OPS-24-A4
Lead Intelligence
Continuous lead intelligence system for a B2B SaaS sales team
Daily multi-source enrichment of 40k+ accounts, intent signals, decision-maker mapping. Hands-off since Q2 2024.
9.4×pipeline
OPS-24-B1 — Marketplace Scraping
Real-time pricing intelligence across 1,200 marketplaces
Resilient scraping with anti-bot routing, SKU normalization and 5-minute price-change webhooks into the client's repricing engine.
2.1Mproducts/day
Challenge
The client — a large e-commerce company with a private portfolio of 50 own brands — needed a view of their and direct competitor prices across 1,200+ marketplaces in EU and US. They previously bought data from two providers. Problems: 24h delay (their pricing cycle is in hours), low SKU coverage (~65%), no visibility into data quality, cost €38k/mo.
Goal: monitor 2M+ products per day, <5 minute delay from price change at source to webhook in the repricing engine, SKU coverage >95%, operating cost <€15k/mo.
Approach
Three-layer architecture. Scraping layer: 480 Playwright workers in Kubernetes across three regions (EU-West, EU-Central, US-East), each with isolated fingerprint and its own residential proxy pool. Distribution per marketplace optimized for their individual rate limits and detection patterns.
Normalization layer: SKU matching via combination of EAN/UPC/MPN, fuzzy name matching (Levenshtein + embedding similarity for atypical cases), canonical product graph in Postgres with 18M nodes. Every new record hits this graph — either as a match to an existing SKU or a new node.
Delivery layer: change detection on every record, webhooks under 5 minutes from change (most <90 seconds), hourly batch export to S3 in Parquet, Next.js dashboard for client analysts.
Outcome
SKU coverage: 97.2%. Median price-change detection time: 84 seconds. Operating cost: €11,800/mo (proxies, infra, LLM for edge cases).
The client saved €26k/mo versus prior providers, gained 23× shorter latency, and over 30 points higher SKU coverage.
The system survived two Cloudflare anti-bot engine updates in 2024 (each patched within 6h of detection), and a full Amazon EU UI migration in August 2025 (patched within 18h via an API backup path).
Stack
PlaywrightKubernetesBright Data + OxylabsPostgresS3 + ParquetTemporalNext.jsGrafana
Metrics
- 2.1M — Products/day
- 1,200+ — Marketplaces
- 84s — Median latency
- 97.2% — SKU coverage
- −68% — Saving vs prior
- 99.94% — Uptime
More cases
Back to case studies ↗OPS-25-C2
Agent Operations
Autonomous research agent for a financial intelligence firm
Goal-driven agent crawling filings, press, social and internal sources — producing structured analyst briefings every morning before 7 AM ET.
−72%analyst hours
OPS-25-D3
Social Operations
Multi-account content & outreach orchestration for a creator network
47 brand accounts, four platforms, one operator console. Scheduling, engagement, analytics and human-in-the-loop review built end-to-end.
11.6×output
Similar problem in your business?
Every project is different, but patterns repeat.
If you recognise pieces of this case study in your own situation — write. We usually see in the first call whether it is hours-per-week scale or months of infrastructure.