Skip to content

Responsibilities (RACI): who owns what

This page explains Responsibilities (RACI): who owns what and how it fits into the RecSys suite.

Who this is for

Engineering leads, product owners, SRE/on-call, and security reviewers planning a pilot or production rollout of the RecSys suite.

What you will get

  • A shared-responsibility model for adopting the RecSys suite
  • A RACI matrix you can use during procurement and delivery planning
  • Clear handoffs between integration, data, and operations

Default responsibility model

By default, the RecSys suite is self-hosted: your organization runs the infrastructure and owns the data. The maintainers provide the software, documentation, and support channels.

If you operate the suite as a managed service, the same responsibilities exist but the owner per row may change.

Roles used in the RACI

  • PO: Product owner (business outcomes, prioritization)
  • App: Application team (frontend/backend integration)
  • Data: Data engineering / analytics (events, pipelines, evaluation)
  • SRE: Platform / SRE / on-call (infra, deploys, monitoring, incidents)
  • Sec: Security / compliance (PII, access controls, risk review)
  • RecSys: RecSys suite maintainers/support (product expertise, reviews, escalation)

Legend:

  • R = Responsible (does the work)
  • A = Accountable (owns the outcome)
  • C = Consulted (gives input)
  • I = Informed (kept in the loop)

Pilot RACI (DB-only mode)

In DB-only mode you can validate the full “serve → log → eval” loop without object storage or pipelines.

Activity PO App Data SRE Sec RecSys
Define surfaces and success metrics A/R C C I I C
Provision dev/staging environments (DB, deploy) I I C A/R C C
Integrate recsys-service API into one surface C A/R C C I C
Propagate request_id and stable IDs C A/R C C I C
Emit exposure logs and outcome logs C R A/R C C C
Run offline evaluation (recsys-eval) and review results C C A/R I I C
Operate runbooks (empty recs / not ready) I C C A/R I C
Decide what to ship (gate/rollout plan) A/R C C C C C

Production RACI (artifact/manifest + pipelines)

Artifact/manifest mode adds a second “supply chain”: pipelines publish versioned signals and a manifest; the service reads the manifest pointer.

Activity PO App Data SRE Sec RecSys
Deploy recsys-service to production (auth, tenancy, limits) I C C A/R C C
Run pipelines on schedule and publish manifests I I A/R C I C
Monitor freshness, latency, and empty-recs rate I I C A/R I C
Roll back config/rules and/or manifest on incident I C C A/R I C
Security review (PII, retention, access, audit) I C C C A/R C
Post-incident review and follow-ups A/R C C C C I