From
Review to
Decisions
From
Review to
Decisions
From
Review to
Decisions
60,000
Document Processed
80%
Productivity improvement
Hours → Minutes
Contract Review
CIO Excellence Award
Recognition
60,000
Document Processed
80%
Productivity improvement
Hours → Minute
Contract Review
CIO Excellence
Award
Recognition
60,000
Document Processed
80%
Productivity improvement
Hours → Minute
Contract Review
CIO Excellence
Award
Recognition
Context
Where Docsmart was?
Previous flow
Previous flow
01
Create Binder
02
Upload Documents
03
Process using template
04
Relevant clauses
01
Create Binder
02
Upload Documents
03
Process using template
04
Relevant clauses
01
Create Binder
02
Upload Documents
03
Process using template
04
Relevant clauses
At this point finding information was no longer the hardest part of Docsmart
At this point finding information was no longer the hardest part of Docsmart
Understanding the opportunity
The product answered what? Conversations questioned why and how?
Despite having access to the right information,
analysts still spent significant effort manually comparing, validating, and synthesizing contract insights.
We Were Designing For
The Gaps We Couldn't Ignore
The Future
DocSmart's goal was to create a unified review workspace where analysts could discover, analyze, compare, and validate contracts before reaching confident decisions—with AI seamlessly supporting the workflow.
Docsmart
Despite having access to the right information,
analysts still spent significant effort manually comparing, validating, and synthesizing contract insights.
We Were Designing For
The Gaps We Couldn't Ignore
The Future
DocSmart's goal was to create a unified review workspace where analysts could discover, analyze, compare, and validate contracts before reaching confident decisions—with AI seamlessly supporting the workflow.
Docsmart
Despite having access to the right information,
analysts still spent significant effort manually comparing, validating, and synthesizing contract insights.
We Were Designing For
The Gaps We Couldn't Ignore
The Future
DocSmart's goal was to create a unified review workspace where analysts could discover, analyze, compare, and validate contracts before reaching confident decisions—with AI seamlessly supporting the workflow.
Docsmart
Despite having access to the right information,
analysts still spent significant effort manually comparing, validating, and synthesizing contract insights.
We Were Designing For
The Future
The Gaps We Couldn't Ignore
DocSmart's goal was to create a unified review workspace where analysts could discover, analyze, compare, and validate contracts before reaching confident decisions—with AI seamlessly supporting the workflow.
Docsmart
But there were three things standing in the way.
The review workflow had no clear starting point
Searching across thousands of contracts wasn't scalable.
AI needed to fit the product—not redefine it
But there were three things standing in the way.
The review workflow had no clear starting point
Searching across thousands of contracts wasn't scalable.
AI needed to fit the product—not redefine it
But there were three things standing in the way.
The review workflow had no clear starting point
Searching across thousands of contracts wasn't scalable.
AI needed to fit the product—not redefine it
But there were three things standing in the way.
The review workflow had no clear starting point
Searching across thousands of contracts wasn't scalable.
AI needed to fit the product—not redefine it
Understanding the problems was only the first step. The real design challenge was defining an experience that balanced analyst workflows, business priorities, and technical constraints.
Working Through Ambiguity
There wasn't one obvious answer—only better questions.
REJECTED
Redesign the whole platform.
Make AI the primary experience.
Treat search as end of review.
Solve everything in one release.
What
I Already
Knew
Relevant clauses are already extracted
Review continues outside DocSmart
AI capability is expected later
Repository continues growing
Questions?
that shaped every design
Rejected
Design Decision
What capabilities could reduce the most manual effort without adding another unnecessary steps?
What information turns extracted contract data into relevance inputs for analysis and evaluation?
How should metadata, extracted entities and relevance signals work together to narrow search results?
Should natural-language search extend the existing filter-based search, or introduce a new way to navigate the product?
Why do traditional filters become insufficient at enterprise scale?
How can AI reduce analysis effort without changing how users already work?
What should be automated, AI-assisted, or left entirely to the user?
How should the system distinguish between a broad match and a useful result?
How should important result be saved and organized so they remain part of the analysis workflow?
What happens to the workflow when processing is incomplete or delayed?
How can the experience communicate system state without making the workflow feel technical?
What needs to remain consistent as new capabilities are introduced across the workflow?
Thinking canvas
REJECTED
Redesign the whole platform.
Make AI the primary experience.
Treat search as end of review.
Solve everything in one release.
What
I Already
Knew
Relevant clauses are already extracted
Review continues outside DocSmart
AI capability is expected later
Repository continues growing
Questions?
that shaped every design
Rejected
Design Decision
What capabilities could reduce the most manual effort without adding another unnecessary steps?
What information turns extracted contract data into relevance inputs for analysis and evaluation?
How should metadata, extracted entities and relevance signals work together to narrow search results?
Should natural-language search extend the existing filter-based search, or introduce a new way to navigate the product?
Why do traditional filters become insufficient at enterprise scale?
How can AI reduce analysis effort without changing how users already work?
What should be automated, AI-assisted, or left entirely to the user?
How should the system distinguish between a broad match and a useful result?
How should important result be saved and organized so they remain part of the analysis workflow?
What happens to the workflow when processing is incomplete or delayed?
How can the experience communicate system state without making the workflow feel technical?
What needs to remain consistent as new capabilities are introduced across the workflow?
Thinking canvas
REJECTED
Redesign the whole platform.
Make AI the primary experience.
Treat search as end of review.
Solve everything in one release.
What
I Already
Knew
Relevant clauses are already extracted
Review continues outside DocSmart
AI capability is expected later
Repository continues growing
Questions?
that shaped every design
Rejected
Design Decision
What capabilities could reduce the most manual effort without adding another unnecessary steps?
What information turns extracted contract data into relevance inputs for analysis and evaluation?
How should metadata, extracted entities and relevance signals work together to narrow search results?
Should natural-language search extend the existing filter-based search, or introduce a new way to navigate the product?
Why do traditional filters become insufficient at enterprise scale?
How can AI reduce analysis effort without changing how users already work?
What should be automated, AI-assisted, or left entirely to the user?
How should the system distinguish between a broad match and a useful result?
How should important result be saved and organized so they remain part of the analysis workflow?
What happens to the workflow when processing is incomplete or delayed?
How can the experience communicate system state without making the workflow feel technical?
What needs to remain consistent as new capabilities are introduced across the workflow?
Thinking canvas
REJECTED
Redesign the whole platform.
Make AI the primary experience.
Treat search as end of review.
Solve everything in one release.
What
I Already
Knew
Relevant clauses are already extracted
Review continues outside DocSmart
AI capability is expected later
Repository continues growing
Questions?
that shaped every design
Rejected
Design Decision
What capabilities could reduce the most manual effort without adding another unnecessary steps?
What information turns extracted contract data into relevance inputs for analysis and evaluation?
How should metadata, extracted entities and relevance signals work together to narrow search results?
Should natural-language search extend the existing filter-based search, or introduce a new way to navigate the product?
Why do traditional filters become insufficient at enterprise scale?
How can AI reduce analysis effort without changing how users already work?
What should be automated, AI-assisted, or left entirely to the user?
How should the system distinguish between a broad match and a useful result?
How should important result be saved and organized so they remain part of the analysis workflow?
What happens to the workflow when processing is incomplete or delayed?
How can the experience communicate system state without making the workflow feel technical?
What needs to remain consistent as new capabilities are introduced across the workflow?
Thinking canvas
With these questions in mind, I stopped designing for individual features and started shaping the experience around the decisions users needed to make next.
Design Approach
From defining problem to designing the experience
This was a design led-project with heavy iteration and collaboration
This was a design led-project with heavy iteration and collaboration
01
Expert View turned processed contract results into a working review space—bringing extracted attributes, source clauses, reviewer context, and selected findings together for contract comparison and analysis.
Expert View
Review
Analyze
Collaborate
Docsmart - Phase 1
Phase:1
01
Expert View turned processed contract results into a working review space—bringing extracted attributes, source clauses, reviewer context, and selected findings together for contract comparison and analysis.
Expert View
Review
Analyze
Collaborate
Docsmart - Phase 1
Phase:1
01
Expert View turned processed contract results into a working review space—bringing extracted attributes, source clauses, reviewer context, and selected findings together for contract comparison and analysis.
Expert View
Review
Analyze
Collaborate
Docsmart - Phase 1
Phase:1
Swipe to view comparison
Experience shift
From finding clauses to working with them
The shift was from locating a relevant clause to analyzing the extracted finding in context.
The shift was from locating a relevant clause to analyzing the extracted finding in context.
The shift was from locating a relevant clause to analyzing the extracted finding in context.
Expert view
What Expert view Changed?
Extracts key contract attributes from relevant clauses across multiple contracts after document processing, reducing document-by-document analysis and helping analysts compare contractual terms when developing client recommendations.
Allows reviewers to add analytical context and classify relevant contract attributes missed during extraction, giving subsequent reviewers a more complete view of the findings when comparing contracts and developing client recommendations.
Shortlisting relevant findings into the Summary View made it easier for analysts and consultants to revisit key results, compare them across contracts, and share the analysis with stakeholders for client recommendations.
Shared result-level feedback gave analysts visibility into how others assessed the same finding, making it easier to validate extraction accuracy and align on the analysis.
Expert view
What Expert view Changed?
Extracts key contract attributes from relevant clauses across multiple contracts after document processing, reducing document-by-document analysis and helping analysts compare contractual terms when developing client recommendations.
Allows reviewers to add analytical context and classify relevant contract attributes missed during extraction, giving subsequent reviewers a more complete view of the findings when comparing contracts and developing client recommendations.
Shortlisting relevant findings into the Summary View made it easier for analysts and consultants to revisit key results, compare them across contracts, and share the analysis with stakeholders for client recommendations.
Shared result-level feedback gave analysts visibility into how others assessed the same finding, making it easier to validate extraction accuracy and align on the analysis.
Expert view
What Expert view Changed?
Extracts key contract attributes from relevant clauses across multiple contracts after document processing, reducing document-by-document analysis and helping analysts compare contractual terms when developing client recommendations.
Allows reviewers to add analytical context and classify relevant contract attributes missed during extraction, giving subsequent reviewers a more complete view of the findings when comparing contracts and developing client recommendations.
Shortlisting relevant findings into the Summary View made it easier for analysts and consultants to revisit key results, compare them across contracts, and share the analysis with stakeholders for client recommendations.
Shared result-level feedback gave analysts visibility into how others assessed the same finding, making it easier to validate extraction accuracy and align on the analysis.
Expert view
What Expert view Changed?
Extracts key contract attributes from relevant clauses across multiple contracts after document processing, reducing document-by-document analysis and helping analysts compare contractual terms when developing client recommendations.
Allows reviewers to add analytical context and classify relevant contract attributes missed during extraction, giving subsequent reviewers a more complete view of the findings when comparing contracts and developing client recommendations.
Shortlisting relevant findings into the Summary View made it easier for analysts and consultants to revisit key results, compare them across contracts, and share the analysis with stakeholders for client recommendations.
Shared result-level feedback gave analysts visibility into how others assessed the same finding, making it easier to validate extraction accuracy and align on the analysis.
The roadmap shifted - brief was simple: bring AI into Expert View. The opportunity was clear, but its role within the existing workflow was still unknown.
Question was how? & Where? …….> Phase 2
02
Integrated Ask AI into Expert View, enabling users to query processed contract documents, explore their contents, and extend analysis beyond the attributes identified through predefined extraction.
EXTENDING EXPERT VIEW WITH Ask AI
Ask
Explore
Analyze
Docsmart - Phase 2
Phase:2
02
Integrated Ask AI into Expert View, enabling users to query processed contract documents, explore their contents, and extend analysis beyond the attributes identified through predefined extraction.
EXTENDING EXPERT VIEW WITH Ask AI
Ask
Explore
Analyze
Docsmart - Phase 2
Phase:2
02
Integrated Ask AI into Expert View, enabling users to query processed contract documents, explore their contents, and extend analysis beyond the attributes identified through predefined extraction.
EXTENDING EXPERT VIEW WITH Ask AI
Ask
Explore
Analyze
Docsmart - Phase 2
Phase:2
02
Integrated Ask AI into Expert View, enabling users to query processed contract documents, explore their contents, and extend analysis beyond the attributes identified through predefined extraction.
EXTENDING EXPERT VIEW WITH Ask AI
Ask
Explore
Analyze
Docsmart - Phase 2
Phase:2
Experience shift
From an emerging capability to an integrated AI experience
Everything was aligned. It moved to development. But…..
Automatically initiating document processing at upload carried significant processing-cost implications.
Automatically initiating document processing at upload carried significant processing-cost implications.
Automatically initiating document processing at upload carried significant processing-cost implications.
Ask Ai
What Ask AI Changed?
Natural-language queries identified relevant clauses and contractual terms across multiple documents without requiring clause-level navigation or exact terminology.
Cross-contract comparison became query-driven, making it easier to identify which agreements contained a particular term, how contractual terms differed, and where exceptions occurred across the contract set.
Ad-hoc questions could extend analysis beyond predefined extraction attributes, allowing the contract content to be explored based on the needs of the review rather than the limits of the configured template.
Conversational follow-ups turned individual findings into deeper analysis—allowing users to ask, clarify, summarize, and understand contract content without first completing the full structured review process.
ASK AI
What Ask AI Changed?
Natural-language queries identified relevant clauses and contractual terms across multiple documents without requiring clause-level navigation or exact terminology.
Cross-contract comparison became query-driven, making it easier to identify which agreements contained a particular term, how contractual terms differed, and where exceptions occurred across the contract set.
Ad-hoc questions could extend analysis beyond predefined extraction attributes, allowing the contract content to be explored based on the needs of the review rather than the limits of the configured template.
Conversational follow-ups turned individual findings into deeper analysis—allowing users to ask, clarify, summarize, and understand contract content without first completing the full structured review process.
ASK AI
What Ask AI Changed?
Natural-language queries identified relevant clauses and contractual terms across multiple documents without requiring clause-level navigation or exact terminology.
Cross-contract comparison became query-driven, making it easier to identify which agreements contained a particular term, how contractual terms differed, and where exceptions occurred across the contract set.
Ad-hoc questions could extend analysis beyond predefined extraction attributes, allowing the contract content to be explored based on the needs of the review rather than the limits of the configured template.
Conversational follow-ups turned individual findings into deeper analysis—allowing users to ask, clarify, summarize, and understand contract content without first completing the full structured review process.
ASK AI
What Ask AI Changed?
Natural-language queries identified relevant clauses and contractual terms across multiple documents without requiring clause-level navigation or exact terminology.
Cross-contract comparison became query-driven, making it easier to identify which agreements contained a particular term, how contractual terms differed, and where exceptions occurred across the contract set.
Ad-hoc questions could extend analysis beyond predefined extraction attributes, allowing the contract content to be explored based on the needs of the review rather than the limits of the configured template.
Conversational follow-ups turned individual findings into deeper analysis—allowing users to ask, clarify, summarize, and understand contract content without first completing the full structured review process.
Once Ask AI was integrated within a binder, the next question was: why stop within the binder?
Expanding AI across binders …….> Phase 3
03
Search Across Binder evolved filter-based search into LLM-based content discovery across binders and documents, adding natural-language search, AI-assisted refinement, document shortlisting, and continued analysis in Expert View.
Search across binders
Binder Search
Content search
Standard filter
Result
Docsmart - Phase 3
Phase:3
03
Search Across Binder evolved filter-based search into LLM-based content discovery across binders and documents, adding natural-language search, AI-assisted refinement, document shortlisting, and continued analysis in Expert View.
Search across binders
Binder Search
Content search
Standard filter
Result
Docsmart - Phase 3
Phase:3
03
Search Across Binder evolved filter-based search into LLM-based content discovery across binders and documents, adding natural-language search, AI-assisted refinement, document shortlisting, and continued analysis in Expert View.
Search across binders
Binder Search
Content search
Standard filter
Result
Docsmart - Phase 3
Phase:3
03
Search Across Binder evolved filter-based search into LLM-based content discovery across binders and documents, adding natural-language search, AI-assisted refinement, document shortlisting, and continued analysis in Expert View.
Search across binders
Binder Search
Content search
Standard filter
Result
Docsmart - Phase 3
Phase:3
Experience shift
From asking within a document to searching across a binder
Everything was working smoothly until !……
Everything was working smooth until !……
User testing exposed a problem ~ content-based searches were taking too long to load.
User testing exposed a problem ~ content-based searches were taking too long to load.
SEARCH ACROSS
What Search Across Binder Changed?
Extracts key contract attributes from relevant clauses across multiple contracts after document processing, reducing document-by-document analysis and helping analysts compare contractual terms when developing client recommendations.
Extended search from contract metadata to contractual content, enabling analysts to identify relevant clauses, terms, conditions, and requirements across documents without reviewing agreements individually.
LLM-based query interpretation brought refinement into the search flow, identifying relevant contractual terms and metadata for broad content queries and narrowing large contract datasets before result retrieval—reducing unnecessary result volume and manual filtering for contract analysts.
Relevant search results could be verified through Quick View, shortlisted into a focused virtual binder, and carried into Expert View—turning cross-binder search into a structured path from contract discovery to detailed analysis.
SEARCH ACROSS
What Search Across Binder Changed?
Natural-language contract search replaced multi-step filter selection with direct query-based retrieval, enabling contract analysts and procurement specialists to locate relevant contracts by describing the required vendor, client, period, or contractual attribute directly.
Extended search from contract metadata to contractual content, enabling analysts to identify relevant clauses, terms, conditions, and requirements across documents without reviewing agreements individually.
LLM-based query interpretation brought refinement into the search flow, identifying relevant contractual terms and metadata for broad content queries and narrowing large contract datasets before result retrieval—reducing unnecessary result volume and manual filtering for contract analysts.
Relevant search results could be verified through Quick View, shortlisted into a focused virtual binder, and carried into Expert View—turning cross-binder search into a structured path from contract discovery to detailed analysis.
SEARCH ACROSS
What Search Across Binder Changed?
Natural-language contract search replaced multi-step filter selection with direct query-based retrieval, enabling contract analysts and procurement specialists to locate relevant contracts by describing the required vendor, client, period, or contractual attribute directly.
Extended search from contract metadata to contractual content, enabling analysts to identify relevant clauses, terms, conditions, and requirements across documents without reviewing agreements individually.
LLM-based query interpretation brought refinement into the search flow, identifying relevant contractual terms and metadata for broad content queries and narrowing large contract datasets before result retrieval—reducing unnecessary result volume and manual filtering for contract analysts.
Relevant search results could be verified through Quick View, shortlisted into a focused virtual binder, and carried into Expert View—turning cross-binder search into a structured path from contract discovery to detailed analysis.
SEARCH ACROSS
What Search Across Binder Changed?
Natural-language contract search replaced multi-step filter selection with direct query-based retrieval, enabling contract analysts and procurement specialists to locate relevant contracts by describing the required vendor, client, period, or contractual attribute directly.
Extended search from contract metadata to contractual content, enabling analysts to identify relevant clauses, terms, conditions, and requirements across documents without reviewing agreements individually.
LLM-based query interpretation brought refinement into the search flow, identifying relevant contractual terms and metadata for broad content queries and narrowing large contract datasets before result retrieval—reducing unnecessary result volume and manual filtering for contract analysts.
Relevant search results could be verified through Quick View, shortlisted into a focused virtual binder, and carried into Expert View—turning cross-binder search into a structured path from contract discovery to detailed analysis.
What I learned
This project taught me how valuable it is to first ~ pause, question assumptions and understand feasibility before making a design decision. When a technical constraint challenged the direction, it pushed me to think beyond the existing solution and explore new possibilities without losing sight of the product goal. It has made me a more analytical and pragmatic designer.
What I learned
This project taught me how valuable it is to first ~ pause, question assumptions and understand feasibility before making a design decision. When a technical constraint challenged the direction, it pushed me to think beyond the existing solution and explore new possibilities without losing sight of the product goal. It has made me a more analytical and pragmatic designer.
Key takeaway
