Project Result
Reduction in time to close offers
AI recommendation acceptance rate
SUS Score
Problem Statement
Contract negotiations relied on memory and judgement, producing inconsistent terms and poorly documented decisions.
Under time pressure, negotiators conceded liability and ESG protections or accepted discounts without checking comparable deals.
The goal: an AI co-pilot that uses past deals to recommend a defensible position, explains its reasoning and hands control back when needed.
Requirements Gathering
Discovery with procurement, legal and category managers identified three requirements:
REQUIREMENT
Visible reasoning
Show comparable deals and assumptions so negotiators can assess the recommendation and Legal can sign off.
REQUIREMENT
Clear Lines of Authority
Define what AI can agree to and when it must pause for human review.
REQUIREMENT
Auditability
Record the evidence, assumptions and rationale behind each position for audit and training.
Research: start with the stalled deal
I spoke with procurement and negotiation leads, Legal and Finance, and category and operations stakeholders. Rather than asking which AI features they wanted, I asked them to walk through their last stalled deal: who needed convincing, what evidence was missing, what they needed before signing and where AI should never act alone. The recurring need was a shared decision record that connected commercial terms, delivery, liability and ESG to the people authorised to make the call.
Using AI to accelerate discovery
Lovable turned early ideas into working prototypes in days, giving stakeholders something concrete to challenge.Maze tested whether users understood the AI’s task and sources, could review and amend its output, and saved time without losing confidence or control.The focus was trust: recommendations had to be understandable, controllable and accountable.
Tools Used
We mapped mandates and escalation, prototyped negotiation paths and examined how users accessed evidence. AI accelerated exploration; people judged the results.
Miro
Stakeholder collaboration, early concepts and user flows.
Easy to collaborate
Non-visual
Perfect for user-flow diagrams
Figma
Reusable components, interactive prototypes and developer handoff.
Component-based architecture
Design and Prototyping
Easy developer handoffs
Google Analytics
Track user behaviour and performance to inform design decisions.
Comprehensive user tracking
Bespoke trend tracking
Integration with Power BI
Claude AI
Explore concepts, refine interactions and accelerate content design.
UX/UI Ideation and exploration
Design feedback and refinement
Faster content and interaction design
Shareable decision records
PROCUREMENT + LEGAL
Capture terms, clauses, alternatives and approval rationale in one shareable record so Legal, Finance and ESG can review the same evidence.
Explainable recommendations
PROCUREMENT LEAD
Show evidence, assumptions, uncertainty and what would change the recommendation. Make countering and escalating as clear as approving.
Comparable contract context
CATEGORY MANAGER
Show supplier, category, scope, value and timing so category managers can judge whether a precedent is relevant.
Historical deal benchmarks
COMMERCIAL ANALYST
Include comparison periods, sample sizes and deal conditions. Expose uncertainty when benchmarks are sparse or outdated.
Decisions under time pressure
SENIOR BUYER
Keep deadlines, trade-offs and mandate limits together, with clear actions to approve, counter or escalate.
Clause-level evidence
LEGAL REVIEWER
Break risk down by liability, delivery, warranty, discount and ESG. Link each contribution to a clause or precedent for Legal to inspect.
01 / A discount can hide a worse deal
INTERVIEW LEARNING / COMMERCIAL TRADE-OFFS
Interview learning: a 5% saving can be consumed by late delivery, weaker liability protection and lost ESG reporting. Design response: show the net consequences of every option together, with ranges and assumptions beside the recommendation. The figures above illustrate the trade-off; they are not measured project results.
02 / Six perspectives, one decision
INTERVIEW LEARNING / STAKEHOLDER ALIGNMENT
Interview learning: negotiation touched Procurement, Legal, Finance, Operations, ESG and the supplier, while one negotiator had to move it forward. Design response: answer those stakeholders’ questions in a single workspace and retain the terms, evidence and rationale as one shareable decision record.
03 / Four hours to make a defensible call
INTERVIEW LEARNING / DECISIONS UNDER PRESSURE
Interview learning: buyers could have only four hours to move a negotiation forward, with time spent consulting other teams. Design response: support a ten-second first read, keep the deal state stable when users return, expose assumptions and missing evidence, and record both the decision and its rationale.
Personas
Sarah Chen is a composite persona built from conversations with procurement, legal, finance, category and operations stakeholders. Her job is to move a stalled negotiation forward without trading away delivery, liability or ESG protections. A four-hour decision window made the challenge concrete: she needed comparable terms, visible assumptions and a clear route to approve, counter or escalate. Legal and Finance needed the same evidence and a record of why each concession was made. That synthesis changed the brief from “ask AI a question” to “make a defensible decision within your mandate”.
Logic Map
I mapped what happens when AI reaches its mandate: show the breached boundary, supporting evidence and options to approve, counter or escalate.
The flow also records who decided, what evidence they used and why, then adds the outcome to Contract Memory.
Wireframes and iteration
I used wireframes to test the decision model before refining the interface. Customer interviews established the constraints; qualitative feedback showed where comparison, evidence and control still broke down. The iterations below connect those findings to the direction we chose.
Iteration 01 / Conversation or decision workspace?
CUSTOMER FEEDBACK / COMPARISON AND CONTROL
Interviews showed that buyers had to balance savings, delivery, liability and ESG within a short decision window. I tested a conversational thread against a prepared decision workspace. Feedback exposed the thread’s memory burden: users had to hold one option in mind while reading the next, repeat questions about ESG and scroll back to find actions. The workspace gave the comparison a stable home and made approve, counter and escalate easier to find.
Iteration 02 / Why we chose this direction
DESIGN RESPONSE / EVIDENCE, AUTHORITY AND ACTION
The early workspace still put charts before the reasoning and clipped the third option. A revised version improved explanation but left the evidence cramped. The chosen direction kept all three negotiation paths visible above the analysis, with consistent terms, consequences and risk against the authority limit. Customer needs drove the detail: a stable deal inbox supported interruptions; option-specific evidence made recommendations defensible; and a fixed evidence set with annotations avoided hiding findings that challenged the preferred option. The rationale and final action stayed on the deal record.
Designing the Interface
The interface combines comparable terms, visible uncertainty and supporting evidence. Sarah can approve, counter or escalate, with her rationale retained on the deal record.
UI DESIGN
Comparing negotiation paths
Side-by-side options help users assess alternatives instead of over-trusting the AI’s first answer.
Identical cards and field order give each path equal weight. Risk appears last to support the final comparison.
UI DESIGN
AI Suggestion and analysis
Ranges on a common scale expose uncertainty and show whether one option clearly outperforms another.
The chart shows the trade-off; the explanation states why the AI recommends a path.
UI DESIGN
Compare your deals
Users need evidence to verify an AI recommendation.
An always-visible comparable-deals panel keeps evidence beside the conclusion and reduces context switching.
Behavioural analysis: evidence at the decision
Heatmaps show interaction around negotiation choices, supporting evidence and final actions. They reinforce the need to keep terms, rationale and consequences together. Heatmaps show where users interact; feedback helps explain why.
MEASURE
System Usability Score
The System Usability Scale (SUS) is a quick and reliable tool for measuring the usability of a system or product through a 10-item questionnaire that provides a single score reflecting users' overall satisfaction and ease of use.
MEASURE
SUS Distribution
Reviewing the decision
DECISION REVIEW HEATMAP
The review dialog has dense click clusters around the option selectors and bottom confirmation controls, with lighter activity across the terms. The concentration is strongest where users choose and confirm a negotiation path.
Inspecting the AI’s reasoning
RECOMMENDATION HEATMAP
Clicks are spread across the recommendation text, but concentrate most heavily on the bottom action controls. The explanation receives interaction alongside the controls for responding to it.
Comparing negotiation paths
WORKSPACE HEATMAP
In the main workspace, clicks cluster on the deal list, negotiation-path selectors and final action buttons. The strongest hotspots mark deal selection and decision-making, with lighter activity across the analysis.
Checking comparable deals
EVIDENCE PANEL HEATMAP
The open comparable-deals panel shows click clusters across its benchmark cards and details. Strong hotspots remain on the final action buttons, showing interaction with both the supporting evidence and decision controls.
Across the four heatmaps, click concentration shifts from selecting a deal and negotiation path to reviewing recommendations and comparable evidence. Final action controls attract the strongest clusters. These patterns show where interaction happens; they do not establish whether users understood the content.
“Easy to manoeuvre and clear overviews”
"Looks great."
Go-to-market: trust before scale
Position the boundary clearly, validate a controlled pilot and support the first live negotiation.
01 / POSITION
A pause is control
Reframe escalation as AI Control Mode: an agreed authority boundary has been reached. Sell evidence-led negotiation with human control, rather than an AI that claims to decide everything.
02 / VALIDATE
Pilot before expansion
Start with roughly 10% of live negotiations and compare against the existing process. Agree gates for closing speed, overrides, usability and review workload before expanding availability.
03 / ENABLE
Make the first deal work
Proposed enablement: guide the first live deal, confirm authority limits and share its record with Legal and Finance. Review activation, repeat use, overrides and escalations weekly; pause expansion if quality suffers.
Outcomes, learning and next steps
The results were encouraging. The next test is whether faster decisions hold up over time.
01 / REPORTED RESULTS
Speed and confidence
31% faster offer closing. 82% recommendation acceptance. SUS 71. These signals support usability and speed; acceptance alone does not prove a better commercial outcome.
02 / DESIGN LEARNING
Test the decision model first
A paper comparison could have exposed the conversation’s memory burden sooner. Test whether people can compare, inspect evidence and act within their mandate before investing in visual polish.
03 / NEXT EVALUATION
Follow decision quality
Track repeat use, overrides, disputes, review workload and realised commercial outcomes. Feed the learning into Contract Memory and explore a portfolio view without hiding individual authority limits.
