# 30-60-90 rollout plan
*Module 12, Lesson 1. Fix the data before automating anything that depends on it.*

## Days 1–30 — measure, do not deploy
- [ ] Clean stage data: every candidate in the right stage
- [ ] Every rejection has a substantive reason (not "not a fit")
- [ ] Run the funnel once by hand — you need a before picture
- [ ] Build the system inventory (expect to find a tool nobody listed)
- [ ] Run the 25-resume parse audit on your chosen tool
- [ ] Pick **one** pilot role family: repeated, enough volume, low stakes

## Days 31–60 — one role family, fully
- [ ] Intake and scorecard on the pilot family
- [ ] Screening with explicit criteria + disagreement check every run
- [ ] Structured interview loop with written anchors
- [ ] Candidate notice published **before** the first automated stage runs
- [ ] Baseline bias audit — the number you will compare against

## Days 61–90 — connect and extend
- [ ] Screening-to-scheduling handoff with one human checkpoint
- [ ] Weekly dashboard live
- [ ] Second role family
- [ ] Review: funnel change, what broke, what the disagreement checks caught

## Do not touch in month one
| Leave alone | Why |
|---|---|
| Automated rejection | Irreversible, candidate-facing, no calibration yet |
| Candidate-facing messaging automation | Errors are public and unrecallable |
| Learned ranking | You have not audited what it learned from |
| Async video assessment | Highest regulatory exposure in the stack |
| Your whole req load at once | Nothing to compare against, nowhere to retreat |

## Measure the rollout
| Measure | Baseline | Day 30 | Day 60 | Day 90 |
|---|---|---|---|---|
| Coverage — % of applications actually reviewed | | | | |
| Days in stage: screen → first interview | | | | |
| Disagreement rate at the check | | | | |
| Screen-in rate | | | | |
| Interview-to-offer rate | | | | |
| Bias audit vs baseline | | | | |

**Screen-in rate dropping sharply after deployment usually means over-filtering, not
better filtering.** It is easiest to catch in the first month.

## Stopping rule
*Decide now what would make you turn it off.*

We will pause the rollout if: ______________________________________________

---
## Hiring manager one-pager
*Everything a hiring manager needs. If it is longer than this, it will not be read.*

1. **The criteria** are the ones you gave me in intake. Here they are: [link]
2. **Read bands, not ranks.** The difference between #6 and #7 is noise. The difference
   between "strong match" and "weak" is not.
3. **If you disagree**, say so and I will record it as an override with your reason.
   Overrides are information, not failures.
4. **What you produce:** scores against the anchors, submitted before the debrief.