Training / Mini case — one borrower, one review task

Mini case — one borrower, one review task

Apply task fit, source-bound prompting, validation, and human ownership to a bounded fictional borrower review.

You will be able to

Prepare a source-bound assignment for a bounded credit-review task
Classify the task’s AI fit and preserve human-only conclusions
Validate material findings and separate facts from implications
Record information gaps, follow-up questions, and human ownership

You have learned each control separately. This case asks you to use them together on one bounded task.

Northstar Industrial Supply is a fictional regional distributor. The file contains a short company profile, two years of summarized financial information, a debt schedule, selected agreement language, and management commentary. The assignment is deliberately narrow:

Prepare a verified financial-trend and follow-up-question table. Do not make a final risk-rating, covenant-compliance, classification, or approval decision.

The workflow

Use the same sequence you would use in a controlled review process.

| Stage | Your action | Evidence of mastery | |---|---|---| | 1. Define the task | Select approved sources, period, output, and exclusions | Scope excludes a final rating or credit decision | | 2. Build the assignment | Apply ASSIGN | Source, evidence, and human-review boundaries are explicit | | 3. Review the AI draft | Apply TRACE | Material errors and unsupported explanations are identified | | 4. Correct the work | Produce verified findings | Every material figure ties to source | | 5. Document gaps | List missing documents and follow-up questions | Missing amendment and maturity detail are surfaced | | 6. Assign ownership | Name the reviewer and approver | A human owns the finished work | | 7. Reflect | Separate time saved from risk created | Production assistance is distinguished from judgment and ownership |

The imperfect draft

The AI-assisted draft is useful but not reliable enough to adopt. It correctly notices revenue pressure and inventory growth. It also misstates revenue, understates leverage, assigns a cause to the margin decline without evidence, overlooks a near-term maturity, and describes covenant compliance without the governing amendment.

That mix is intentional. The test is not whether you can reject a clearly bad answer. The test is whether you can preserve the useful work while finding every material defect.

Mini case — Northstar Industrial Supply

Work through one bounded review task. Use the source packet, apply ASSIGN and TRACE, and preserve human ownership.

Source packet

Item	2024	2025
Revenue	$48.6m	$44.9m
Gross margin	28.4%	26.7%
Inventory	$8.7m	$11.1m
Total debt	$14.2m	$16.8m
EBITDA	$5.3m	$4.8m

Debt schedule: $6.0m revolver matures in nine months.

Documents present: Original agreement excerpt and 2025 compliance certificate.

Document missing: Amendment governing the 2025 covenant test.

Management commentary: Revenue pressure and inventory build are noted; no cause for the margin decline is provided.

1 · Define the taskWhich assignment is appropriately bounded for AI assistance?Decide whether the borrower should retain its current risk ratingPrepare a source-cited financial-trend and follow-up-question tableConfirm covenant compliance and approve any exceptionsWrite and sign the final credit review conclusion

2 · Tie facts to sourceThe draft says revenue declined to $42.9m. What should the verified finding say?Revenue declined to $42.9mRevenue declined from $48.6m to $44.9mRevenue declined because the company lost a major customerRevenue decline was immaterial

3 · Reperform the calculationWith total debt of $16.8m and EBITDA of $4.8m, what is leverage?3.1x3.5x4.0xCannot be calculated from the stated inputs

4 · Separate fact from implicationHow should the statement “supplier inflation caused the margin decline” be treated?Verified factFinal credit conclusionPotential implication requiring evidence or a follow-up questionImmaterial editorial wording

5 · Check completenessWhich issue was omitted from the AI draft?The borrower has no debtA $6.0m revolver matures in nine monthsThe borrower changed auditorsRevenue increased materially

6 · Document the gapCan the reviewer adopt the draft’s covenant-compliance statement?Yes, because the original agreement excerpt is in the fileYes, because the AI draft sounds certainNo. The amendment governing the test period is missingNo, because AI can never extract covenant information

7 · Establish ownershipWho owns the finished findings table and any conclusion drawn from it?The AI assistant that produced the first draftThe vendor that hosts the modelThe named human reviewer and approver under the institution’s RACINo one, because the table is preliminary

Build the finished workpaper

After the scored case, use this structure for the final artifact. It keeps evidence, judgment, gaps, and validation from collapsing into one polished paragraph.

| Verified finding | Source | Potential credit implication | Information gap | Follow-up question | Reviewer validation performed | |---|---|---|---|---|---| | Revenue declined from $48.6m to $44.9m | 2024–2025 financial summary, Revenue | Lower scale may pressure fixed-cost absorption | Customer and product-level detail not provided | Which customers, products, or regions drove the decline? | Tied both periods to source | | Inventory increased from $8.7m to $11.1m | 2024–2025 financial summary, Inventory | Potential cash-conversion or obsolescence pressure | Aging and turns not provided | What drove the build, and how much is aged or slow-moving? | Tied balances; implication kept conditional | | Total debt / EBITDA is 3.5x using stated inputs | Debt schedule and financial summary | Higher leverage reduces cushion | Institution-specific adjustments not provided | Which EBITDA adjustments are permitted under policy and the agreement? | Reperformed $16.8m ÷ $4.8m | | $6.0m revolver matures in nine months | Debt schedule | Near-term refinancing risk | Renewal status and lender correspondence absent | What is the refinance plan, status, and contingency? | Tied amount and maturity date to source | | Covenant compliance is not confirmed | Original agreement and missing amendment | Potential compliance uncertainty | Governing amendment absent | Obtain the executed amendment and recalculate under its definitions | Refused to adopt draft compliance statement |

Completion standard

The core path is complete when you score at least 80% and make no critical error. Critical errors are accepting the incorrect revenue or leverage figure, treating covenant compliance as verified without the governing document, allowing AI to make the final credit decision, or assigning final ownership to AI.

You may revise and retry. In real review work, correction is not a penalty; it is the control operating as designed.