From Manual Grind to AI Magic
How We Made Contract Data Extraction Faster, Smarter, and Scalable
Contracts are at the heart of procurement but managing them was tedious, inconsistent, and prone to error. We set out to design an AI-powered extraction experience that transforms how users add and view contracts, reducing effort while improving data quality and speed.
Imagine if you had to manually read through lengthy contracts and input data into the system, a time-consuming, error-prone process. Some of our customers experimented with ChatGPT, but without validation or clear context, trust in AI suggestions was low. Even reviewing existing contracts for meetings meant digging through PDFs just to find key terms like payment schedules or termination clauses.
As one procurement manager put it:
“It took like two full days of just reading contracts and putting things in — lots of coffee.”
Slow, boring onboarding
Adding contract data into the system was a manual chore, often requiring users to refer back-and-forth between documents and fields.
Unstructured inputs, unclear confidence
Customers tried using ChatGPT to extract key details but didn’t trust the AI’s accuracy or know where the data came from.
Missing context during review
Reviewing older contracts meant opening PDFs and manually hunting for details, a time-consuming task especially before meetings.
Our solution streamlined workflows, gave confidence in AI suggestions, and helped teams reduce manual work from 2 days to just a couple hours.
Smart extraction meets clarity
AI suggestions are now presented with source traceability and clear accept/reject flows.
One interface, two powerful use cases
Unified design supports both: adding new contracts and reviewing old ones.
Scalable and future-proof
Designed to grow with more fields, clauses, and automation capabilities.
With AI handling extraction and review, teams can now shift their time from manual work to something more strategic. What once took days now takes hours, boosting efficiency, building trust in the data, and empowering teams to prepare smarter for negotiations and supplier meetings.
Procurement Lead
Main goals:
Access key terms quickly before meetings
Spot negotiation levers like payment terms
Ensure data quality on contracts
Contract manager
Main goals:
Save time on manual contract entry
Access key terms quickly before meetings
Maintain complete, accurate contract records to ensure good data
Design trust, not just automation
Users won’t accept AI suggestions blindly. Explain, trace, and empower them to decide.
Be biased toward simplicity first
Start with the most common document; add edge-case complexity only after user insights.
Don’t separate UX for create and view
Designing for both creation and viewing modes together gave us a coherent, scalable pattern.
While designing for trust and clarity was essential, we also had to ensure technical accuracy. Without it, even the best UX would fall apart. Early in the project, developers validated AI performance across different contract formats. We decided to start small, thus, showing only basic and high-confidence fields, while placing experimental fields in a separate section. This gave users clarity, while also letting us gather valuable feedback without mixing reliable and untested outputs.
The next step is bringing bulk AI upload into the product, so users can onboard multiple contracts at once instead of handling them one by one. This feature is already in progress. We are also excited to hear feedback on experimental fields once users start relying on them more frequently helping us balance accuracy, trust, and usability as the product matures.
Closing Thoughts
This project reminded me that trust is a design problem, especially when users rely on AI for critical, high-stakes work.
UX METHODS
User journey mapping
Affinity mapping
User interviews
Usability testing
Desk research
Tools
Figma
FigJam
Notion
Notably on iPad
