Multi-tenant analytics: 240ms p99 across 600 customer workspaces

The situation

• A Berlin-based product analytics SaaS hit 600 paying workspaces faster than expected. The Lovable-generated schema put every customer's events into a single shared table with a `workspace_id` column.
• Queries against the largest 20 workspaces were locking the database, and dashboards were timing out for the customers paying the most. A Series A term sheet was on the table.
• The lead investor's tech diligence partner was scheduled to spend 4 hours with the codebase. The CTO needed a real multi-tenant data architecture, not a slide deck about one.

What we did

• Postgres + ClickHouse architecture with per-workspace partitioning
• Query planner routing between transactional and analytical stores
• Per-dashboard Redis cache with event-driven invalidation
• Materialize-powered live dashboard with sub-second freshness
• Natural-language query feature scoped per workspace
• Per-workspace latency dashboards on Grafana for SLA reporting

The result

• Series A closed two weeks after the diligence call at a 3.4× higher valuation than the previous round.
• Largest workspace dashboards now load in under 300ms. The natural-language query feature converted into a paid add-on within 60 days and is now used by 22% of accounts.