Three proposals arrive at the same meeting. One has a pilot behind it. One has a conviction. One has a deadline attached to someone else's budget. A comparison table feels like the way to settle it. It is only useful if the columns hold honest input.
Why a single number fails first
The instinct when comparing initiatives is to score each one and rank the scores. That instinct borrows authority from arithmetic it has not earned. A score compresses evidence, assumptions, cost and timing into one figure, and in doing so hides exactly which parts of the decision are supported and which are guesses. When the initiative underperforms, the score cannot tell you whether the evidence was thin or the assumption was simply wrong.
Substantiation rules push in the same direction. The FTC's guidance for small businesses states that under the Federal Trade Commission Act, advertising must be truthful and non-deceptive, advertisers must have evidence to back up their claims, and ads cannot be unfair (ftc.gov). The agency describes a “reasonable basis” as objective evidence supporting the claim, with the required kind of evidence depending on the claim itself (ftc.gov). If a proposed campaign will make claims, the comparison is not only about budget and timeline. It is also about whether the evidence to support those claims exists yet, or has to be produced.
Split the columns before you compare
The most useful structural change to any comparison is to give evidence and assumptions separate homes. Evidence is something you can point to and someone else can check: a prior test, a published reference, a measured baseline. An assumption is a belief you are acting on because checking it is expensive, slow, or impossible right now. Both are legitimate inputs to a decision. Only one of them can be cited later if the decision is questioned.
This separation is what the companion guide on keeping evidence separate from assumptions sets out in practice. The point is not to eliminate assumptions — you cannot plan without them — but to keep them visible. A comparison that names its assumptions is auditable. A comparison that buries them in a score is not.
What to put side by side
A practical comparison holds a small, fixed set of rows per initiative:
- The decision question the initiative is meant to answer. If two proposals answer different questions, they are not rivals and should not be ranked against each other.
- Evidence you can show, with where it came from.
- Assumptions the plan rests on, stated plainly.
- Cost, including the cost of finding out whether the assumptions are true.
- Owner and the time constraint that governs the work.
- What remains unknown, written as a question rather than a blank.
The worksheet at /decision-brief/ is built around exactly these fields for up to three proposals. It produces no score and promises no outcome. It exports a private Markdown brief in which the unknowns are left standing, on the view that a plan which conceals its gaps is harder to review than one which names them.
What the frameworks say about uncertainty
Government and platform guidance converges on the same caution. NIST describes its AI Risk Management Framework as intended for voluntary use, to help organizations incorporate trustworthiness considerations into the design, development, use and evaluation of AI products and services (nist.gov). Its Generative AI Profile, released 26 July 2024, is described as a resource to help organizations identify risks unique to generative AI and propose actions for managing them (nist.gov; airc.nist.gov).
NIST also notes that current approaches to evaluating AI often fail to account for real-world risks and impacts, and that its ARIA program pairs people with AI applications to study behaviours in scenario-based interactions (airc.nist.gov). The implication for a marketing comparison is modest but real: where a proposal depends on an AI-assisted step, the comparison should record what is known about how that step behaves in your context, rather than assuming a capability.
Google's search documentation makes a related point about how content is judged. It states that its systems prioritise helpful, reliable information created to benefit people rather than content made to manipulate rankings, and that fabricating creator profiles is deception and a signal of a low-quality page (developers.google.com). It also treats interactive features such as calculators and online tools as main content when they directly help the page achieve its purpose (developers.google.com). None of this promises a result. It does suggest that the quality of the comparison instrument itself is part of the work.
A comparison you can defend
When the brief goes to the deciding group, three questions should be answerable from the page alone: which claims rest on evidence and which on assumption; who owns each initiative and against what deadline; and what is still unknown. If the table cannot answer those, it is a preference dressed as analysis. The route from question to comparison is set out in starting with a decision, and the follow-through is in setting a review cadence so the assumptions get revisited rather than fossilised.