Reducing Hours of Manual Work to Minutes
Simplifying estimate comparison by designing a document extraction and comparison tool, eliminating 3-4 hours of manual effort and delivering automated comparisons in minutes (~90% faster), and enabling 10x+ scale.
ABOUT DATAX.AI
dataX.ai is an AI-first company that automates the process of extracting, Organising, and matching product data from semi-structured and unstructured sources.
PROJECT OVERVIEW
Delta automates estimate comparison across documents, cutting down manual effort and enabling faster, more efficient decision-making to reach conclusions faster.
IMPACT
MY ROLE
Handled the end to end design process, from initial concept to high-fidelity designs, over a span of 2 months.
CONTEXT
This started when a construction company approached us at a conference, looking for a better way to handle their workflow.
In the housing industry, contractors create detailed estimates for repair or replacement work. These documents can easily run over 100 pages. Insurance companies then generate their own estimates for the same work.
Each line item includes details like quantity, unit price, depreciation, and actual cash value (ACV). To find differences, these estimates are usually compared manually.
This process takes hours, is prone to errors, and slows things down, especially when multiple documents are involved.
To solve this, we built a tool that automates the comparison and makes the process much faster and easier to manage.
This tool is designed with focus on the USA and Canada Market.
RESEARCH
Through deeper conversations with clients and users, the problem became much clearer.
For every new project, contractors first prepare a comparison estimate. This alone takes around 3–4 hours, including marking differences, comparing documents, and creating a final draft. This is then sent to the insurance company, which responds with its own analysis.
On average, it takes 2-3 back and forth cycles to reach an agreement, and each cycle can stretch over weeks.
The core users are construction managers who handle multiple clients at once, along with all estimates and documentation. Many of them are not very comfortable with digital tools and tend to avoid anything that feels complex or time consuming to learn.
Even then, comparing estimates for just one client takes up a significant amount of their time. With multiple projects running in parallel, this quickly becomes overwhelming.
Because of this, they rely heavily on manual processes. When timelines get tight, they often agree to reduced scopes just to keep work moving directly impacting their margins.
These insights highlighted the gaps and opportunities in the current flow. From this, we shaped a set of goals to guide the solution forward.
GOALS
USER FLOW
After understanding the users' pain points and behaviour, I set out to design a user flow that was fast, and frictionless, minimising steps and keeping the process straightforward. Here’s how the flow was structured
ONBOARDING
The onboarding process was designed to be quick and straightforward. Construction companies subscribed to the platform and were provided with a dashboard that included an admin account. Admins could then invite contractors via email.
Contractors completed a simple signup by entering their name and setting a password, allowing them to get started without friction.
To enhance the experience beyond a standard onboarding flow, company-specific branding was incorporated. Each organisation saw a customised interface from the very first interaction, creating a sense of familiarity and ownership.
Each user has an individual account, and any data they upload remains specific to them. This ensures that information is not shared across other users.
A key insight was that many users had limited familiarity with digital tools. To support this, the interface was designed to be
The primary action was made highly prominent, with minimal surrounding elements, so users could immediately understand what to do next without confusion. This helped reduce friction and made the experience more accessible for first-time users.
CREATING BATCHES
Once the user accesses the upload screen, they’re presented with four simple input fields:
Batch Name: An identifier provided by the user to label the comparison project for future reference.
Files: Users are required to upload both the contractor’s estimate and the insurance estimate. To streamline the process, we enabled both manual upload and the option to import files directly from their drive
Description (Optional): While not mandatory, this field can be very helpful for adding context, such as claim numbers, property details, or any other notes.
HOME SCREEN
As soon as the files are uploaded, processing begins instantly, and the project is listed in a structured table view. The table includes the following columns:
Batch Name: The identifier given by the user during upload.
Batch Status: Indicates the current state of the project
Uploaded On & Description: Displays the upload date along with any optional description provided for added context.
Input Files: These are the documents uploaded by users for comparison. These files remain accessible and can be downloaded at any time when needed.
Actions: Allows users to download the output file and share the output by entering receivers email (mostly to themselves)
EXTRACTION PROCESS
Before diving into the comparison screens, let me walk you through the contents of the uploaded documents, what exactly gets extracted and what the user is comparing.
Whenever an estimate for the repair or replacement of a damaged property is created, it is typically organised into the following structure:
Category: Refers to specific areas of the property, such as Master Bedroom, Kitchen, etc.
Description: Details the specific repair or replacement tasks within each category.
Attributes: These usually include key metrics like Quantity, Unit Price, Tax, Overhead & Profit (O&P), Replacement Cost Value (RCV), Depreciation, and Actual Cash Value (ACV).
Insurance companies follow a very similar structure in their estimates, making this format consistent across both parties and ideal for direct comparison.
DOCUMENT COMPARISION
For the Comparison results, I didn’t just want to show a side-by-side number comparison.
I aimed to add visual cues and strong hierarchy to help users quickly and easily make decisions. This involved multiple iterations.
Once the the data is extracted it is categorised into 5 types -
All , Exact Match, Matches with Discrepancies, Not In Insurance Estimate, Not in Contractor Estimate
COMMENTS AND FILTERS
Users can filter data based on predefined categories available within the columns.
Additionally, filters can be applied based on comparison status (available on the “All” screen), including:
Matching
Not Matching
Overestimated (Contractor > Insurance)
Underestimated (Contractor < Insurance)
Not Available (N/A)
This allows users to quickly narrow down and focus on specific discrepancies or patterns in the data.
Users can add comments to each description (row) for their own reference.
IMPACT & RESULTS
Thank you for taking the time to read this case study! I hope it offered valuable insights and a glimpse into my design process.
Reducing Hours of Manual Work to Minutes
Simplifying estimate comparison by designing a document extraction and comparison tool, eliminating 3-4 hours of manual effort and delivering automated comparisons in minutes (~90% faster), and enabling 10x+ scale.
ABOUT DATAX.AI
dataX.ai is an AI-first company that automates the process of extracting, Organising, and matching product data from semi-structured and unstructured sources.
PROJECT OVERVIEW
Delta automates estimate comparison across documents, cutting down manual effort and enabling faster, more efficient decision-making to reach conclusions faster.
IMPACT
MY ROLE
Handled the end to end design process, from initial concept to high-fidelity designs, over a span of 2 months.
CONTEXT
This started when a construction company approached us at a conference, looking for a better way to handle their workflow.
In the housing industry, contractors create detailed estimates for repair or replacement work. These documents can easily run over 100 pages. Insurance companies then generate their own estimates for the same work.
Each line item includes details like quantity, unit price, depreciation, and actual cash value (ACV). To find differences, these estimates are usually compared manually.
This process takes hours, is prone to errors, and slows things down, especially when multiple documents are involved.
To solve this, we built a tool that automates the comparison and makes the process much faster and easier to manage.
This tool is designed with focus on the USA and Canada Market.
RESEARCH
Through deeper conversations with clients and users, the problem became much clearer.
For every new project, contractors first prepare a comparison estimate. This alone takes around 3–4 hours, including marking differences, comparing documents, and creating a final draft. This is then sent to the insurance company, which responds with its own analysis.
On average, it takes 2-3 back and forth cycles to reach an agreement, and each cycle can stretch over weeks.
The core users are construction managers who handle multiple clients at once, along with all estimates and documentation. Many of them are not very comfortable with digital tools and tend to avoid anything that feels complex or time consuming to learn.
Even then, comparing estimates for just one client takes up a significant amount of their time. With multiple projects running in parallel, this quickly becomes overwhelming.
Because of this, they rely heavily on manual processes. When timelines get tight, they often agree to reduced scopes just to keep work moving directly impacting their margins.
These insights highlighted the gaps and opportunities in the current flow. From this, we shaped a set of goals to guide the solution forward.
GOALS
IMPACT & RESULTS
Thank you for taking the time to read this case study! I hope it offered valuable insights and a glimpse into my design process.
USER FLOW
After understanding the users' pain points and behaviour, I set out to design a user flow that was fast, and frictionless, minimising steps and keeping the process straightforward. Here’s how the flow was structured
ONBOARDING
The onboarding process was designed to be quick and straightforward. Construction companies subscribed to the platform and were provided with a dashboard that included an admin account. Admins could then invite contractors via email.
Contractors completed a simple signup by entering their name and setting a password, allowing them to get started without friction.
To enhance the experience beyond a standard onboarding flow, company-specific branding was incorporated. Each organisation saw a customised interface from the very first interaction, creating a sense of familiarity and ownership.
Each user has an individual account, and any data they upload remains specific to them. This ensures that information is not shared across other users.
A key insight was that many users had limited familiarity with digital tools. To support this, the interface was designed to be
The primary action was made highly prominent, with minimal surrounding elements, so users could immediately understand what to do next without confusion. This helped reduce friction and made the experience more accessible for first-time users.
CREATING BATCHES
Once the user accesses the upload screen, they’re presented with four simple input fields:
Batch Name: An identifier provided by the user to label the comparison project for future reference.
Files: Users are required to upload both the contractor’s estimate and the insurance estimate. To streamline the process, we enabled both manual upload and the option to import files directly from their drive
Description (Optional): While not mandatory, this field can be very helpful for adding context, such as claim numbers, property details, or any other notes.
HOME SCREEN
As soon as the files are uploaded, processing begins instantly, and the project is listed in a structured table view. The table includes the following columns:
Batch Name: The identifier given by the user during upload.
Batch Status: Indicates the current state of the project
Uploaded On & Description: Displays the upload date along with any optional description provided for added context.
Input Files: These are the documents uploaded by users for comparison. These files remain accessible and can be downloaded at any time when needed.
Actions: Allows users to download the output file and share the output by entering receivers email (mostly to themselves)
EXTRACTION PROCESS
Before diving into the comparison screens, let me walk you through the contents of the uploaded documents, what exactly gets extracted and what the user is comparing.
Whenever an estimate for the repair or replacement of a damaged property is created, it is typically organised into the following structure:
Category: Refers to specific areas of the property, such as Master Bedroom, Kitchen, etc.
Description: Details the specific repair or replacement tasks within each category.
Attributes: These usually include key metrics like Quantity, Unit Price, Tax, Overhead & Profit (O&P), Replacement Cost Value (RCV), Depreciation, and Actual Cash Value (ACV).
Insurance companies follow a very similar structure in their estimates, making this format consistent across both parties and ideal for direct comparison.
DOCUMENT COMPARISION
For the Comparison results, I didn’t just want to show a side-by-side number comparison.
I aimed to add visual cues and strong hierarchy to help users quickly and easily make decisions. This involved multiple iterations.
Once the the data is extracted it is categorised into 5 types -
All , Exact Match, Matches with Discrepancies, Not In Insurance Estimate, Not in Contractor Estimate
COMMENTS AND FILTERS
Users can filter data based on predefined categories available within the columns.
Additionally, filters can be applied based on comparison status (available on the “All” screen), including:
Matching
Not Matching
Overestimated (Contractor > Insurance)
Underestimated (Contractor < Insurance)
Not Available (N/A)
This allows users to quickly narrow down and focus on specific discrepancies or patterns in the data.
Users can add comments to each description (row) for their own reference.