ProCura: Transforming Contract Management Through Intelligent Automation
A smart contract management solution that automates redaction, improves compliance, and accelerates procurement workflows.
Overview
Procura is an AI-assisted document redaction platform designed for complex, regulated procurement environments. I led the UX research and product design process, working with procurement specialists, contract managers and technical stakeholders to understand how AI could reduce the burden of reviewing extremely large documents without removing the human judgement required when handling sensitive information.
The resulting experience combines automated document analysis with reusable redaction profiles, confidence-based review, manual intervention and auditability.
The problem
The project originated from a wider investigation into how AI and automation could reduce increasing workload across procurement.
One opportunity stood out: contract redaction.
Large procurement documents can contain sensitive commercial, personal and legal information, while existing processes rely heavily on people finding and removing that information manually. Contracts could exceed 100 pages, with research also identifying requirements for documents exceeding 1,000 pages.
Automation appeared to offer an obvious solution, but it introduced a second problem.
If users still had to reread every page after the AI completed its work, much of the efficiency benefit disappeared. If they simply trusted the system, they risked sensitive information being incorrectly disclosed.
The problem therefore became less about automating redaction and more about designing the relationship between automation, uncertainty and human responsibility.
Understanding the real workflow
Before designing the redaction experience, I needed to understand where it existed within the wider procurement journey.
I facilitated collaborative workshops with stakeholders working across sourcing, contracts, procurement technology, policy and AI development. Rather than immediately presenting a solution, we mapped the journey from selecting a contract through strategy, drafting, assurance, negotiation, award and eventual publication.
The research uncovered a fragmented ecosystem.
Documents moved between repositories, Word and PDF software, procurement systems and email. Contracts could pass through multiple rounds of stakeholder endorsement and assurance, while users remained responsible for collecting comments, updating documents and resubmitting them for approval.
One stakeholder explained that the organisation did not yet have a clear established process or dedicated tool for the emerging redaction requirement. Significant document collaboration still happened through email despite enterprise systems existing elsewhere in the workflow.
This changed my understanding of the problem.
Procura could not be designed as an isolated AI button. It needed to work within a complex organisational system while avoiding becoming another sprawling enterprise platform.
What the research changed
The original concept was much broader than redaction.
Early thinking included metadata extraction, summarisation, risk identification, clause intelligence, contract guidance and integration with existing management systems. The initial product hypothesis explicitly explored an intelligent workflow spanning ingestion, AI processing, human review and eventual system integration.
Research validated some of those opportunities, but it also helped create boundaries.
For example, users experienced genuine difficulty choosing the appropriate contract type. Less experienced staff could select overly complex forms, reuse whatever had been used previously, or rely heavily on internal experts for guidance.
That represented a useful AI opportunity, but it did not necessarily belong inside the redaction workflow.
This became particularly clear when participants explained that redaction could also be used for other documents, including information requests and contract changes. Forcing every user through an end-to-end procurement journey would make the redaction product less useful.
The solution therefore became much more focused:
Procura would specialise in intelligent document redaction and review while integrating with the wider ecosystem rather than attempting to replace it.
This was one of the most important product decisions in the project.
The Procura workflow
Turning a complex task into four guided stages
Rather than presenting users with a large document editor and hundreds of controls, I organised the experience into four clear phases.
01
Upload
The user begins by adding one or multiple documents.
Procura validates the files, analyses their structure and attempts to identify the document type. Instead of applying that classification silently, the system presents its recommendation back to the user for confirmation.
This creates an early human checkpoint before the classification influences everything that follows.
Where appropriate, Procura also recommends a Redaction Profile, containing the relevant rules for that type of document.
02
Analyse
The AI begins processing the document.
One UX challenge was avoiding the typical ambiguous AI loading state. Instead of presenting a spinner labelled Analysing, Procura communicates the work taking place, such as reviewing document structure, checking metadata, identifying personal information and applying profile rules.
The interface can progressively surface the number of findings, confidence distribution and potential exceptions.
The user therefore sees the system working without being overwhelmed by technical model information.
03
Review
This became the core experience.
The document sits alongside a structured findings queue, allowing users to move directly between proposed redactions instead of manually searching hundreds of pages.
Each finding can communicate its category, location, confidence and reason for detection.
Users can approve a suggestion, reject it, change its category, edit the selection or create a completely new manual redaction.
Low-confidence results can be prioritised automatically so that human attention follows uncertainty.
04
Finalise
Before export, Procura provides a high-level summary of what happened.
The user can see how many redactions were applied, how many required manual intervention, whether unresolved findings remain and whether metadata or image warnings still require attention.
Only once the appropriate review is complete does the document move towards export or integration with the organisation's existing systems.
The four-stage model was grounded in the original workflow hypothesis of ingestion, automated processing, human review and finalisation.
Turning organisational knowledge into Redaction Profiles
AI should not have to rediscover the same rules every time
One of the strongest product concepts to emerge from the workshops was the idea of reusable Redaction Profiles.
I used the analogy of a recipe.
The document is the finished cake. The profile is the recipe describing how that type of document should be treated.
Instead of retaining every previous contract simply so the AI can imitate what happened, Procura can retain reusable organisational rules.
A profile might define the types of information normally removed, sensitive sections, known protected content, metadata requirements, exceptions and areas that should always require human review.
This creates several benefits.
It allows Procura to understand organisational context that generic redaction tools do not have.
It also makes the system adaptable.
If policy or document templates change, an authorised user can update the profile rather than waiting for an entirely new product or model. Profiles can be versioned so that an audit trail can show exactly which rules were applied when a document was processed.
Importantly, user corrections do not need to automatically retrain the organisation's rules. Repeated corrections could instead become suggested profile improvements, reviewed by an authorised administrator before being adopted.
This balances learning with governance.
Designing for the exceptions
AI UX becomes most important when the AI is wrong
The workshops repeatedly exposed situations that a happy-path prototype could easily overlook.
Sensitive information might appear inside document metadata.
It might appear within an image rather than text.
A technical drawing might require only a particular area to be hidden.
An AI classification might be incorrect.
A document could contain content that does not match any existing profile.
These edge cases made manual control essential.
Participants explicitly requested the ability to manually redact documents, amend AI-generated redactions and undo changes. Image content was particularly important because a user might need to hide only part of a drawing or embedded image rather than removing it entirely.
Auditability also emerged as a fundamental requirement.
Users wanted the ability to understand what had been changed, why it had been changed and by whom. The broader solution concept therefore included audit trails alongside user approval and override capabilities.
The solution
From manual redaction to supervised automation
The resulting Procura concept brings automated analysis, confidence-based review, reusable profiles, manual intervention and auditability into one guided experience.
Instead of manually searching an entire document for everything that might be sensitive, users supervise a system that performs the repetitive analysis and directs their attention towards the decisions that require expertise.
The product concept also respects the wider enterprise ecosystem. It prepares and validates documents for downstream systems rather than attempting to replace existing contract-management infrastructure.
The project remained at research, design and proof-of-concept stage, so I would not claim measured production efficiencies or deployment outcomes.
The value of the work was demonstrating a credible interaction model for introducing AI into a high-risk workflow.
Reflection
Procura changed how I think about automation.
Traditional UX often aims to remove friction. In high-risk AI systems, some friction is valuable.
Confirming a document classification, reviewing an uncertain redaction or acknowledging an unresolved warning adds an extra interaction, but those moments protect against far more serious mistakes.
The most important question was therefore not how much of the workflow AI could automate.
It was deciding where automation should stop.
The final concept uses AI where machines are strongest, such as searching, pattern recognition and processing large volumes of information, while preserving human judgement for ambiguous and consequential decisions.
For me, that became the defining idea behind Procura:
AI should reduce the work required to make a decision, not remove the person responsible for making it.
Ready to transform your digital experience?
From initial concept to final implementation, we provide comprehensive UX services that transform your ideas into user-centred digital experiences.