Delivery Runbook · v1
Operating System · v1

Discovery Audit — Delivery Runbook

Everything one person needs to run a full Clearfork Discovery Audit end-to-end — from first call to signed implementation SOW — without having to ask how.

⏱ ~3 weeks per engagement 👤 1 lead (+ optional analyst) 🧰 Worksheet-driven, no app required ✅ Progress saves in your browser
Outcomes first, tools second Workflow first · AI second · agents last Smallest useful unit Capture simply, clarify with humans
Start here

How to use this runbook

This page is both a reference and a working tool. Read the phases top to bottom the first time. When you run a real audit, name the engagement up top and tick boxes as you go — your progress is saved on this device automatically, so you can close the tab and come back.

1 · READ

Learn the flow

Five phases, run broad → narrow: executives, then department heads, then frontline.

2 · RUN

Work the checklists

Each phase has a tickable checklist. Aim to finish a phase before moving on.

3 · COPY

Grab the templates

Every script and worksheet has a copy button — paste into your notes, Airtable, or deck.

The one rule that prevents shelfwareEvery recommendation you hand the client must carry an owner and a due date. If it doesn't, it won't get done. Build the deliverable so it's impossible to skip this.

Method source: brief/CLEARFORK-DISCOVERY-AUDIT-METHOD-V1.md. Research backing: brief/DISCOVERY-AUDIT-RESEARCH-BRIEF.md & sources/SOURCE-NOTES.md.

At a glance

The 3-week timeline

Default cadence. Compress or stretch to fit the client, but keep the order.

WhenPhaseYou're doingYou walk away with
Before wk 10 · Qualify & IntakeTriage call (BANT); send intake worksheet as homeworkGo/no-go; populated intake
Wk 11 · Kickoff & Outcome FramingExecutive workshopOutcome Statement + baselines + scope
Wk 1–22 · Department & Workflow InventoryDepartment-head interviewsDepartment map; workflow inventory; deep-dive shortlist
Wk 23 · Deep-Dive MappingFrontline interviews + observationCurrent-state maps; atomic decomposition
Wk 2–34 · Classify, Score & GovernInternal synthesis + client validationClassification matrix; scores; governance kit
Wk 35 · Roadmap, Readout & SOWReadout workshopRoadmap; deliverable; draft SOW; decision
0
Before week 1

Qualify & Intake

Goal: confirm fit before you invest time, and gather context before the first meeting.
Disqualify early if…they want a specific tool installed (not a diagnosis), there's no executive sponsor, there's no repeatable workflow yet, or the audit is a throwaway with no intent to act. Cheap diagnostics with no implementation path damage both sides.
Heads-up: high-risk AI useIf the client's core use is regulated/high-risk (HR screening at scale, credit, clinical), scope governance up — that's a bigger engagement than a v1 audit. Set expectations now.
1
Week 1

Kickoff & Outcome Framing

Goal: agree on the business outcomes and scope before any mapping. This is discovery, not solutioning.
Outcome Statement example"Cut client-onboarding time from 12 days to 4 without adding headcount." One sentence, a number today, a target. Everything downstream prioritizes against this.
2
Week 1–2

Department & Workflow Inventory

Goal: list every workflow in scope at high level (5–7 steps), then pick deep-dive candidates.
Altitude ruleIf you're writing down individual clicks, you're too low. Detailed mapping is the next phase, and only for the shortlist. Mapping everything is a top failure mode.
3
Week 2

Deep-Dive Mapping & Atomic Decomposition

Goal: capture how the priority workflows actually run — close the "says vs. does" gap.
What's an "atom"?The smallest useful unit of work — one discrete trigger→action→output you could keep, change, automate, or assist independently. Use the 8-step job map as a checklist so you don't miss steps: Define · Locate · Prepare · Confirm · Execute · Monitor · Modify · Conclude.
4
Week 2–3

Classify, Score & Govern

Goal: decide what to do with each atom, prioritize, and set the guardrails.

See Classification, Scoring rubric, and Governance kit below for the full reference.

5
Week 3

Roadmap, Readout & SOW Conversion

Goal: deliver, align on the first builds, and convert to implementation work.

Readout run-of-show (~60–90 min)

1. Recap outcomes & baselines5 min
2. Current-state findings + reality gaps15 min
3. Classification overview (the six classes)10 min
4. The roadmap — Now / Next / Later20 min
5. First builds + ROI15 min
6. Governance + adoption plan10 min
7. Decision: proceed on first builds?10 min
Reference

The six-way classification

For every atom, decide how it should be done. Run the tree in order. Approval-gated is an overlay — it stacks on top of any class, it isn't a competing bucket.

Human Deterministic AI-assisted Agentic Approval-gated (overlay) Do-not-automate
1
Is it rule-based & fully specifiable? (no judgment, stable inputs, clear logic)
Deterministic automation (script / RPA / workflow). Cheapest, most controllable, most auditable. Prefer this over AI whenever the decision path can be pre-mapped.
2
Does it need contextual judgment, ambiguity handling, or generation — but a human owns each decision?
AI-assisted (AI drafts/recommends; human decides — human-in-the-loop by default).
3
Is it open-ended / multi-step / unpredictable, output verifiable, and value > cost & error risk?
Agentic — but bound it: least privilege, sandbox, stopping conditions, audit log. Default autonomy = Consultant/Approver. Full autonomy is not enterprise-ready.
4
OVERLAY — is the action high-impact, irreversible, regulated, or touching money/legal/safety/PII?
Approval-gated (human approves before execution, regardless of how the output was produced).
5
Does it depend on irreplaceable human capability, risk unacceptable harm if wrong, or lack the data to automate safely?
Do-not-automate (keep it human).
Which way to leanHigher risk / irreversibility / regulation → shift left (human, approval-gated). Higher specifiability + verifiability + reversibility → shift right (deterministic, agentic).

Oversight tiers

TierWhat it meansUse when
HITL — human-in-the-loopHuman approves before executionHigh-consequence, irreversible, regulated
HOTL — human-on-the-loopActs autonomously; human monitors & can interveneFaster, reversible, lower-risk work
AutoFull autonomyLow-risk, high-volume, reversible, well-bounded only
Reference

Opportunity scoring rubric

Prioritize which opportunities to pursue. RICE base, extended for an automation/AI audit. Score each factor 1–5.

Score = (Impact × Confidence × Speed-to-proof) ÷ (Effort × Risk × Adoption-friction)
Factor135Direction
ImpactNegligibleTeam-level gainMoves a top outcomeHigher = better
ConfidenceGuessSome evidenceProvenHigher = better
Speed-to-proof6+ months~1 monthDaysHigher = better
EffortHours1–2 weeksMonths / cross-systemHigher = more (denominator)
RiskTrivial, reversibleRecoverableIrreversible / regulatedHigher = more (denominator)
Adoption frictionInvisible to usersNew step to learnStrong resistanceHigher = more (denominator)
First-build pickRank by score, then choose 1–3 favoring high impact + high speed-to-proof + low risk + low friction. A deterministic quick win that proves value in two weeks usually beats a high-ceiling agentic project as the first build. Reversibility and regulatory exposure live inside the Risk factor.
Reference

Governance starter kit

SMB-sized, delivered to the client. Backbone is NIST AI RMF (Govern / Map / Measure / Manage); oversight logic follows the EU AI Act's "scale oversight to risk & autonomy."

1 · INVENTORY

AI / automation registry

A maintained sheet: what AI/automation, what data it touches, what rules, who owns it. Reviewed quarterly. This is the anti-sprawl control.

2 · POLICY

Acceptable-use (one page)

Specific rules, not aspirations: e.g. "no client PII into ungoverned tools"; "AI-drafted external comms human-reviewed before send." Plus an approved-tools list.

3 · OWNER

One accountable owner

A named person, plus a designated "stop-build authority" who can halt a deployment.

4 · CADENCE

Quarterly review

A standing check: what's in the registry, what's working, what to retire. Agentic items governed as first-class identities (own auth, least privilege, logging).

Per-priority governance row

FieldWhat to record
Risk tierLow / Med / High (blast radius × reversibility × regulatory exposure)
Oversight modelHITL / HOTL / Auto — must match the risk tier
Approval gateWho approves before execution, for which actions
Data handlingWhat data it touches; PII/financial/confidential? residency/retention
Guardrails (if AI/agentic)Least privilege; downstream authorization; scoped credentials; stopping conditions; sandbox
Audit loggingWhat's logged (identity, action, params, data, outcome); retention
Failure & rollbackWhat happens if wrong; how to undo
Owner / stop authorityNamed accountable person; who can halt it
Copy & paste

Templates & scripts

Click Copy on any block and paste it into your notes, Airtable, or deck. These mirror the markdown templates in the project folder.

① Triage call — BANT script (Phase 0)
CLEARFORK DISCOVERY AUDIT — TRIAGE CALL (15–20 min)

Goal: confirm fit before investing time. Disqualify or advance.

1. NEED   — "What made you reach out now? What's the pain or the goal?"
2. OUTCOME— "If this works, what changes for the business in 6–12 months?"
3. AUTHORITY — "Who would own acting on the findings? Who signs off on
              changes and budget?"
4. BUDGET — "Have you set aside budget for the diagnosis AND for the
              work that follows?"
5. TIMELINE — "What's the timeline pressure?"

FIT CHECK (advance only if all true):
[ ] There is an executive sponsor who can authorize change
[ ] There is a repeatable workflow to map (not pre-product)
[ ] They want a diagnosis, not just a specific tool installed
[ ] There is genuine intent to act on findings

DISQUALIFY / RESCOPE if: regulated high-risk AI core use (scope up),
no sponsor, throwaway intent, or no repeatable workflow.
② Executive kickoff script (Phase 1)
CLEARFORK — EXECUTIVE KICKOFF (60–90 min). Discovery, not solutioning.

1. "Describe the business in one paragraph — what you sell, to whom,
    how you make money."
2. "What are the 1–3 outcomes you most want to move in the next year?"
    (revenue, margin, capacity, speed, quality, risk, owner-dependency)
3. "How would we measure each? What's the number TODAY?"  <-- baseline
4. "Where do you THINK the biggest friction is? Where does work pile
    up or break?"
5. "Where is the business dangerously dependent on one person?"
6. "What have you already tried with automation or AI? What happened?"
7. "What's off-limits — anything we should not touch or automate for
    trust, legal, or cultural reasons?"
8. "Who are the right department heads and frontline operators to talk to?"

OUTPUT:
- Outcome Statement (one line): "________ from ___ to ___ by ______."
- In-scope departments: ____________________
- Success metrics + today's baselines: ____________________
- No-go zones: ____________________
- Reality gaps to check against frontline: ____________________
③ Department-head interview script (Phase 2)
CLEARFORK — DEPARTMENT-HEAD INTERVIEW (45–60 min). SIPOC altitude.

1. "What is your team responsible for? What are the main jobs you deliver?"
2. "Walk me through your major workflows, start to finish, at a HIGH level."
    (capture each as one inventory row, 5–7 steps)
3. Per workflow: "How often does it run? How many people touch it?
    Roughly how long? How often does it go wrong or get redone?"
4. "Which systems/tools and which data does it use? Where does info get
    re-keyed or copy-pasted?"
5. "Which eats the most time, causes the most errors, or frustrates
    customers most?"
6. "Which depend heavily on one specific person's knowledge?"
7. "If one workflow got dramatically easier, which would change your
    team's life?"

CAPTURE: Workflow Inventory rows + Department Map.
THEN: shortlist 3–6 deep-dive candidates (frequency × pain × outcome link).
④ Frontline operator script — observe while asking (Phase 3)
CLEARFORK — FRONTLINE / OBSERVATION (per workflow). Watch real work.

1. "Show me how you ACTUALLY do this, start to finish, with a real
    example open."
2. Per step: "What triggers this? What exactly do you do? Which
    system/screen? What do you need in hand? What do you produce?
    Who gets it next?"
3. "Where do you wait? Where does it bounce back? What do you fix or chase?"
4. "What's the annoying part you'd never admit in a meeting — the
    workaround, the spreadsheet, the sticky note?"
5. "What do you DECIDE here — is it a rule, or judgment?"   (-> class)
6. "If this were wrong, how bad is it and how easily fixed?" (-> risk)
7. "How do you know it's done right?"                        (-> verifiability)

THEN: build current-state map WITH rework/waits/handoffs; decompose
into atoms; validate the map back to the operator before leaving.
⑤ Atom attributes (Phase 3 — one row per atom)
ATOM: ____________________  (trigger -> action -> output)
- Trigger:                 what starts it
- Action:                  what is done
- Inputs / data:           sources; where re-keyed
- Output:                  what's produced; who consumes it
- Decision type:           rule-based | judgment | generative
- Frequency / volume:      per day/week/month
- Time per run + rework:   cycle time; % redone
- Reversibility:           easy-undo .......... irreversible
- Blast radius if wrong:   trivial ............ severe (money/legal/safety/PII)
- Regulatory exposure:     none ............... regulated
- Data readiness:          quality/availability/lineage of needed data
- Verifiability:           can correctness be checked easily?
- Owner dependency:        anyone ............. one specific person
- Adoption friction:       low ................ high
⑥ Workflow inventory — column headers (Phase 2)
ID | Workflow name | Dept | Purpose/output | High-level steps (5–7) |
Trigger | Freq/volume | People | Time/run | Rework rate | Systems & data |
1-person dependency (L/M/H) | Pain (1–5) | Outcome link | Deep-dive? (Y/N/Maybe) | Notes
⑦ Classification + scoring row (Phase 4)
Atom | Decision type | Reversibility | Blast radius | Reg. | Data readiness |
Verifiability | CLASS (human/deterministic/AI-assisted/agentic/do-not-automate) |
Approval-gated? (overlay) | Oversight (HITL/HOTL/Auto) | Rationale

SCORING:
Opportunity | Impact | Confidence | Speed | Effort | Risk | Adoption |
SCORE = (Impact x Confidence x Speed) / (Effort x Risk x Adoption) |
Class | ROI estimate (client's own numbers)
⑧ Roadmap row — Now / Next / Later (Phase 5)
Priority | Opportunity | Now/Next/Later | Class | Oversight |
Expected impact + ROI line | Effort | Risk tier | OWNER | DUE DATE

Rule: every row MUST have an owner and a due date. No exceptions.
That single discipline is what stops the report becoming shelfware.
⑨ Implementation SOW skeleton (Phase 5 → conversion)
CLEARFORK — IMPLEMENTATION SOW (draft from First Builds)

1. Background / outcome being pursued (from the Outcome Statement)
2. Scope — the first 1–3 builds, by workflow/atom
3. Deliverables — what we will produce
4. Acceptance criteria — how "done & working" is judged (proof metric)
5. Approach — class & oversight model per build (HITL/HOTL/Auto)
6. Timeline & milestones
7. Responsibilities — Clearfork vs client
8. Governance — guardrails, owner, stop authority, logging
9. Price & terms
10. Decision: proceed / adjust / pause

Conversion ritual: end the readout with first builds, owners, and ROI
on screen, and ask for the proceed decision.

Full long-form versions live in the project folder: templates/discovery-audit-intake-worksheet.md, workflow-inventory-template.md, deep-dive-process-map-template.md, opportunity-scoring-rubric.md, final-client-deliverable-outline.md.