
Kept AI logic hidden from RMs to avoid cognitive overload - only results mattered
TIMELINE
2 Weeks
User Stories
I created business stories that encompassed feature requirements at a high level and reviewed them with the PMs to confirm which ones we wanted for the MVP.
Hypothesis
I created a Hypothesis document to summarise the Scope of our Project.
Persona
I reviewed existing user insights gathered over time from past discussions with PMs, team managers, and end users, along with their understanding of the product, market competition, and industry trends. I also conducted interviews with three end users to understand their daily workflows, pain points, and goals. Based on these insights, I synthesized a persona representing the primary user of this product: the Relationship Manager.
Information Architecture
Once I had a clear understanding of the user, I created an Information Architecture map and collaborated with PMs to define and prioritise what would be included in the MVP.
Doc AI - Custom Document Type - Training Journey
Doc AI - Consumption Journey
Doc AI - Extraction Journey
Ideation
I started by sketching the user flows on paper to quickly explore how each journey would work. These sketches were then converted into mockups and prototypes. Through multiple iterations, the designs were reviewed and validated with stakeholders, engineering, and end users through usability testing.
Doc AI - Custom Document Type - Training Journey
Level 1
Level 2
Level 3
Final Output Screens
Doc AI - Consumption Journey
Level 1
Level 2
Final Output Screens
Doc AI - Extraction Journey
Level 1
Level 2
Final Output Screens
Results & Impact
Faster RM workflows: Automatic document training, extraction, and processing minimise manual data entry and review efforts across all three use cases, allowing RMs to complete form filing and case management much more quickly.
High accuracy & trust: The Document AI achieved approximately 99% extraction accuracy, enhancing data consistency and decreasing rework caused by manual errors.
Scalable adoption: Admin-led template training and CRM configuration allowed the solution to expand across various document types, while RMs benefited from a straightforward, seamless auto-fill workflow without backend complexity.ies