# AI & Your Work

## Examine the Work Before You Automate It

### Participant Workbook · 90-minute flagship masterclass

Use this workbook with the Aloha AI flagship masterclass. Hold one recurring piece of work privately in mind or use the fictional Kōkua Studio case. Do not record confidential, privileged, proprietary, personal, health, financial, employee, client, tax, payment, credential, or security-sensitive information.

This workbook supports general education. It does not recommend a vendor, authorize implementation, certify readiness or compliance, or provide individualized legal, tax, security, clinical, accounting, employment, financial, or investment advice.

---

# Part 1 — Define

## Translate the tool request into a work problem

**Tool request or proposed technology:**

**Recurring work beneath the request:**

**Desired outcome:**

**Constraint that cannot be violated:**

**Accountable owner:**

**People doing, receiving, reviewing, or repairing the work:**

Complete this statement:

> We need to decide how to improve **[recurring work]** for **[affected people]**, owned by **[role]**, so that **[observable outcome]** improves while preserving **[constraint, right, quality, or relationship]**.

If the statement still names a product, rewrite it.

---

# Part 2 — Examine

## Opportunity Map

### 1. Friction

Friction is repeated loss of time, attention, money, capacity, or trust. It is a signal—not automatically the cause.

- Where is the visible delay, repetition, rework, error, or frustration?
- Who experiences it?
- How often does it occur?
- What evidence establishes it?
- What might actually cause it?
- Which current strengths must not be destroyed?

### 2. Inputs

Correct work depends on information that is available, current, complete, authoritative, and permitted for the intended use.

| Required input | Current source | Authority/owner | Current? | Permitted? | Missing or conflicting? |
|---|---|---|---|---|---|
|  |  |  |  |  |  |
|  |  |  |  |  |  |
|  |  |  |  |  |  |

### 3. Judgment

Mark every interpretation, approval, empathy, negotiation, taste, professional duty, or responsibility a person must own.

| Judgment | Current owner | Evidence used | Consequence if wrong | May AI assist? | Must remain human-owned? |
|---|---|---|---|---|---|
|  |  |  |  |  |  |
|  |  |  |  |  |  |

### 4. Risk

Design how failure will be detected—not merely how it might be prevented.

| Failure | Person/information affected | Detection method | Response owner | Stop trigger | Repair/fallback |
|---|---|---|---|---|---|
|  |  |  |  |  |  |
|  |  |  |  |  |  |

### 5. Value

Define the measure before a demonstration can rewrite the standard.

- Current baseline:
- Intended improvement:
- Measurement method:
- Review period and sample:
- Guardrail against a worse substantive outcome:
- Correction/review workload:
- Evidence that would show the intervention is not worthwhile:

---

# Part 3 — Make the current system visible

Follow one representative case from trigger to destination. Mark judgment **J**, exception **E**, handoff **H**, repair **R**, and evidence gap **?**.

| Step | Person/system | Input | Action or judgment | Output/handoff | J/E/H/R/? |
|---|---|---|---|---|---|
| Trigger |  |  |  |  |  |
| 2 |  |  |  |  |  |
| 3 |  |  |  |  |  |
| 4 |  |  |  |  |  |
| 5 |  |  |  |  |  |
| Destination |  |  |  |  |  |

## Evidence ledger

Label each claim **measured**, **observed**, **reported**, **inferred**, or **unknown**.

| Claim | Status | Source/date | What it supports | Limitation | What would verify it |
|---|---|---|---|---|---|
|  |  |  |  |  |  |
|  |  |  |  |  |  |
|  |  |  |  |  |  |

---

# Prepared fictional case — Kōkua Studio

Kōkua Studio is a fictional composite, not a client story or performance claim.

Two owners provide customized services in Hawaiʻi. Each week, they spend approximately six owner-hours apiece turning discovery notes into proposals. Notes are scattered. Service details, current prices, availability, prior language, and client-specific fit must be checked. One owner drafts; another reviews before delivery.

Important judgments include client fit, scope, price, promises, exceptions, and final responsibility. Important risks include confidential notes entering an inappropriate service, invented deliverables, stale prices, omitted limitations, and confident-sounding errors.

## Your initial view

- What is the decision—not the tool request?
- What appears to be friction?
- What might be the underlying cause?
- Which inputs require authority or cleanup?
- Which judgments remain human-owned?
- What baseline and guardrail would you define?
- Which of the four findings would you choose now?

## Prepared disposition

Assuming the inputs can be stabilized, the lowest responsible AI-related intervention is a bounded experiment:

- ten fabricated or deliberately redacted historical cases;
- one current, approved source packet;
- draft-only output;
- owner review of fit, scope, price, promises, missing facts, exceptions, and prohibited language;
- comparison with baseline time, turnaround, and correction burden;
- current process remains available;
- stop if confidential material enters the system;
- stop if an unapproved promise survives review; and
- stop if correction burden eliminates the time benefit.

---

# Part 4 — Decide

## Four legitimate findings

Choose one. Do not treat AI adoption as the successful answer by default.

- [ ] **Bounded experiment:** enough evidence and control exist to test one narrow claim reversibly.
- [ ] **Redesign first:** ownership, sources, permissions, steps, or handoffs are not stable enough to test AI responsibly.
- [ ] **Keep human-led:** relationship, craft, care, discretion, duty, or accountability is central to the value of the work.
- [ ] **Insufficient information:** the responsible next step is to gather evidence rather than manufacture certainty.

**Finding:**

**Evidence supporting it:**

**Contradictory evidence or dissent:**

**What could change the finding:**

## Intervention ladder

Choose the lowest sufficient intervention capable of producing the required result.

1. Clarify ownership.
2. Improve a template or checklist.
3. Use ordinary rules-based automation.
4. Add one AI-assisted step with review.
5. Create an integrated workflow.
6. Build a custom system.

**Selected rung and why lower rungs are insufficient:**

**Setup, testing, training, maintenance, security, and exit work introduced:**

---

# Part 5 — Build the smallest next action

## Bounded experiment specification

Complete only if your finding supports an experiment.

| Element | Definition |
|---|---|
| Claim being tested |  |
| Users and task |  |
| Explicit exclusions |  |
| Permitted/prohibited inputs |  |
| Approved references |  |
| Structured instruction |  |
| Draft/output boundary |  |
| Human review checklist |  |
| Decision owner |  |
| Destination |  |
| Evidence log |  |
| Baseline and measure |  |
| Guardrail |  |
| Duration/sample |  |
| Stop rule |  |
| Rollback/fallback |  |

## Testable human review

“A person checks it” is not a control.

- Reviewer:
- Evidence and checklist:
- Authority to reject or stop:
- Review capacity and response time:
- Escalation path:
- Record produced:

## Stop rule

> Pause **[scope]** within **[time]** when **[event or metric]** reaches **[threshold]** during **[sample/window]**. **[owner]** determines whether the work stops, is repaired, or is separately authorized to resume. Until then, **[fallback]** continues.

---

# Monday Plan

Do these before buying a platform:

1. Document one recurring process exactly as it happens.
2. Mark every point containing sensitive inputs or human judgment.
3. Define one reversible improvement and its measure.

| First action | Owner | Due date | Evidence produced | Review point |
|---|---|---|---|---|
|  |  |  |  |  |

- Evidence required to advance:
- Evidence or event requiring pause/stop:
- People who must review or participate:
- Next review date:

---

# Completion check

- [ ] I translated the tool request into a work problem.
- [ ] I mapped friction without assuming it was the cause.
- [ ] I identified the inputs, authorities, permissions, and conflicts.
- [ ] I preserved human-owned judgments and responsibilities.
- [ ] I designed detection, response, fallback, and stop rules.
- [ ] I defined value and guardrails before a test.
- [ ] I considered all four legitimate findings.
- [ ] I chose the lowest sufficient intervention.
- [ ] Any experiment is bounded, reversible, reviewable, and measured.
- [ ] My Monday Plan names an action, owner, evidence, and review date.

Completion demonstrates that you recorded a decision method. It does not establish that the evidence is correct, the tool is appropriate, a test is authorized, legal or professional requirements are satisfied, or implementation is safe.