Highest Seller NPS in the Industry for India's Top Seller Aggregator
9%
Uplift in Seller NPS (Industry-leading)
50%
Faster seller approval rates
80%
Reduction in QC turnaround time
About the Client
India's largest seller aggregator and marketplace enablement platform, helping small and medium businesses succeed on major e-commerce platforms. The company provides end-to-end services including catalogue management, inventory optimization, pricing intelligence, and fulfilment-serving as the operational backbone for tens of thousands of sellers.
Challenge
The platform's seller onboarding and catalogue management process had become the primary source of seller frustration, directly impacting retention and growth:
Manual, Complex, and Error-Prone Cataloguing: Sellers were required to fill 30+ attributes per product listing, many of which were redundant or irrelevant for their category. The average time from product receipt to live listing was 4-6 days.
Attribute Overload Slowed Sellers Down: For sellers managing hundreds of SKUs, this meant weeks of operational overhead before products could generate revenue. Competitors with simpler onboarding flows were winning seller acquisition battles.
Inconsistent QC Rejections (False Flags): Quality control processes were plagued by inconsistency. Sellers reported 40%+ rejection rates, with many rejections being false positives.
Incorrect Category Mapping: Products were frequently miscategorized, either by sellers unfamiliar with marketplace taxonomies or by automated systems that couldn't handle edge cases.
Low Seller NPS Compared to Competitors: The platform's Seller NPS had dropped to the bottom quartile of the industry. Exit surveys revealed that "catalogue hassles" were the #1 reason sellers churned to competing aggregators.
Our Approach
BergLabs implemented a comprehensive AI-powered catalogue transformation, simplifying workflows while dramatically improving accuracy and speed.
Simplified and Automated Seller Workflows
Intelligent automation to reduce manual data entry and streamline seller experience.
- Intelligent Attribute Inference: Instead of requiring sellers to manually fill every field, BergForge agents pre-populated attributes by analyzing product images and titles. Sellers only needed to verify and correct-reducing average attribute entry time from 12 minutes to 2 minutes per product.
- Smart Defaults by Category: Based on historical data, we implemented category-specific defaults for low-variance attributes. "Country of Origin" defaulted to India for 94% of sellers; "Return Policy" defaulted to the platform standard.
AI Auto-Fill from Image + Text
Computer vision and NLP to extract structured data from unstructured inputs.
- Computer Vision Attribute Extraction: BergForge's vision agents analyzed product images to extract physical attributes (size, color, shape), text recognition (brand names, model numbers, specifications), and contextual features.
- Natural Language Processing on Titles: Product titles were parsed to extract structured data. "Apple iPhone 15 Pro Max 256GB Space Gray Unlocked" was automatically decomposed into brand, model, variant, storage, color, and carrier compatibility.
- Cross-Reference Enrichment: When confidence was low, agents cross-referenced against existing catalogue data to infer missing attributes.
Computer Vision Auto-Detected Categories
Automated category classification using image analysis.
- Hierarchical Classification: Instead of requiring sellers to navigate 5-level category trees, our vision agents predicted the correct category path from product images alone. Accuracy reached 94% for top-level categories and 87% for leaf-level placement.
- Confidence-Based Routing: High-confidence classifications (>90%) were auto-approved. Medium-confidence (70-90%) were presented to sellers for confirmation. Low-confidence (<70%) were routed to human experts.
- Category Conflict Resolution: When image analysis suggested one category but title analysis suggested another, the system flagged the conflict for human review.
Simplified QC Guidelines + Contextual Auto-Blocks
Streamlined quality control with actionable feedback.
- Rule Reduction: We audited the platform's 500+ QC rules and eliminated 60% that were redundant, outdated, or generating high false-positive rates.
- Contextual Rejection Logic: Instead of binary pass/fail, the new system provided specific, actionable feedback with clear instructions for correction.
- Auto-Block for Critical Violations: Certain violations (trademark infringement, prohibited products, adult content) triggered immediate blocks without human review.
Reduced Manual QC Touchpoints
Risk-based quality assurance to improve efficiency.
- Statistical Sampling QA: Instead of reviewing every listing, we implemented risk-based sampling. New sellers received 100% review for their first 20 listings; established sellers with <5% historical rejection rates received only 10% spot-checks.
- Automated Duplicate Detection: BergForge agents identified duplicate listings before QC review, preventing wasted reviewer time.
- Seller Self-Service Fixes: For minor issues, sellers received automated suggestions and could fix-and-resubmit without entering the QC queue again.
┌─────────────────────────────────────────────────────────────┐
│ Seller Upload Interface │
│ (Images + Basic Product Info) │
└─────────────────────────────┬───────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────┐
│ BergForge Agent Layer │
├──────────────────┬──────────────────┬───────────────────────┤
│ Vision Agent │ NLP Agent │ Enrichment Agent │
│ - Image analysis│ - Title parsing │ - Cross-reference │
│ - Color detect │ - Entity extract│ - Default inference │
│ - Category pred │ - Brand detect │ - Gap filling │
└──────────────────┴──────────────────┴───────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────┐
│ Confidence Routing Layer │
│ │
│ >90% ───────▶ Auto-Approve ───────▶ Live Listing │
│ │
│ 70-90% ─────▶ Seller Confirm ─────▶ Live Listing │
│ │
│ <70% ───────▶ Human Expert ───────▶ Manual Review │
└─────────────────────────────────────────────────────────────┘Impact
The AI-powered catalogue transformation delivered industry-leading seller satisfaction:
| Metric | Before | After | Improvement |
|---|---|---|---|
| Seller NPS | Baseline | +9% | Industry's highest |
| Seller Approval Rate | 60% | 90% | 50% faster |
| QC Turnaround Time | 48 hrs | 10 hrs | 80% reduction |
| Manual QC Touchpoints | 100% | 30% | 70% reduction |
| Average Listing Time | 4-6 days | 1 day | 80% faster |
Seller Retention: 30-day seller retention improved by 22%, with catalogue experience cited as the primary driver in follow-up surveys.
Operational Efficiency: The platform reduced QC headcount by 40% while processing 3x more listings daily-reallocating resources to seller success and growth initiatives.
Competitive Advantage: The platform moved from bottom-quartile to top-quartile Seller NPS within 6 months, becoming a key differentiator in seller acquisition.
Testimonial
“Everything feels connected now-just upload and go. Before BergLabs, listing a new product felt like filling out a tax form. Now it feels like posting to Instagram. My team can list 50 products in the time it used to take us to do 10.”
Founder
Multi-Category Seller (5,000+ SKUs)
Engagement Model
Type
Assisted Operations (AI platform + client workforce)
Duration
8 weeks implementation + ongoing automation
Team
BergForge platform deployment, 10 BergLabs engineers, client QC team training
Platforms
BergForge (agent factory), BergFlow (human review workflows)