EN UA RU

Case studies

Work that speaks
for itself.

In-depth breakdowns of real projects — architecture decisions, timelines, and measurable outcomes.

$140M+Client revenue
12Industries
3.8×Avg. project ROI
14wkAvg. delivery
01
Retail & E-commerce 18 weeks · 2024

NovaMart survived Black Friday — and turned 14 hours of downtime into €2.4M in new revenue

Built with
340%
Peak capacity increase
Nov 2023 – Mar 2024
6 engineers · 1 PM
Frankfurt, Germany
The situation

A €120M/year fashion retailer whose Magento 1 infrastructure collapsed under load every sale season. Their November 2023 Black Friday resulted in 14 hours of downtime and €890K in lost orders — while competitors scaled effortlessly. The team had been patching the same infrastructure issues for three years with no sustainable fix.

What we did

Platform migration — Full rewrite to Magento 2.4 with headless PWA storefront decoupled from the backend, enabling independent scaling of each layer

Caching layer — Redis object cache, Elasticsearch catalog, Varnish full-page caching. Cache hit rate reached 94.2% within 2 weeks of go-live

AWS Auto Scaling — Event-aware scaling groups with pre-warming scripts deployed 2 hours before planned sales. No more surprise traffic spikes

Observability — Real-time dashboards, automated incident alerts, and a 15-minute runbook that any engineer can execute solo

“Our first Black Friday on the new platform was the calmest day of the year. Zero incidents. The team actually enjoyed it — which hadn't happened in four years.”

MR
Markus Riedel CTO, NovaMart GmbH · Frankfurt
console.aws.amazon.com / cloudwatch / novamart-prod
NovaMart — Black Friday 2024
Live — Nov 29
Peak RPS 14,200
Response 0.8 s
Uptime 99.97%
Instances 12 → 38
Request rate req/s
00:00 06:00 12:00 14,200 peak 18:00 00:00
p95 response time ms
EC2 instances count
Service CPU Latency Status
magento-web 62% 340 ms
elasticsearch 45% 18 ms
redis-cache 28% 2 ms
varnish-cdn 34% 5 ms
rds-primary 51% 12 ms
340%
Peak capacity increase
0.8s
Page load (was 2.1s)
+18%
Conversion rate
€2.4M
Additional revenue, yr 1
02
FinTech · Payments 14 weeks · 2023

Meridian Capital cut payment failures from 8.4% to 0.03% — and recovered $4.1M in annual revenue

Built with
99.97%
Payment success rate
Aug – Nov 2023
4 engineers · 1 PM
Amsterdam, Netherlands
Payment orchestration Live
Success rate99.97%
Avg routing time14 ms
Monthly volume$48.3M
The situation

Meridian's payment infrastructure was three PSPs wired together with no intelligent routing. When one provider degraded, transactions failed silently. Their 8.4% failure rate in Q3 2023 was costing $4.1M/year in lost transactions — while the finance team burned 40+ hours per week on manual reconciliation across all three systems.

What we did

Orchestration engine — Routes each transaction to the optimal PSP in real time by card type, geography, failure history and cost. Decision in <14 ms

Automatic fallback routing — Provider degrades mid-session? The engine re-routes instantly without the user ever seeing an error

Webhook normalisation — Unified event stream across 7 providers feeding one source of truth, eliminating the manual reconciliation gap

Automated accounting sync — Reconciliation runs every 15 min and syncs directly to their ERP. From 40 hours of manual work per week to zero

“We had been living with payment failures for so long they felt normal. Flexor showed us they weren't — and fixed it faster than we thought possible. The ROI was visible within the first month.”

SL
Sophie Laurent VP Engineering, Meridian Capital · Amsterdam
0.03%
Failure rate (was 8.4%)
14ms
Routing decision time
0h
Manual reconciliation/wk
$4.1M
Annual revenue recovered
03
Integrations eCommerce · 150K SKUs

How a 150K-product store stopped losing orders — by replacing REST chaos with RabbitMQ

Built with
99.9%
Sync reliability
Q3 2025 — 10 weeks
4 engineers
EU — remote
The situation

A large eCommerce retailer with 150,000 products and 5,000 orders per day ran constant synchronisation between their store, PIM system and CRM — products, stock levels, orders, prices, customer data, and attribute updates. All integrations were built on synchronous REST API calls. During peak traffic the integration layer would collapse: timeouts piled up, orders were lost or duplicated, stock figures went out of sync, and every failure cascaded into the next. On an average week 12–15% of API requests simply failed, and the operations team spent hours every morning patching data by hand.

What we did

Integration audit — Mapped all 27 data flows between store ↔ PIM ↔ CRM. Identified the four deadliest bottlenecks: bulk product imports (150K SKUs), real-time stock sync across 3 warehouses, order status callbacks, and attribute propagation for 80+ custom attributes.

RabbitMQ migration — Replaced every synchronous REST call with asynchronous message queues. Each data type — products, inventory, orders, attributes — got its own exchange, routing topology and dedicated consumers with independent scaling.

Retry & dead-letter logic — Built a three-tier retry strategy: immediate retry, delayed retry (exponential backoff up to 30 min), and dead-letter queue with Slack alerts. No message is ever silently lost — every failure is tracked and re-processed.

Observability stack — Deployed centralized logging (ELK), real-time queue dashboards, and automated alerting. The team now sees every message lifecycle — from publish to acknowledgement — in one place. Average incident response dropped from hours to minutes.

“We used to start every Monday untangling weekend sync failures. Now 200,000 messages a day flow through the queues without a hiccup. The operations team finally trusts the data — and we haven't lost a single order in three months.”

DK
Daniel Kravchuk Head of eCommerce Operations
grafana.internal / d / queue-health
RabbitMQ — Queue Health Last 24h
Messages / day 203,841
Failed 3
Avg latency 284 ms
Consumers 12
Message throughput msg/min
Queue depth messages
Failed / retried count
Queue Ready Rate Status
products.sync 24 82/s
orders.create 3 58/s
inventory.update 11 145/s
attributes.propagate 0 34/s
crm.customer 1 21/s
99.9%
Sync reliability
<5
Failed syncs per day
0.3s
Avg sync latency
0
Manual interventions / month
04
Architecture B2B · €45M GMV

A B2B wholesale platform escaped its 8-year-old monolith — and cut deploy time from 4 hours to 8 minutes

Built with
97%
Deploy time reduction
Q1–Q2 2025 — 16 weeks
5 engineers · 1 architect
Berlin, Germany
github.com / wholesalehub / platform / actions
Deployment Pipeline
Live
Tests 1m 42s
Build & Push 3m 18s
Deploy to staging 1m 05s
Deploy to production 1m 47s
Total: 7m 52s
Service Pods CPU Status
catalog-svc 3/3 24%
pricing-svc 2/2 18%
orders-svc 4/4 31%
inventory-svc 2/2 12%
invoicing-svc 2/2 9%
api-gateway 3/3 15%
The situation

A B2B wholesale distributor with 2,000+ business customers and €45M annual GMV ran everything on an 8-year-old Symfony monolith. A single deployment took 4 hours of downtime, one bug in the pricing module could crash the entire platform, and the team had stopped shipping features — spending 70% of their time firefighting. Three failed attempts to refactor in-place had drained the team's confidence.

What we did

Domain mapping — Identified 6 bounded contexts (catalog, pricing, orders, customers, inventory, invoicing) and mapped all cross-domain dependencies before writing a single line of code.

Strangler fig migration — Extracted services one by one behind an API gateway, keeping the monolith running in production throughout. Zero big-bang cutover — each service went live independently.

Kubernetes & CI/CD — Containerized every service with Docker, orchestrated via Kubernetes on AWS EKS. Built a full CI/CD pipeline — push to main triggers automated tests, build, and rolling deploy in under 8 minutes.

Event-driven sync — Services communicate via RabbitMQ events instead of direct API calls. Eventual consistency with saga patterns for complex workflows like order → inventory → invoice.

“We went from being afraid to deploy on Fridays to shipping multiple times a day. The platform finally feels like it belongs to us again, not the other way around.”

TB
Thomas Bergmann CTO, WholesaleHub GmbH · Berlin
97%
Deploy time reduction
2
Incidents per month
18+
Deploys per week
+35%
GMV growth, year 1
05
Order management Omnichannel · Retail

How a retailer on 5 sales channels stopped overselling — and grew revenue 22% in 6 months

Built with
0
Oversells eliminated
Q4 2024 – Q1 2025 — 12 weeks
3 engineers · 1 PM
London, UK
The situation

A mid-market fashion retailer selling through their own Shopify store, Amazon, eBay, and 4 physical locations had no single source of truth for inventory. Stock levels were synced manually via spreadsheets twice a day. The result: 30+ oversells per week, angry customers, marketplace penalties, and a warehouse team spending half their day fixing order routing by hand. Peak season made everything worse — Black Friday 2024 alone caused 200+ oversold orders.

What we did

Centralized OMS — Built a custom order management hub that ingests orders from all 5 channels in real time, normalizes them into a single format, and routes to the optimal fulfillment location based on stock proximity and shipping cost.

Real-time inventory sync — Connected all channels via APIs (Shopify, Amazon SP-API, eBay) with sub-second inventory updates. When a unit sells anywhere, all channels reflect it within 2 seconds.

Smart order routing — Automated fulfillment logic: ship from nearest warehouse, split orders across locations when needed, auto-generate shipping labels. Manual routing dropped from 4 hours/day to zero.

Analytics dashboard — Real-time visibility across all channels: stock levels, order velocity, fulfillment SLA, channel profitability. The team makes data-driven decisions instead of gut calls.

“We used to dread every marketplace notification — half of them were oversell complaints. Now we run five channels from one screen, and haven't had a single oversell in two months.”

JW
James Whitfield Operations Director, Thread & Co · London
oms.threadco.io / dashboard
Order Management Hub
Live
Today's orders 342
Oversells 0
Inv. accuracy 99.8%
Avg fulfill 1.4 h
Channel Orders Revenue Status
Shopify Store 148 £12,420
Amazon UK 94 £8,130
eBay 52 £3,670
Retail (4 stores) 48 £5,890
Inventory sync latency seconds
0
Oversells eliminated
−60%
Fulfillment time
99.8%
Inventory accuracy
+22%
Revenue growth, 6 months