Client retained at $11K/month · Phase 3: AI natural language query interface ("How much did I spend on marketing in Q3?") · Client referred i-verve to 2 other SaaS companies in their investor portfolio
Who we worked with
A Series B SaaS company offering a cloud-based accounting and financial management platform used by 8,000+ small and mid-sized businesses across 30+ countries. Approximately 120 employees. The platform handled invoicing, expense tracking, bank reconciliation, and financial reporting.
The platform wasn't failing — it was hitting architectural limits that blocked growth. A 7-year-old monolithic Node.js codebase, single-tenant architecture, and zero AI features were creating churn as competitors shipped intelligent automation. The founding engineering team was burned out from firefighting scaling issues while needing to ship AI simultaneously.
Architecture limits becoming a competitive threat
The platform worked — but scaling past 10,000 customers was operationally unsustainable, reports were slow for power users, and competitors were pulling customers away with AI features the monolith couldn't support.
800K-Line Monolith Choking Releases
The codebase had grown to 800,000+ lines. Every feature deployment required testing the entire application — stretching the release cycle to 2–3 weeks and blocking the product roadmap.
2–3 wk release cycleSingle-Tenant Architecture — Unscalable
Each customer had a separate database instance. Managing 8,000 individual databases was already painful — and scaling past 10,000 customers was operationally unsustainable without a complete rethink.
10K customer ceilingReports Taking 60 Seconds for Power Users
Customers with large datasets (50,000+ transactions) waited 30–60 seconds for reports to generate — a daily frustration that was showing up in support tickets and churn surveys.
30–60 sec report timesCompetitors Shipping AI — Churn Rising
Xero, FreshBooks, and others were shipping smart categorization, auto-reconciliation, and cash flow prediction. The client had none of these — and churn was increasing 2% per quarter as customers switched.
+2% churn per quarterIsolated Data — No Path to ML
The single-tenant architecture meant each customer's data was completely siloed. Training ML models across customer data for AI features was architecturally impossible without a shared analytics layer.
Zero shared ML layerEngineering Team Burned Out
The founding team was consumed firefighting scaling issues and had no capacity to build AI features simultaneously. They needed a partner who could own both the migration and the AI development at the same time.
No AI delivery capacityTwo phases, one AI-native SaaS platform
We ran the architectural migration and AI feature development in overlapping phases — resolving the scaling constraints while simultaneously building the intelligence layer that stopped the churn.
- Migrated to multi-tenant microservices — Node.js + TypeScript, organized by domain: invoicing, expenses, reconciliation, reporting, user management, billing
- Shared PostgreSQL cluster with row-level security — tenant data isolation maintained, but cross-tenant analytics now possible for ML training with full anonymization
- Shared data warehouse on Snowflake — aggregates opt-in anonymized transaction data across tenants for ML model training, with full GDPR compliance and consent management
- Event-driven architecture using AWS SQS/SNS for inter-service communication — decoupled services that scale independently
- Deployment cycle: 2–3 weeks → 3–5 per day — GitHub Actions CI/CD with feature flags (LaunchDarkly) for safe, incremental rollouts
- Query performance optimized — report generation for large datasets reduced from 30–60 seconds to under 3 seconds through indexing, query restructuring, and caching
- Intelligent expense categorization — NLP model trained on 12M+ anonymized transactions automatically classifies expenses at 94.7% accuracy out of the box, rising to 97.3% after 30 days of per-customer feedback learning
- Automated bank reconciliation — fuzzy matching + ML matches bank feed entries to invoices and expenses, auto-reconciling 82% of all entries without any human input (previously 100% manual)
- Predictive cash flow forecasting — trained on 12 months of each customer's own transaction history, predicts incoming and outgoing cash for the next 30, 60, and 90 days, visualized as an interactive dashboard chart
- AI anomaly detection — proactively flags unusual transactions, duplicate expenses, missing invoices, and tax discrepancies before end-of-month reconciliation
- AI Insights panel — natural language summaries on the dashboard: "Your accounts receivable is 23% higher than last month. 3 invoices over 60 days past due are driving this — here they are."
Full-stack AI-native SaaS architecture
Results that changed how they compete
Measurable improvements across churn, deployment velocity, report speed, reconciliation automation, and product adoption — all within 20 weeks, with AI becoming the platform's primary competitive differentiator.
Reduction
Auto-Reconciled
Generation
Adoption in 3 Months
5 engineers, one platform-defining delivery
A compact, senior-weighted team that owned the full scope — multi-tenant migration, AI model development, and production rollout to 8,000+ live customers — in parallel, without disrupting a single user.
Architect / Tech Lead
Microservices design, multi-tenant strategy, Snowflake data warehouse, GDPR consent architecture
Senior Node.js / TypeScript Developer × 2
Domain microservices, event-driven SQS/SNS architecture, query optimization, API layer
ML Engineer
Python NLP categorization model, fuzzy reconciliation, Prophet cash flow forecasting, SageMaker deployment
DevOps / QA Engineer
AWS ECS, GitHub Actions CI/CD, LaunchDarkly feature flags, opt-in beta rollout to 8,000+ tenants
20-Week Full Delivery
AI features launched as opt-in beta to 500 customers, then rolled out to all 8,000+ — zero downtime throughout the migration
What the client said
"i-verve solved two problems at once that our team couldn't tackle alone — the architectural migration we'd been putting off for two years, and the AI features we desperately needed to stop the competitive bleeding. In 20 weeks they shipped what would have taken us two years. The churn impact was immediate. Our AI features are now the number one reason new customers choose us over Xero and FreshBooks. This engagement genuinely changed our trajectory as a company."