# Migration → FinOps handoff checklist

The migration program (MAP, data-center exit, or lift-and-shift) ends. The bill
does not. This is the explicit handoff that stops a freshly-migrated estate from
becoming next year's "why is AWS so expensive" ticket. Work it in the **first 30
days post-cutover**, while the migration team still has context.

Treat each line as Done/Not-done. The point of failure is almost always an
*ownership* gap, not a technical one: nobody owns the bill after the migration
SI rolls off.

## 0. Ownership & cadence (do this first)

- [ ] Named **FinOps owner** (or owning team) assigned, in writing, before the
      migration team rolls off.
- [ ] Monthly cost-review cadence on the calendar with eng + finance + product.
- [ ] Anomaly alerting on (AWS Cost Anomaly Detection) routed to a channel a
      human reads.
- [ ] "Definition of done" for handoff agreed: tagging coverage %, RI/SP
      coverage target, and a first optimization backlog (see CSV template).

## 1. Visibility (you cannot optimize what you cannot see)

- [ ] Cost allocation tags activated in the Billing console (they are inactive
      by default even if resources are tagged).
- [ ] Tagging coverage measured (% of cost carrying owner + environment + app).
      Migration commonly leaves 30–60% untagged on lifted resources.
- [ ] Tag policy + (optional) SCP/`RequireTags`-style guardrail in place so new
      resources arrive tagged.
- [ ] **Cost Optimization Hub** opted in (org-wide). It consolidates 18
      recommendation types across accounts/Regions, dedupes them, and prices
      them with *your* RI/SP discounts.
- [ ] Per-team/app cost views (Cost Explorer saved reports or CUR + Athena)
      shared with the people who can act on them.

## 2. The post-migration waste that lift-and-shift always leaves

- [ ] **Right-size** over-provisioned EC2/RDS (migrations copy on-prem specs,
      which were sized for peak + headroom). Drive from Compute Optimizer via
      Cost Optimization Hub.
- [ ] **Idle / orphaned** resources: unattached EBS, idle RDS, old snapshots,
      unused Elastic IPs, idle NAT Gateways, zombie load balancers.
- [ ] **Storage tiering**: S3 lifecycle / Intelligent-Tiering; gp2 → gp3 on EBS.
- [ ] **Graviton** candidates (managed services first: RDS/ElastiCache/OpenSearch,
      then stateless compute).
- [ ] **Dev/test scheduling**: stop non-prod nights/weekends.
- [ ] Modernization backlog seeded (not done now): monolith → containers/serverless
      where it pays for itself — feeds the next quarter, not the handoff.

## 3. Commitment discipline (only AFTER right-sizing)

- [ ] Right-size **before** buying commitments. Buying RIs/SPs on
      over-provisioned instances locks in the waste.
- [ ] Establish a coverage target (e.g. 70–80% of stable baseline on
      Compute Savings Plans), reviewed monthly — don't commit 100%.
- [ ] Decide ownership of commitment purchases (usually FinOps/finance, informed
      by eng).

## 4. Guardrails so it doesn't regress

- [ ] AWS Budgets per team/app with alerts (and, where appropriate, Budgets
      Actions to cap runaway non-prod).
- [ ] Unit-cost metric defined (cost per tenant / per order / per 1k requests) so
      "the bill went up" can be judged against "did the business grow."
- [ ] Cost added to architecture-review and Well-Architected cadence.

## "If you only do one thing"

Assign a named owner for the bill **before** the migration team leaves, and put
a recurring monthly cost review on the calendar. Every other line on this list
gets done eventually if an owner exists, and none of them get done if one
doesn't.