Model Scope, Drivers, And AI Brief

You are building a board-ready CFO pack and someone suggests using AI to speed up the model build. The professional risk is not the tool. It is letting AI invent scope, drivers, or assumptions and then discovering the mismatch when the board asks why ARR does not tie to deferred revenue, why cash does not reconcile, or why valuation is based on metrics your auditors will not accept.

Your objective in this lesson is to define a modeling scope, a driver set, and success criteria tightly enough that AI output becomes usable model input. You still own the model’s integrity and its interpretability. AI can draft structures and tables, but it cannot take accountability for what the model implies.

Before you brief AI, align on what the model must contain and how it will be consumed. Use the architecture below as the target shape so every driver and output has a designated home.

Lock the scope to board use

Scope is not a feature list. It is a contract between you and the decision makers about what questions the model will answer and what questions it will explicitly not answer. If you skip this contract, you get a fragile workbook that looks complete but fails under scrutiny when scenario logic breaks, circularities appear, or KPIs cannot be reconciled to GAAP or IFRS reporting.

For a SaaS CFO pack, define the minimum board-use outputs up front. A 3-statement model that ties, scenarios that are auditable, and a valuation module that is explicit about method and drivers are table stakes because directors will ask tie-out questions, not only growth questions.

Build the AI briefing packet

AI needs a briefing packet the same way a junior analyst does. Without it, it fills gaps with plausible numbers or category defaults, and you end up with hallucinated inputs that quietly drive your cash runway or valuation. The consequence is not cosmetic. It is decision error that you cannot defend because you cannot trace the assumption to a source.

Your packet has three working documents. An assumptions log that states each driver, the value, the time basis, and the source. A data dictionary that defines each metric and its calculation so ARR, revenue, billings, and deferred revenue are not conflated. Constraints that tell AI what it is not allowed to invent, including any nonpublic operating metrics and any accounting policy choices that require your judgment.

Use the next exercise to convert an unstructured planning memo into structured driver inputs that a model can actually consume.

Specify outputs with acceptance tests

A prompt that asks for a model is a prompt for improvisation. A prompt that specifies formats, naming conventions, and check requirements is a prompt for deliverables you can review. If you do not define acceptance tests, you will spend the time you saved generating work reformatting tables, hunting broken links, and discovering late that the model does not balance.

Include four elements in your blueprint. Role and context that pins the domain and the entity type. The exact output formats, including tables shaped to match your Excel layout and named ranges like Rev_Drivers or WC_Assumptions. Explicit neutrality where needed, for example US GAAP and IFRS framing unless you direct otherwise. Acceptance tests that AI must satisfy, such as balance sheet balancing, cash reconciliation, and ARR movement checks.

Draft your prompt blueprint using a template that forces structured tables and check totals, so you can paste with minimal interpretation.