> `YOUR GRACE AGENCY  /  AUTOMATION LAUNCH KIT  /  ROLLOUT  /  09 OF 11`
>
> **AUTOMATE. DELIVER. EMPOWER.**
> *Build deliberately. Observe closely. Let evidence decide the next iteration.*

# 7-Day Automation Rollout Guide

**A deliberate path from one process to a trusted v1.** Use this plan with one template, not the entire kit at once.

| Rhythm | Focus | Proof to keep |
| --- | --- | --- |
| Days 1–2 | define the process and minimum record | owner, trigger, field map, baseline |
| Days 3–5 | build and test a narrow v1 | test results, exceptions, rollback notes |
| Days 6–7 | observe limited use and decide | review log, metric comparison, next action |

> **ROLLOUT PRINCIPLE**
> Limited live launch is not permission to remove safeguards. Keep the approval path, review the output daily, and pause when evidence says to pause.

## Before day 1

Choose a workflow from `01-workflow-templates/`, assign an owner, and set aside 45–60 minutes each day. You need access to the workflow’s source, destination, and a safe test space. No new software purchase is required to use this guide.

## Day 1 — Select and define

**Goal:** choose a narrow, measurable first workflow.

1. List three recurring processes using Prompt 1.
2. Choose the one with a clear trigger and manageable consequence if it pauses.
3. Complete sections 1 and 2 of the setup checklist.
4. Write a baseline: current response time, number of handoffs, manual minutes, or error count.

**Deliverable:** a one-sentence workflow outcome, owner, trigger, destination, and baseline.

**Do not proceed if:** the trigger depends on a judgment call, you have no owner, or an unsafe external action is required.

## Day 2 — Build the record and source of truth

**Goal:** make the data clear before adding automation.

1. Use Prompt 4 to produce a small field schema.
2. Create or update the destination table, CRM view, board, or folder template.
3. Decide how a duplicate will be recognized.
4. Mark data that must not enter notifications or AI prompts.

**Deliverable:** one source-to-destination field map and a duplicate rule.

**QA:** Enter a fictional record manually and confirm every required field is visible to the owner.

## Day 3 — Map the happy path

**Goal:** turn the template into a minimum viable sequence.

1. Translate the selected template into your platform’s trigger, search, condition, create/update, and notify actions.
2. Keep only the happy path plus one manual-review route.
3. Name the workflow and each action clearly.
4. Use a test label or test destination.

**Deliverable:** a switched-off or test-only v1 with no external auto-send action.

**QA:** Review each action against the original template. Remove any extra step that does not directly serve the outcome.

## Day 4 — Add safety and failure handling

**Goal:** prevent quiet errors and accidental repeats.

1. Use Prompt 5 to identify failure modes.
2. Add missing-field handling, duplicate handling, and a failure notification.
3. Write the rollback steps where the owner can find them.
4. Set human approval for external communications and sensitive decisions.

**Deliverable:** a documented failure path and pause procedure.

**QA:** Disable the test destination once to confirm the owner sees a useful failure signal rather than a silent error.

## Day 5 — Test with evidence

**Goal:** prove the workflow handles normal and awkward records.

1. Use Prompt 16 to create 10 fictional test records.
2. Run them one at a time.
3. Use Prompt 17 to compare actual to expected results.
4. Fix blockers and rerun every affected case.

**Deliverable:** a completed test table with notes, screenshots, or run IDs.

**Launch gate:** Do not launch if you see duplicate creation, lost data, a wrong owner, or an external action that can happen without intended review.

## Day 6 — Limited launch

**Goal:** process a small amount of live work under close review.

1. Turn on the trigger only for a narrow segment or manually selected records.
2. Review every output before the end of the business day.
3. Log each exception, edit, and manual intervention.
4. Pause immediately if a record reaches the wrong destination, repeats an external action, or exposes data.

**Deliverable:** a limited-launch log: count processed, exceptions, owner time, and customer-impact notes.

**Expected outcome:** You learn from real work without turning a small mistake into a broad one.

## Day 7 — Review and decide

**Goal:** make an evidence-based next move.

1. Use Prompt 22 to summarize the first week.
2. Compare the baseline with actual outcomes; avoid claiming improvement from a tiny sample.
3. Decide `keep`, `adjust`, or `pause`.
4. If keeping, document the v1 as an SOP with Prompt 21 and set a 30-day review.

**Deliverable:** a short owner handoff using Prompt 25, plus one prioritized improvement.

## Daily review questions

- Did every record end in the expected place with the expected owner?
- Did a duplicate, missing value, or delayed event change the result?
- Did the automation create a draft or take an action beyond the intended boundary?
- Could a teammate understand why the workflow made each decision?
- Is the workflow saving useful time without adding hidden review work?

## Troubleshooting guide

| Signal | Likely cause | First response |
| --- | --- | --- |
| No workflow run | trigger mismatch or source permission | create one test record; verify fields and connection access |
| Duplicate record/task | repeated event or missing lookup | pause external actions; add a search/upsert step |
| Wrong routing | labels too broad or missing data | tighten allowed labels; route uncertain cases to review |
| Empty/garbled AI output | prompt lacks fixed format or source fields | require JSON/headings; validate output before use |
| Client message sent too early | approval boundary was skipped | pause workflow; review logs; restore a draft-only step |
| Owner never sees exception | notification target/rule incorrect | test failure path with owner present |

## Definition of a healthy v1

A healthy v1 is boring in the best way: it routes known work reliably, surfaces unknown work visibly, keeps consequential judgment human, and leaves enough evidence for the owner to improve it next month.

---

`YOUR GRACE AGENCY  ·  ROLLOUT / 7 DAYS  ·  TEST FIRST, THEN LAUNCH`