Cloud Services in Stockholm

Technology / B2B SaaS

A Stockholm-based B2B SaaS company (SEK 650M ARR, 4,500 enterprise customers across Europe and North America, 450 employees — providing analytics infrastructure for data teams) needed to re-architect its cloud platform to support international expansion, reduce infrastructure costs as a percentage of revenue, and meet the data residency requirements of enterprise customers in regulated industries. (1) Single-region limitation: all infrastructure ran in a single AWS eu-north-1 region — providing excellent performance for Nordic customers but 180-280ms latency for North American customers (representing 35% of ARR and 60% of new pipeline). US enterprise prospects frequently cited performance as a concern during evaluation, and 3 deals over USD 500K had been lost to competitors with US cloud presence. (2) Infrastructure cost trajectory: cloud spending was SEK 85M annually (13% of ARR) — above the 8-10% benchmark for efficient SaaS companies at this scale. Infrastructure costs were growing faster than revenue (40% YoY vs. 30% YoY revenue growth), and the engineering team had limited visibility into which customers, features, or services drove costs. Without intervention, infrastructure costs would exceed 15% of ARR within 18 months. (3) Data residency requirements: EU enterprise customers (40% of ARR) required EU data residency — currently satisfied by the Stockholm deployment. But US enterprise customers (35% of ARR) increasingly required US data residency (CCPA compliance, federal government FedRAMP requirements, and enterprise data governance policies mandating US processing). The company could not serve both requirements from a single region. (4) Reliability concerns: the platform had experienced 3 significant outages in the past year — each lasting 2-4 hours. With 4,500 enterprise customers depending on the platform for analytics, each outage generated hundreds of support tickets, executive escalations, and contract negotiation challenges. The published SLA was 99.9% — but actual availability was 99.7%. Enterprise customers were beginning to request stronger SLAs (99.95% or 99.99%) that the current architecture could not support. (5) Scaling bottleneck: the platform's monolithic data processing engine was reaching scale limits — individual customer workloads could consume disproportionate resources, creating noisy-neighbour issues that affected other customers. The engineering team spent 30% of its time on scaling-related firefighting rather than feature development.

We re-architected the cloud platform over 14 weeks. (1) Multi-region deployment: three-region architecture — AWS eu-north-1 (Stockholm — EU primary, serving European customers), us-east-1 (Virginia — US primary, serving North American customers), and eu-west-1 (Ireland — EU failover). Data residency: customer data pinned to the region matching their data residency requirement — EU customers in Stockholm, US customers in Virginia. Configuration-driven: region assignment configurable per customer without code changes. Control plane: centralised management in Stockholm with regional data planes — customer data never left the assigned region, while management operations spanned regions. (2) Cost optimisation: Kubernetes migration — remaining monolithic services decomposed into microservices on EKS, enabling independent scaling of components. Right-sizing: automated instance right-sizing based on actual utilisation — identified SEK 18M in immediate over-provisioning. Spot instances: analytics processing workloads (interruptible by nature) migrated to spot instances — 65% cost reduction for these workloads. Storage tiering: hot/warm/cold data tiering for customer analytics data — recent data on SSD, older data on standard storage, archived data on Glacier. Savings plans: 3-year compute savings plans for baseline capacity — 40% savings vs. on-demand. Customer cost attribution: per-customer cost calculation enabling unit economics analysis — identifying unprofitable customer segments and informing pricing decisions. (3) Multi-tenant re-architecture: workload isolation — each customer's processing workload isolated in dedicated Kubernetes namespaces with resource quotas. Fair scheduling: resource allocation ensuring no single customer could consume disproportionate capacity — eliminating noisy-neighbour issues. Burst capacity: customers could burst above their allocation when capacity was available — providing good performance without permanent over-allocation. Tiered performance: different SLA tiers (standard, premium, enterprise) with corresponding resource guarantees — enabling pricing differentiation based on infrastructure commitment. (4) High availability: multi-AZ deployment in each region (3 AZs per region for production workloads). Database: Aurora Global Database with multi-AZ writes — automated failover within region (sub-30 seconds) and cross-region failover (under 5 minutes) for regional disasters. Load balancing: ALB with health checking and automatic instance replacement — unhealthy instances removed and replaced without customer impact. Chaos engineering: regular failure injection (service failures, AZ failures, database failovers) validating resilience — automated and scheduled monthly. (5) Observability and SRE: distributed tracing across all services (OpenTelemetry), SLO-based alerting (alert on customer-impacting issues, not infrastructure metrics), error budgets (each service with defined error budget — automatic deployment freezes when budget exhausted), and incident management (PagerDuty integration with automated runbooks for common failure scenarios).

International expansion: US region launched within 4 months. 8 enterprise deals (USD 500K+ ACV each) closed within 12 months citing US data residency as a deciding factor — SEK 52M in additional ARR. US customer satisfaction improved from 3.5/5 to 4.6/5 for platform performance (latency from 180-280ms to 15-35ms). Pipeline: US enterprise pipeline increased 85% — the multi-region capability unlocked previously inaccessible market segments (US government, US healthcare, US financial services). Infrastructure costs: from SEK 85M to SEK 58M annually (32% reduction) — while handling 40% more traffic. Infrastructure as percentage of ARR: from 13% to 8.9% (below the 10% benchmark). Per-customer cost: from SEK 18,900 to SEK 11,200 (41% reduction). Cost visibility: 98% of cloud spending attributed to specific customers and features — enabling informed product and pricing decisions. Reliability: zero customer-impacting outages in 12 months (from 3 the previous year). Measured availability: 99.98% (from 99.7%). Enterprise SLA: now offering 99.95% SLA to enterprise customers — a competitive differentiator. Noisy-neighbour incidents: from 15 per quarter to zero — workload isolation eliminating cross-customer impact. Engineering productivity: scaling-related firefighting from 30% of engineering time to 5% — redeploying engineering capacity to feature development (equivalent to 12 additional engineers). Annual value: SEK 52M (US expansion — new ARR from deals requiring US data residency), SEK 27M (cost reduction — infrastructure spending from SEK 85M to SEK 58M), SEK 18M (engineering productivity — 25% engineering time recovered from scaling firefighting), SEK 15M (churn reduction — improved reliability reducing churn by 2.5% on the customer base), SEK 8M (enterprise upsell — premium SLA tier driving higher ACV for enterprise customers) = SEK 120M (approximately USD 11.5M). Development cost: SEK 3.8M (approximately USD 365K). Payback: under 12 days.