Review the evidence signals before interviewing. Then use the anchored descriptions—not instinct alone—to choose the score that best matches each answer.
01
Evaluation factor
Technical command
35% weight
Check depth in three-statement modeling, DCF and variance analysis alongside applied AI: Python or SQL pipelines, LLM APIs, retrieval over 10-K filings, and Copilot or Excel automation.
Evidence to listen for
Commands the instruments, models, or reporting standards the role turns on
Can build the analysis rather than only interpret someone else's
Knows the assumptions inside a model and which ones actually drive the answer
Fluent in the frameworks and disclosure regimes that apply
Five-point scoring guide
1
Poor
Cannot explain the instruments or standards they claim to work with.
2
Needs Improvement
Interprets others' analysis but cannot build or defend it.
3
Satisfactory
Solid working command; thin on unfamiliar structures or standards.
4
Very Good
Builds the analysis and knows which assumptions actually move the answer.
5
Excellent
Builds and defends models in Excel and Python, and names specific AI tooling used on real forecasting or filings work.
02
Evaluation factor
Deals and deliverables that closed
25% weight
Probe deliverables that reached decision-makers: board decks, monthly close packs, budget cycles, valuation memos, or an AI agent that cut reporting turnaround, with named figures.
Evidence to listen for
Names transactions, filings, or reports they worked, with size, counterparties, and their own scope
Distinguishes their contribution from the deal team's
Knows what happened afterwards, including what underperformed
Can describe one that fell over and why
Five-point scoring guide
1
Poor
No completed work; describes process rather than outcomes.
2
Needs Improvement
Involved in transactions but cannot state their own scope.
3
Satisfactory
Real deliverables with adequate ownership; outcomes described loosely.
4
Very Good
Named transactions or filings with clear personal scope and honest post-mortems.
5
Excellent
Cites concrete outputs with dollar or hour impact, such as a close cycle shortened or a model backing a funded investment.
03
Evaluation factor
Risk judgement
25% weight
Test how they treat model risk and hallucination: source citation, reconciliation to the general ledger, audit trails for AI outputs, and when they refuse to automate.
Evidence to listen for
Distinguishes a modelled risk from a real one
States confidence and what would change their view
Comfortable disagreeing with a number that suits everybody
Knows the limits of the data behind a projection, especially over long horizons
Five-point scoring guide
1
Poor
Treats model output as truth; no sense of data limits.
2
Needs Improvement
Reports numbers without qualifying them; avoids unwelcome conclusions.
3
Satisfactory
Reasonable judgement; qualifies findings when prompted.
4
Very Good
States confidence unprompted and will hold an unpopular position on evidence.
5
Excellent
Validates every AI-generated figure against source data and can describe an instance where they overrode or rejected an automated output.
04
Evaluation factor
Explaining it to decision-makers
15% weight
Assess how they translate variance drivers and model assumptions for CFOs, controllers and non-finance leads without leaning on jargon or unexplained AI outputs.
Evidence to listen for
Explains a technical position to an investment committee, board, or regulator so they can act on it
Writes to the standard the audience is held to
Handles challenge without either caving or digging in
Works across legal, operations, and external counterparties
Five-point scoring guide
1
Poor
Cannot communicate beyond technical peers.
2
Needs Improvement
Explanations lose the audience or oversimplify to the point of error.
3
Satisfactory
Adequate with familiar audiences; less effective under challenge.
4
Very Good
Explains clearly to committees and regulators and holds up under challenge.
5
Excellent
Explains assumptions and sensitivities in plain language, showing the drivers behind a number rather than just the number.
Put this rubric to work
Score every candidate against the same standard
Add these weighted factors to Hirevire and let AI evaluate recorded answers against your rubric.
