Handoff Checklist

Why handoff is the deliverable

An FDE deployment that you can never leave is a failure, not a success. The work is only "done" when the customer's team can operate the deployment without paging you.

The economic version: FDE time is the most expensive line item in a customer deployment. Platforms that scale this category systematically reduce FDE-hours per customer over time. Your individual handoff quality is a direct input to the platform's gross margin.

The professional version: you don't grow into senior FDE work by holding onto deployments. You grow by handing them off well and taking on harder ones.

When to start handoff

Not "in the last week." Handoff starts when scale-to-SLA stabilizes — usually around day 45–60. You're still active on the deployment, but you're shaping the work for someone else to inherit. Concretely:

• Week 6–8: start writing runbooks. Don't wait until you're done shipping; runbooks written in the heat of shipping are more accurate than runbooks written from memory.
• Week 8–10: bring the customer's data team into incident response. They observe; you handle. This is training, not handoff yet.
• Week 10–12: formal handoff. Customer's team is primary; you're backup. Transition gradually, not abruptly.

Artifacts to leave behind

The full set, in priority order:

1. Runbooks (chapter 06)

Six core runbooks for the most-common failure modes, customized with the customer's stack and contacts. Written for their analysts, not for you.

2. The deployment overview doc

A single document that explains what was built, why, and how the pieces fit. The reader is a new customer-side engineer joining the team six months from now.

# [Customer] × [Platform] — Deployment Overview
Last updated: YYYY-MM-DD

## What this is
- Business outcome being driven
- Scope (in / out)
- Success criteria

## Architecture
- Diagram (system map + data flow)
- Each component, what it does, who owns it

## The customer's stack we touch
- ERP, CLM, warehouse, BI, identity
- For each: auth pattern, integration mechanism, ownership

## Our pipelines
- Ingestion (sources, schedules, locations)
- Extraction (which models, configurations, prompts)
- Persistence (schema layout, retention)
- Marts (dbt models, lineage)

## The SLA
- Per-class thresholds
- Measurement methodology
- Escalation if breached

## How to operate it
- Daily / weekly / monthly checks
- Where the alerts go
- Runbook directory: [link]

## Where to look when something's wrong
- Status dashboard: [link]
- SLA dashboard: [link]
- Pipeline runs: [link]
- HITL queue: [link]

## Who to contact
- Customer Success: [name, email]
- FDE backup contact: [name, email]
- Platform escalation: [tracking system link]

3. The data dictionary

Column-level documentation of every mart table. What it means, how it's derived, what NULLs imply. The customer's analysts will reuse the marts; they need to read this.

4. The eval / SLA reproducibility kit

If the SLA is contested at renewal, you'll need to reproduce the measurement. Leave behind:

• The gold-set audit sample (or instructions to regenerate).
• The scoring scripts.
• The thresholds, with rationale.
• Historical measurements with timestamps.

5. The customer-specific configuration

Whatever you customized for this customer — extraction prompts, doc-class taxonomies, field mappings, supplier-master overrides. All in version control. All documented.

6. The "what we tried that didn't work" file

The most underrated artifact. List what you tried, why it didn't work, what the alternative was. Future FDE coming back to this deployment in a year will save weeks.

Customer-team training

The artifacts are necessary; the training is what makes them stick.

Two formats

• Pair sessions. Work through a real task together. The customer's analyst drives; you coach. 60–90 minutes. Do at least three across the deployment.
• Office hours. Optional weekly hour in the last 4 weeks of deployment. Customer's team brings what they want to discuss. Builds confidence; surfaces gaps in the docs.

What to cover

• How the pipeline runs end-to-end (so they can reason about failures).
• How extraction is configured per class (so they can onboard new classes).
• How the SLA dashboard works (so they can interpret quality alerts).
• How to reprocess a date range (backfill runbook).
• How to file a platform bug (so they can do it without you).
• How to escalate to platform / CS (so they have the path when you're gone).

The signal you're done training

The customer's analyst handles a real incident from start to finish using the runbook, without paging you. If that happens twice, training is done.

On-call transition

Three stages over the last 4–6 weeks of deployment:

Stage 1: FDE-primary, customer-observer

Customer's team joins incident calls. Watches you triage. Asks questions.

Stage 2: Joint, customer-driven

Customer's team is primary; uses runbooks. You're on the call but stay back. Step in only when the runbook doesn't resolve.

Stage 3: Customer-primary, FDE backup

Customer's team handles incidents themselves. You're available for platform-team escalations only. Weekly check-in for the first month; monthly thereafter.

The transition criterion

Customer's team resolves the top-5 most-common incidents without paging you. If you've gone through stage 2 for 2–3 incidents and they handled them, move to stage 3.

Customer success transition

CS picks up the ongoing relationship. The transition needs three things:

1. A joint introduction

You + CSM + customer's data lead + executive sponsor on one call. CSM is introduced; their role is explained; the customer leaves with their direct contact for ongoing.

2. A knowledge-transfer doc to CSM

Specifically for CSM: who's who at the customer, what relationship dynamics matter, what risks are open, what expansion conversations are alive, what's been promised, what hasn't.

3. The CSM's first 30 days plan

What CSM should do in their first month with this customer. You write a draft; CSM customizes. Includes who to meet, what to ask, what to confirm.

The graceful exit

Don't disappear. Don't extend forever either.

The 30-day decay

After formal handoff, plan a 30-day decay:

• Week 1: weekly check-in.
• Week 2: weekly check-in if needed; CSM-led.
• Weeks 3–4: monthly check-in starts.
• After day 30: quarterly check-in unless there's an active expansion conversation.

Signal that exit went well

• The customer's data team mentioned a new analyst who joined and onboarded using your docs alone.
• An incident happened and they handled it from the runbooks.
• The CSM-led monthly review continues without you.
• The customer asks for an expansion conversation, not for you to come back and operate.

Final handoff checklist

# Handoff Checklist — [Customer]

## Artifacts (all in customer-accessible location)
- [ ] Deployment overview doc — current, comprehensive
- [ ] Six runbooks — written for their analysts, tested
- [ ] Data dictionary for all mart tables
- [ ] SLA reproducibility kit (gold set, scoring scripts, history)
- [ ] Customer-specific configuration documented + version-controlled
- [ ] "What we tried that didn't work" file
- [ ] Contact list: customer-side owners + our team contacts

## Training
- [ ] At least 3 pair sessions completed with customer's analysts
- [ ] Office hours run for at least 4 weeks
- [ ] Customer's team has resolved at least 2 real incidents
      from runbooks without paging FDE

## On-call
- [ ] Stage 1 done: customer observed FDE handling incidents
- [ ] Stage 2 done: customer drove incident response, FDE backup
- [ ] Stage 3 active: customer primary, FDE for platform escalations only
- [ ] On-call rotation documented in their tooling

## CS transition
- [ ] Joint introduction call held (FDE + CSM + customer)
- [ ] Knowledge-transfer doc delivered to CSM
- [ ] CSM's first 30 days plan drafted and reviewed
- [ ] CSM has access to all customer-facing artifacts

## Commercial
- [ ] Renewal date noted; CSM owns
- [ ] Expansion opportunities documented and handed to AE
- [ ] Any open commercial commitments listed

## Internal
- [ ] Deployment learnings written up (for next FDE on similar deployment)
- [ ] Reusable artifacts contributed back (templates, runbooks, configs)
- [ ] Post-deployment retro held (30 min, with FDE + platform PM + AE)

## Customer-perceived
- [ ] Customer's data lead has confirmed in writing that
      handoff is complete and they have what they need
- [ ] Executive sponsor knows the new contact structure