Case 01
UX Case Study
AI‑Powered Emissions Estimation Integration via Climatiq in SiGREEN
Personalized, component-aware suggestions (unit, region, material) with confidence and one-tap overrides—so teams can add secondary data when suppliers can’t.
Project Overview
"I spend more time hunting for emission factors than making real decisions." That was the voice of one procurement manager during a workflow review and it stuck with us.
In SiGREEN, when suppliers don’t provide a Product Carbon Footprint (PCF), procurement managers turn to secondary data. But entering this data was entirely manual. They had to juggle multiple tabs, copy-paste values from databases, and match descriptions by instinct a time-consuming and error-prone task.
That’s where I saw an opportunity: what if an AI assistant could surface the most relevant emission factor suggestions with confidence scores directly inside the form?
Project
SiGREEN
My Role / Team
Me - Product Designer
1 PO
1 Tester
2 Front End Devs
1 Backend Architect
Platform
Web App
Timeline
January 2024- March 2024
The Challenge
Manually managing secondary data introduced several friction points:
Procurement managers had to interpret technical data from multiple sources
It was easy to pick mismatched factors, leading to inaccuracies
Time per entry ballooned when dealing with large BOMs
This not only slowed decision-making but impacted trust in the system.
Vision and Proposed Solution
To streamline the workflow, SiGREEN integrated Climatiq Autopilot, an AI-driven emissions factor mapping engine.
This integration allows procurement users to automatically match component descriptions to the most relevant emission factors from Climatiq’s verified database eliminating manual lookup and mapping.
Dynamic search by component, location, unit, etc.
AI-powered matching of emission factors with confidence scoring
Centralized, high-quality emissions data from verified sources
Solution
We defined what success would look like:
Procurement managers trust the AI output, but feel in control
Time per entry is drastically reduced
Matching quality is transparent and justifiable
UX Process
Understanding the Terrain
I started with a landscape study of Climatiq’s APIs and data structure.
Our goal: match their strengths to real-world user goals.
What metadata do users rely on most?
Can suggestions adapt based on component type?
What fallback options should exist if no good match is found?
Overview of Climatiq
Climatiq is a sustainability tech platform designed to help businesses calculate, report, and reduce carbon emissions. Its solution is API-first and caters to developers, sustainability officers, and business decision-makers aiming to embed carbon intelligence into digital products, operational systems, and workflows.
Core Capabilities of Climatiq
Secondary research
Alignment Workshops
Workshops helped us converge faster:
Product Owners aligned on pricing plans and rollout roadmap
Backend Architects discussed how to retrieve and rank AI suggestions
Frontend Developers scoped loading states and fallback UI
Initial user Journey discussion
Stakeholders - UX,PO,Architect and Dev lead
Asking questions to Stakeholders
Understanding with Q&As
Deep Diving into Climatiq feature
Understanding the Climatiq offering and challenges
There are 3 APIs which are provided by Climatiq, but we were focusing on 2 APIs
Transforming Customer Needs into Revenue Opportunities
Identifying the importance of the Ecoinvent database for our customers was a turning point in the project.
I proposed splitting our offering into two plans, which allowed us to elevate Ecoinvent into a premium, payable feature within SiGREEN. This strategic move not only enhanced the product’s value perception for customers but also created a new revenue stream, strengthening the business case for sustainability-driven offerings.
Deep Diving the flow for procurement manager
Understanding the working of the flow of filling out secondary data flow
Brainstorming & Ideation
To avoid scope creep, we used an impact-effort matrix:
High Priority: AI mode toggle, confidence scores, metadata preview
Low Priority (Phased): Sector filtering, comparison charts
Final User Flow
"The final flow made the AI feel like a co-pilot"
1. Select Component → 2. Pick AI Mode → 3. Add Year/Quantity/Region → 4. Review Suggestions → 5. Apply → 6. Save Entry
Admin Flow: Enabling the Feature
Company Admins manage access via plan selection:
Basic Plan: Enables both AI modes using open-source datasets
Premium Plan: Adds access to proprietary datasets like ecoinvent
Ideation & Flow Mapping
Through sketches and whiteboard iterations, we debated:
Whether to use side-drawers or inline cards - Spoiler alert! We had to use modals due to development constraints
Should confidence scores be percentages, grades, or tooltips?
How to allow override without causing decision fatigue
Prototyping & Design Decisions
Wireframes helped me to validate:
Modals over drawers (less screen interference)
Progressive disclosure of suggestions
Showing source + confidence for each factor
Final Screens
Procurement manager View
Integrated AI suggestion engine within the existing flow
Company Admin View
Feature enablement and plan management
Outcome
3–5 mins → under 1 min
Time saved per entry with Best Match
190k+ factors
From 16 sectors, powering smarter coverage
>60% usage
Adoption rate within two months of rollout
Increased trust
Due to visible scoring and manual override
Al controls
Multi Suggest (list up to 20)
Best Match (one-shot).
Key inputs:
Declared unit = Piece;
Quantity = 12
Values appear in Production stage rows; users can edit or accept .
Press Save to complete one entry(counts toward volume)
Spotlight notes
Quick talk track for reviews
Results
Speed, corrections, and usage
Interpretation
What it means and what to watch
Entries are faster, people override less often, and adoption is strong by week four. Signals align, suggesting the Al suggestions are relevant and usable.
Trust & Accessibility
Keyboard path documented; visible focus states on all controls.
ARIA labels for the confidence widget and suggestion list.
“Confidence” tooltip: “Estimate based on source match, year, region, and unit.”
Provenance chip on apply: Source: Climatiq / ecoinvent · Year: 2023 · Region: EU.
Controls: Override, Undo, and an audit-log entry on apply.
Edge cases
No good match → prompt to refine filters or use Manual Select; keep AI panel visible.
Rate limit → non-blocking notice, auto-retry after delay; manual path available.
Network error → inline error with Retry; show last known suggestions if cached.
Low confidence (<40%) → badge + microcopy: “Review carefully—low match score.”
Key Learnings
Design for control, not automation
AI works best when users feel assisted, not overridden. Giving users the ability to toggle between Multi Suggest and Best Match helped build trust.
Design for control, not automation
AI works best when users feel assisted, not overridden. Giving users the ability to toggle between Multi Suggest and Best Match helped build trust.
Transparency builds credibility
Displaying confidence scores and data sources reassured users about the AI’s decisions, encouraging adoption.
Backend collaboration accelerates UX
Engaging technical stakeholders early enabled us to shape flows that were both user-friendly and technically feasible, saving weeks in delivery.