Transforming Email Chaos into Intelligent Workflows for a B2B Collections Agency
client
Cadex
Industry
B2B Payments and
Collections Agency
location
USA
project duration
Ongoing
Client Overview
Cadex, a leading B2B payments and collections agency, manages hundreds of daily support emails from clients, partners, and merchants. These emails often contain multiple issues in one thread, mixed formats (text, PDFs, images), and varying levels of urgency—making manual sorting and routing time-consuming and error-prone.
The Challenge
Despite experimenting with machine learning, Cadex struggled to move from proof-of-concept to production accuracy.
Their initial model worked well on test data, boasting over 90% accuracy, but once exposed to real-world emails, performance dropped sharply to 40%.
Key pain points included:
- Mixed topics in the same email
- Unbalanced training data
- Non-text attachments like invoices, scanned letters, and screenshots that the model couldn’t interpret
Streamlining Email Management, Categorization, and Response with Intelligent Automation
Auto Reply – No Info
System detects out-of-office auto reply, classifies as “Auto Reply – No Info,” no action required.
Auto Reply – With Info
System detects auto-reply with relevant information and categorizes it as “Auto Reply – With Info.”
No Reply – With Info
System identifies automated email containing informational content and classifies it as “No Reply – With Info.”
No Reply – No Info
System detects automated notification with no additional details and categorizes it as “No Reply – No Info.”
Invoice Request – No Info
System detects invoice request without billing details, categorizes it as “Invoice Request – No Info,” and triggers an automated response requesting details.
Claims Paid – No Proof
System identifies claim of payment without proof, categorizes it as “Claims Paid – No Proof,” and triggers an automated follow-up requesting supporting documents.
Proposed Solution
Galaxy Weblinks designed a multi-agent AI system that could read, understand, and route every incoming email—regardless of content type or complexity.
- Preprocessing Agent: Cleans and structures emails, extracts text from PDFs and images using OCR.
- Intent Classifier Agent: Detects the main purpose of each email using advanced language models.
- Workflow Router Agent: Automatically assigns each message to the correct operational queue or team.
- Feedback Tuner Agent: Continuously learns from team corrections to improve model performance.
Results
The AI-driven email workflow automation significantly improved operational efficiency and accuracy, resulting in a 50% reduction in workforce requirements.
- Email classification accuracy : Improved from 40% to 98% after the Galaxy Weblinks Solution.
- SLA compliance : Saw a +65% improvement from the baseline after the Galaxy Weblinks Solution.
- Average handling time : Showed a 30% reduction after the Galaxy Weblinks Solution.
- Model adaptability : Changed from Static to Daily self-improvement through feedback after the Galaxy Weblinks Solution.
Business Impact
Innovation Highlight
Galaxy Weblinks’ key innovation lies in its hierarchical classification approach—separating coarse-grained intent detection from fine-grained workflow routing. This allows the system to handle complex emails with multiple intents while maintaining high accuracy. The continuous learning loop ensures adaptability as customer communication patterns evolve.