BuyLead Matchmaking System
Owning the buyer-seller matching algorithm for India's largest B2B marketplace—optimizing lead relevance, reducing friction, and driving transactions.
The Challenge
Beta Industries connects millions of buyers and sellers across India. As APM, I owned the BuyLead matchmaking system—ensuring sellers receive relevant leads that convert to transactions, not noise they reject.
How It Works
Requirement
Algorithm
BuyLeads
or Feedback
Preference Optimization
Improving lead relevance by honoring seller transaction behavior
Dynamic Transaction Order Value (TOV)
High retail NI givers received low-value leads they consistently rejected
Calculated minimum TOV preference only for sellers with ≥1 Retail NI. Daily calculation (vs weekly). Differential thresholds: 10% for high NI givers, 5% for others.
Order Value Slab Deboosting
Leads far below seller's preferred TOV cluttered relevant section
Forcefully deboosted BuyLeads from 2+ slabs below seller's minimum TOV slab. Applied to both Relevant and Recent BL sections.
High Value Lead Prioritization
High-value leads (≥₹10K) mixed with general ranking, missing monetization signals
Re-designed BL display grids to prioritize leads ≥₹10,000 at top position for sellers without pre-defined preferences.
AOV & Quantity in BL Search
Bug caused quantity flags to populate incorrectly, dropping quantity lead consumption
Rebuilt logic to honor seller's preferred AOV and quantity in search. Fixed critical bug in eto_ofr_mapping table.
Ranking & Location Logic
Optimizing lead display based on freshness, proximity, and performance
Recency + Location Based Boosting
Basic recency check without fine-grained distance led to suboptimal conversion
Bifurcated top 10 grids: 1) <2hr & <250km (top priority), 2) 2-24hr & <250km, 3) 0-24hr & >250km.
BA/BB Sub-rank Segregation
All B-rank categories grouped together regardless of transaction history
Segregated B rank into BA (has transactions) and BB (0 transactions). BA given higher display priority.
Hyperlocal MCAT Honoring
40% of transactions occur within 50km—distant leads caused high NI
Set up hyperlocal MCAT flag. Deboosted leads from hyperlocal categories when seller-buyer distance >250km.
Consuming Cities Logic
System ignored cities where seller had successfully done business
Developed "consuming cities" logic: any city with ≥1 transaction in last 12 months gets boosted grid position.
DLP Filter Integration
Broad filters ("All Locations") showed leads outside DLP range, causing Location NI
Integrated DLP logic into location filters. Leads outside seller's DLP displayed below blue strip.
Foreign Seller PMCAT Boost
PMCAT leads suppressed for foreign sellers despite good transaction data
Removed PMCAT suppression for Foreign DLP sellers. Applied to Relevant and Recent BL sections.
Feedback & Catalog Hygiene
Real-time feedback loops and catalog quality improvements
Real-Time Negative Marking (Info NI)
Nightly batch process delayed removing irrelevant categories
Enhanced real-time decision: MCAT marked negative immediately when Insufficient Info NI/Txn ≥10x (60-day window).
Other NI & Brand NI Marking
Negative marking focused only on "Wrong Product" NI, missing other signals
Extended marking to "Other NI" (≥10x ratio) and Brand NI (≥2x ratio). 372 MCATs marked for 288 sellers in 8 days.
New Seller A-Rank Boost
New paid sellers started with B rank, hindering initial visibility
Upgraded default MCAT rank for newly hosted paid sellers to A rank for first 30 days.
Category Taxonomy Cleanup
Key categories had inaccurate taxonomy, mis-mapping, missing product types
Created proper PMCATs (Thermocouple, RTD, Thermistor). 6 new MCATs. Mass cleaned 4,137 products. Enriched 2,924 products.
Operations Infrastructure
Internal tools for faster resolution and monitoring
Automated BLNI Ticketing
New sellers with high NI needed prompt follow-up, but no automated flagging
Developed "BL NI-Welcome" ticket triggered for new sellers on every 2nd BL NI. Real-time operation.
Internal CRM Auto-Assignment
Manual case assignment caused 3-hour resolution TAT
Developed internal CRM that auto-assigned seller cases to call center team.
Gladmin NI Dashboard
Teams switched between 4 screens for audit, ran manual queries
Developed centralized "Not Interested Report" dashboard. Added NI Summary screen with all critical data points.
The Smart Library Analogy
If Beta Industries is a huge library, my work ensured that every new reader (seller) starts with the highest-rated books (A rank boost) that match their preferred genre (MCAT mapping) and size (TOV). When they repeatedly mark books as irrelevant (NI), the librarian (algorithm) learns instantly and hides those specific titles forever—replacing manual sorting with automated, self-correcting intelligence.