creditboard
Training / The RACI Exhibit

Who is accountable when the machine drafts the work?

A RACI over the credit review lifecycle, drawn for the moment an AI assistant enters the workflow. Read the reference below, then clone it, adapt it to your institution, and take the exhibit to your governance committee.

AI can produce work. Only a human can own a conclusion.

Taught mode — the Creditboard reference

CRAA cannot be Accountable. The machine can draft the work; it cannot own the conclusion. Accountability for a credit judgment rests with a named human. Every row below has exactly one Accountable role, and it is always a named human. Rows marked human only are judgment tasks: the machine holds no role in them at all. Each row carries the reasoning, the verification tier the task demands, and what goes wrong in an examination if the line is drawn elsewhere.

Verification tiers

Tier 1 — Full recalculation. Every AI-produced figure independently recomputed against source.

Tier 2 — Traced sample. A defined sample of AI figures traced to source, with a stated expansion trigger.

Tier 3 — Editorial review. AI output is prose with no load-bearing figures; reviewer adopts it as their own words.

Tier 0 — Independent. AI output not used; human performs from source.

The tier is a property of the task, not of how good the model looked today.

Builder mode — adapt it to your institution

Edit any assignment. Add your own tasks. One cell is locked, and it stays locked: the machine’s column under Accountable. Export the result as a policy exhibit — CSV, XLSX, or a print-clean PDF headed with your institution’s name.

Credit Review Function — RACI: AI in the Credit Review Lifecycle
Adapted from the Creditboard reference · creditboard.org/training/raci · 2026-07-12
TaskReviewerReview ManagerHead of Credit ReviewCRAA (AI)
never Accountable
Model Risk / ValidationBusiness Line / Originator
Scoping
Define review scope and objectives
Assess portfolio risk and set review frequency
Sampling
Design sampling methodology
Select the sample
Intake
Extract data from credit files and scanned documents
Analysis
Spread financial statements
Build the EBITDA bridge
Flag quality-of-earnings anomalies
Test covenant compliance
Recalculate DSCR / LTV / borrowing base
Assess collateral valuation adequacy
Evaluate guarantor support
Conclusion
Recommend a risk rating
Assign the final risk ratingHuman only
Determine accrual status / classificationHuman only
Conclude on adequacy of the ACL / allowanceHuman only
Documentation
Draft workpaper narrative
Cite evidence to source documents
Draft the issue / finding
Approve the issue and its severityHuman only
QA
Second-level review of workpapers
QA sample of AI-generated content
Reporting
Draft the review report
Sign off and issue the reportHuman only
Present to Credit Risk Review Committee / Board
Governance
Approve prompts and model version in production
Respond to examiner inquiry on methodology
Matrix is valid: every task has exactly one Accountable human, and the machine owns no conclusion.