Stabilising and extending a multi-tenant HR platform

Our client publishes an HR platform used by mid-sized French companies for absence management, HR records and payroll preparation. The product is a multi-tenant Ruby on Rails application with PostgreSQL and Sidekiq background processing, hosted on AWS in eu-west-3 (Paris), owned by a small senior engineering team.

Enterprise customers were arriving faster than the team could absorb them. Each new prospect brought security questionnaires, single sign-on requirements and demands for integrations with payroll systems.

The same engineers who carried the product roadmap also carried production. Incident response interrupted every sprint, the bug backlog grew, and background jobs that moved payroll data occasionally failed silently — the kind of failure an HR product cannot afford. Meanwhile, two enterprise opportunities had stalled on missing SAML SSO and a payroll export.

Hiring more senior engineers in Paris was slow and expensive, and would still have left nobody on call at night.

We began with a six-week transition: reading the codebase, pairing with the client’s engineers, and writing runbooks for the failure modes they knew by heart. We instrumented the application with Sentry and Datadog APM, added structured logging, and put every Sidekiq queue under explicit monitoring, with dead-letter handling and idempotent retries for any job that touches payroll data.

From there, production became our responsibility under a Managed IT Maintenance (SLA) contract: 24/7 monitoring, P1–P4 incident classification, bug fixing and security updates. Before changing anything in the payroll calculation paths, we added characterisation tests around them. CI runs on GitHub Actions, with a staging environment that mirrors production data shapes through anonymised fixtures.

In parallel, our engineers delivered the features the sales pipeline was blocked on: SAML 2.0 and OIDC single sign-on, built as a dedicated authentication service in front of the monolith; a versioned REST API with webhook delivery; and a payroll-export connector producing files the client’s customers load into their payroll engines. Each integration shipped with tenant isolation and rate limiting from the first release.