Case study · 2026
Designing AI to support complex procurement decisions
I led the UX/UI design of an AI-assisted bid evaluation experience that helps procurement teams analyse large volumes of supplier information, compare bids and interrogate complex submissions through natural-language queries.
Role
Lead Product Designer
Team
Product · Engineering · QA
Stage
Prototype → Development QA
Focus
AI Product Design · Decision Support · Trust · Complex Data
I owned: problem framing · interaction model · UX/UI · prototyping · validation · design QA
The PROBLEM
Procurement teams needed to review supplier submissions containing pricing, exclusions, validity dates, technical responses, method statements and supporting documents.
The information required to make a decision existed across multiple documents and bidders, making comparison time-consuming and increasing the effort required to identify important differences.
The opportunity for AI wasn’t simply to add a chatbot. It was to help users surface relevant information while keeping the underlying evidence and decision-making process understandable.
AI-assisted decision support
Procurement · Enterprise SaaS · Data-heavy workflows · AI
Before defining the assistant, I mapped the existing evaluation workflow and worked with Product and domain stakeholders to identify where evaluators spent effort finding, comparing and verifying information. This helped separate problems suited to AI assistance from decisions that needed to remain with the evaluator.
PRODUCT STRATEGY
I worked with Product, Engineering and stakeholders to understand the needs, constraints and dependencies across buyer and supplier workflows. Rather than treating each side independently, I mapped how catalogue configuration, pricing, fulfilment and returns affected what users experienced on the opposite side.
This helped identify where the marketplace needed shared rules and where buyer and supplier experiences needed to remain deliberately different.
Summarise
Reduce the effort required to review long supplier documents by extracting key information such as pricing, exclusions, validity dates and method statements.
Compare
Help evaluators identify differences between bidders, including pricing, technical scores, document completeness and item-level values.
Interrogate
Allow users to ask specific questions about a bidder or across the full bid set without manually navigating between documents.
This framed the assistant as a decision-support tool rather than an autonomous evaluator.
USER NEEDS
Price
Who is the cheapest bidder?
Who is the most expensive?
What is bidder 7’s total price?
Compare the price of this item across suppliers.
Quality
Who has the highest technical score?
Which bidder offers the strongest price-quality balance?
Completeness
Which bidders are missing BOQ information?
Which submissions have incomplete documents?
What exclusions has this supplier listed?
What is the validity period of the bid?
Capacity / risk
Cross-bid analysis
INTERACTION MODEL
I designed the AI Assistant as a contextual panel within the bid evaluation experience. Users could continue viewing bidder information while asking questions, reviewing summaries and comparing results.
Bid evaluation
→ Ask question
→ AI response
→ Supporting evidence
→ Continue evaluation
Keeping the assistant within the evaluation context reduced the need to move between separate tools and helped users maintain sight of the underlying bid information.
DISCOVERABILITY
Procurement users may understand the evaluation problem without immediately knowing how to phrase an AI query. I introduced suggested questions to demonstrate the assistant’s capabilities and provide low-effort entry points into common evaluation tasks.
INFORMATION ARCHITECTURE
Document summary
Surface the important information inside an individual submission, such as:
totals
exclusions
validity dates
method statements
missing information
Consolidated bidder summary
Separating these levels helped prevent dense AI output from becoming another source of complexity.
TRUST & CONTROL
Some evaluation questions naturally moved beyond retrieval into judgement — for example comparing price and quality, identifying the highest technical score or asking which submission represented better value.
I designed responses to make the reasoning visible by surfacing the relevant values and comparison criteria, rather than presenting an unexplained answer.
AI UX PRINCIPLES
Context
Responses should make clear which bidder, documents or comparison set they refer to.
Evidence
Important conclusions should point users back to supporting information.
Control
Users remain responsible for the evaluation and can continue investigating the underlying submission.
Consistency
AI interactions should use the same design-system patterns and hierarchy as the wider product.
PROTOTYPING
I used scenarios with large bidder sets to explore how the assistant behaved when users moved between individual supplier questions and cross-bid comparisons.
25 Bidders
Bidder set
→ individual supplier query
→ cross-bid comparison
→ missing information
→ return to evaluation
DESIGN QA & DELIVERY
The work continued after the prototype.
Layout
Panel height and header hierarchy
Components
Query chips, textarea, button radius
Content
AI Assistant hierarchy and suggested-query treatment
Consistency
Alignment with the wider design system
WHAT I LEARNED
Useful before impressive
AI needed to reduce real evaluation effort rather than simply demonstrate capability.
Evidence before confidence
Users needed to understand why an answer was being surfaced.
Context before conversation
The strongest interactions happened when AI understood where users were in the evaluation workflow.
The project shifted my focus from designing a chatbot to designing an AI-supported decision system.
DESIGN OUTCOME
A more focused way to navigate complex bid information.
The design established a clear interaction model for introducing AI into bid evaluation — combining document summarisation, cross-bid comparison and conversational querying within the existing procurement workflow.
Faster access
Key information surfaced without manually reviewing every document.
Comparable
Bidder information structured for easier cross-supplier analysis.
Reviewable
AI outputs designed to remain connected to the evidence behind them.
Reflection
This project reinforced that AI product design is as much about boundaries as capability. The strongest experience was not the one where AI did the most, but the one where it reduced effort while preserving context, evidence and user control.