AI pointed at the commercial layer.

The problem it solves

A statement of work is where margin is won or lost, and it is usually reviewed by whoever is free on a Thursday. Scope creeps in through a single unbounded adjective. A fixed-fee engagement carries time-and-materials language. A guarantee gets promised in a sentence nobody costed. The delivery team then spends six months paying for a paragraph.

This is also the least fashionable place to point AI, which is exactly why it is worth doing. When AI compresses delivery hours, the commercial model has to change with it, and the firms that notice first are the ones already looking at their own paper.

What I built

A reviewer that runs two lenses over the same document and converges on a single verdict.

The scope lens asks whether the document is deliverable-led, bounded, phase-separated, customer-obligated, value-framed, and change-controlled. The contract lens asks whether it is the right kind of document, on the right paper, with safe wording, sound math, testable acceptance, and explicit authority.

Commercial model fluency

The taxonomy covers all five models the firm sells, and the review includes a behavior test that checks whether the language actually matches the model it claims to be. A capacity engagement described in deliverable language is a specific, expensive kind of mistake, and it is invisible unless someone is looking for it.

The math never gets eyeballed

A pricing script recomputes every fee table: milestones, budgets, capacity, fixed terms, percentages, and deliverable payment coverage. A bounds scanner distinguishes genuinely vague scope terms from properly bounded ones, so a reviewer is not flagging "up to seven priority datasets" as if it were open-ended.

It edits the actual document

Output is not a memo about the document. It is the document, redlined, with tracked changes inserted at paragraph level and formatting cloned from the surrounding list so the edit does not announce itself as machine-made. It also generates clean documents against the current brand standard, and pulls replacement language from a pre-vetted clause library organized by practice area.

What it gives an organization

• Consistent commercial rigor that does not depend on which reviewer was available.
• Findings that cite the standard, so they survive an argument with a deal owner.
• A read on the other side's paper before it is signed rather than after.
• A place to work out how pricing has to change as AI compresses the hours.