# Expert pilot worksheet

Use one dossier throughout all eight chapters. Below are a fictional completed starting version and a blank version to copy. All Noor data are practice material. Expert here concerns managing business use; this document is not programming certification or approval to use real data.

You do not need your own business. Take turns playing Noor (decision-maker), Alex (user) and Sam (critical reviewer). Record their objections and assess the rubric in a separate second reading. Call this self-assessment; a fictional role is not a genuinely independent expert.

## Completed starting version — fictional plan, no measured results yet

**Dossier:** NOOR-A, version 1. **Owner:** Noor. **Question:** can reviewed reply preparation reduce workload without incorrect promises? **Scope:** general product questions using only fictional question text and an approved fact sheet. **Prohibited:** customer files, invented product claims, price agreements, sending and autonomous external actions.

**Status key:** A = assumption; P = planned; O = your own recorded observation; G = given fictional practice result. Date each later change and add its reason and evidence reference. Never silently turn an assumption into a measurement.

| Chapter | Data and work product | Evidence to keep | Owner | Decision point |
|---|---|---|---|---|
| H1 Strategy | A: 300 questions/month; hypothesis of 12 → 7 minutes including review. Option A: draft replies; B: autonomous quotations; C: newsletter. Provisionally choose A; exclude B. | 01-choice-card: goal, three candidates, exclusion reasons, alternatives and assumptions. | Noor | Day 15: does the trial fit the problem, capacity and permitted scope? |
| H2 Data and choice | P: only product code, general question and approved fact sheet. Name, email, order and medical explanation are unnecessary. Start with manual chat use. | 02-data-flow: field/purpose/necessity/access/retention; four risks; choice with alternative and reassessment trigger. Real provider and processing not yet approved. | Noor as data owner | Before first input; again for real data, a different service or changed scope. |
| H3 Prompt and tests | P: source FICHE-2 version 1; concept/escalatie; human review. Start with the small v1/v2 workshop. | 03-prompt-v1, 03-prompt-v2, reason for the change and your own complete test log. Keep H3 tests separate from pilot cohorts D30/D60. | Alex tests; Noor assesses release | Before a limited trial and after a prompt, source or model change. |
| H4 Research | P: compare the fictional delivery excerpts by product, region and date; do not assume a market standard. | 04-source-register and conflict note: claim, source, scope, uncertainty and decision implications. | Sam as critical reviewer | Before a research claim affects the pilot choice or product text. |
| H5 Commercial work | P: assess fictional claims and a success measure; send no emails. The marketing trial is a side assignment, not automatic expansion of reply pilot A. | 05-claim-check and experiment card with primary metric, complaint metric, comparison and predefined criteria. | Alex; Noor decides scope | Before publication or a campaign; a new task requires a new decision. |
| H6 Operations | P: source → chat → draft → human review. A possible CRM route remains a simulation. Do not blindly repeat an action when its execution status is unknown. | 06-process-map, permissions/mapping, approval fields and runbook with owner and fallback. | Noor as administrator; Alex as user | Before a connection, during an incident and before resuming. |
| H7 Measurement | A: €1,000 upfront, €250/month, €30/hour. Year-1 TCO €4,000. Measure time, quality, rework and workload together. | 07-measurement-plan, raw measurement records and independently recalculated scenarios. Separate potential capacity, realisable value and cash impact. | Noor; Sam recalculates | On days 30, 60 and 90; do not change criteria after seeing results. |
| H8 People and decision | P: Noor decides/pauses; Alex reviews; cover, available time, training and support still to be completed. | 08-schedule, practice report, incident/change log, rubric assessment and decision note. | Noor | Before the first trial, before resuming and on day 90. |

**Source FICHE-2, practice version 1:** LAMP-2 is not dimmable. The fact sheet provides no delivery time, other models or colours. Do not turn missing information into a promise.

**Fixed exercise thresholds:** zero critical errors; at least 95% fully source-supported drafts; an average of no more than 8 minutes including review; a working manual fallback. Critical means an unauthorised promise or prohibited data. These thresholds were chosen for the case; they are not a universal production standard.

**Provisional business case:** 300 × (12−7)/60 = 25 hours/month; at €30/hour, up to €750 in potential capacity value. Year 1: 12 × €250 + €1,000 = €4,000 in costs. Whether freed-up time will be used productively remains to be demonstrated. No proven cash savings.

## Blank version — copy and complete

**Dossier/version/date:** … **Process owner:** … **Problem:** … **Selected task:** … **Alternative without AI:** … **Within scope:** … **Outside scope/prohibited:** … **Fictional roles or actual available reviewers:** …

| Chapter | Data/assumptions and status A/P/O/G | Work product and evidence reference | Owner + available time | Decision point/condition |
|---|---|---|---|---|
| H1: choice | … | … | … | … |
| H2: data and architecture | … | … | … | … |
| H3: prompt and evaluation | … | … | … | … |
| H4: research | … | … | … | … |
| H5: commercial checks | … | … | … | … |
| H6: operations and recovery | … | … | … | … |
| H7: measurement and finances | … | … | … | … |
| H8: people and decision | … | … | … | … |

If a section is unnecessary for your task, explain why. Record the associated lesson exercise separately; do not broaden the pilot merely to apply every chapter operationally.

### Measurement card — one copy per indicator

**Goal/indicator:** … **Definition and formula:** … **Source/unit:** … **Case count/denominator:** … **Baseline and comparable trial:** … **Period:** … **Missing data:** … **Predefined threshold:** … **Owner/frequency:** … **Confounding changes:** …

### Test log — one copy per run

**Date:** … **Visible model name or unknown:** … **Prompt version + full text:** … **Source version:** … **Test ID/repetition:** … **Exact input:** … **Expected behaviour:** … **Verbatim output received:** … **Format check:** … **Source check:** … **Task boundary:** … **Severity:** … **Human judgement + reason:** … **Improvement action:** …

A new version requires its own runs. Keep failures too. Do not label your own observations G; never copy fictional example results as your own measurements.

### Decision log

| Point | Available evidence | Thresholds met/breached | Decision and reasoning | Owner/next step |
|---|---|---|---|---|
| Day 15: starting conditions | … | … | … | … |
| Day 30: first trial | … | … | … | … |
| Day 60: reassessment | … | … | … | … |
| Day 90: next steps/scaling | Not yet available | Still to be assessed | Do not fill in ahead of time | … |

For each incident: **signal → pause → owner → manual fallback → repair → retest → explicit decision to resume**. Also record what the employee does when the usual reviewer is absent.

### Assessment and revision

Assess against the eight criteria from H8: strategic fit; value and baseline; data and risk; technical design; evaluation; human review; adoption and management; decision quality. For each criterion, record **insufficient/sufficient/strong, a specific evidence reference, shortcoming, repair and second judgement**. Ask a critical fellow learner to review it when available. An AI assessment can suggest a question, but does not prove your execution or authority.

For this learning assignment: every criterion must be at least sufficient, with appropriate evidence; critical errors block release. A sufficient design does not yet establish production readiness. Missing outcomes remain unknown.