artifacts/standard-named

Rendered from markdown source. Open raw source on GitHub.

Minimum Viable Continuity Playbook

Outsourced Accounting Firms (AI‑Capable Edition)

Continuity is the capability to preserve intent, consent, and legibility through change—without turning the firm into a bureaucracy.

Outsourced accounting firms sit at the highest‑risk intersection of:

• multi‑tenant client data
• regulated financial workflows
• historical exceptions
• human + AI collaboration

This playbook defines the minimum viable structures required to safely operate AI across clients, providers, and time.

---

0. Firm Scope Boundary

Declare what this playbook governs.

Applies to:

• All client financial data
• All AI‑assisted analysis, summaries, classifications, or recommendations
• All external tools or providers touching client books

Explicitly excludes:

• Personal productivity tools with no client data
• One‑off experiments until promoted

This prevents accidental cross‑client leakage and shadow‑AI production use.

---

1. Client‑Bound Intent Registry (CBIR)

Purpose: Make why work is performed legible per client.

Each client has explicit intents, not just services.

Minimum fields per intent:

• Client ID
• Intent ID
• Plain‑language purpose
• Authorized outputs
• Prohibited uses
• Regulatory sensitivity (tax, payroll, audit‑adjacent, advisory)

Example:

Client: ACME Co
INT‑FIN‑MONTHLY‑001
Purpose: Monthly management P&L summary
Authorized: Internal advisory conversations
Prohibited: Tax filing, lender reporting
Sensitivity: Medium

Same data, different intent ≠ same permission.

---

2. Client Data Domain Map (CDDM)

Purpose: Prevent semantic bleed across clients and contexts.

Core accounting domains (minimum):

• General Ledger
• Revenue
• Expenses
• Payroll
• Taxes
• Forecasts / Projections

For each domain:

• System of record
• Client‑specific semantics
• Known ambiguities
• Allowed AI operations

Example:

Domain: Revenue
Client nuance: Deferred revenue tracked manually
Ambiguity: Cash vs accrual views diverge

Declare disagreement early so AI doesn’t invent agreement.

---

3. AI Client Isolation Matrix (ACIM)

Purpose: Ensure AI never crosses client boundaries.

For each AI system or model:

• Client isolation method (hard / soft / none)
• Data domains accessible
• Allowed operations
• Retention policy
• Logging requirement

Example:

Model: GPT‑4.1
Client scope: Single‑client sessions only
Access: Read‑only financial summaries
Retention: Disabled
Logging: Required

Multi‑client context windows are a hard stop unless explicitly engineered.

---

4. Provider & Tool Continuity Declaration

Purpose: Extend continuity beyond your firm.

Every tool or provider touching client data must declare:

• Data received
• Transformations performed
• AI usage (model + purpose)
• Artifacts returned
• Treatment of derived data

This includes:

• OCR / receipt tools
• Categorization software
• Forecasting platforms
• Offshore bookkeeping teams

If they touch client books, they are inside your continuity boundary.

---

5. Advisory vs Bookkeeping Boundary Rule

Purpose: Prevent AI from silently crossing regulatory or liability lines.

Minimum rule:

• Bookkeeping AI: classify, summarize, reconcile
• Advisory AI: scenario analysis, explanations, options
• Tax / audit decisions: human‑only unless explicitly approved

Each AI output must be tagged:

• Informational
• Advisory
• Decision‑support

This protects both licenses and trust.

---

6. Decision Trace Chain (Lightweight)

Purpose: Preserve explainability without audits.

Any AI‑assisted client‑impacting output must be traceable to:

• Client ID
• Intent ID
• Data domain(s)
• Model + version
• Human sponsor

No narrative required. Just a chain.

---

7. Experiment → Client‑Facing Promotion Gate

Purpose: Stop “we tested this once” from becoming firm infrastructure.

Promotion required when an AI tool becomes:

• recurring
• client‑visible
• relied upon for accuracy or speed

Promotion checklist:

• Assigned client intent(s)
• Data domain mapping
• AI isolation confirmation
• Sponsor named

This is where most accounting firms silently fail today.

---

8. Continuity Steward (Fractional)

Purpose: Make continuity owned, not heroic.

Role:

• 2–4 hours/week
• Maintains registries
• Reviews promotions
• Resolves semantic disputes
• Has veto power on unsafe AI use

This is stewardship, not compliance.

---

9. The Client Safety Question

Every AI‑affected workflow must answer:

“Could this output accidentally affect the wrong client, filing, or financial decision?”

If yes → continuity gap.

---

What This Enables

• Safe AI‑augmented bookkeeping
• Scalable advisory services
• Defensible AI use
• Client trust at scale
• Future regulatory readiness without retrofitting

---

Bottom line:

AI doesn’t reduce responsibility in outsourced accounting. It concentrates it.

Continuity is how you hold that responsibility without breaking the firm.