What "done" means for a marketing asset

Defining done for marketing work

A content calendar is a list of promises. Every row is a thing that does not exist, described by a title that sounded good in a planning meeting.

This is why calendars survive contact with a busy month so badly. When the supplier shipment goes wrong and the founder disappears for three weeks, a calendar produces exactly nothing, because every row still requires the scarcest resource in the business to start from zero. A queue of finished drafts produces output, because all it requires is a decision.

That's the difference the build step of the loop turns on. And it only works if "done" means something specific.

Table of Contents

The definition

An asset is done when it could publish today without anyone doing further work on it.

That is the whole test, and it is stricter than it sounds. Applied honestly it disqualifies:

A thing that needs twenty more minutes is not done, because those twenty minutes will be spent at the worst possible time: when you're trying to ship. Small unfinished tails are what turn a week of drafted work into a week of nothing published.

Things that passed as done and weren't

We ran this definition backwards across our own site in September 2026 — auditing live assets as though they were being submitted for approval. Several failed:

Every one of these was marked complete at the time. They failed the "could publish today" test months after publishing, which is the uncomfortable part: they were never done, they were just live.

The lesson we took is that the definition of done has to include the mechanics, not just the copy. Does the button work. Does the link resolve. Does the image show the current product. Copy gets reviewed because it's the visible part; mechanics get assumed because they're boring.

The brief format

Most unfinished drafts are unfinished because the brief was. A title and a keyword is not a brief — it's the calendar row again, passed along.

This is the structure we use. It's deliberately short enough to fill in for real:

ASSET:        Blog post / email / ad — and where it publishes
AUDIENCE:     Who specifically, at what stage
JOB:          The one thing this asset needs to do
CLAIM:        The single thing we're asserting
EVIDENCE:     The specific data, example or result backing that claim
                (if this line is empty, the asset does not get written)
CTA:          Exact destination URL
MUST INCLUDE: Facts, constraints, product details that must be right
MUST AVOID:   Claims we cannot make, competitors we don't name,
                anything under review
DONE WHEN:    The publishable condition — images in, links live,
                subject line written

The two lines that do the work are EVIDENCE and MUST AVOID.

Evidence, because a brief without it produces a draft padded with generic industry statements — and that is the material that no longer ranks and never converted. If nobody can fill that line, the honest response is to go and get the evidence, or drop the asset. We enforce this on ourselves at the build level: no post ships from this site without something in it that exists nowhere else.

Must-avoid, because it's the only line that prevents the expensive failure. AI drafts don't know which of your capabilities are real. We published a claim that our AI routed support tickets; it didn't. That line was plausible, well-written, and false, and a populated must-avoid list would have stopped it.

Why the brief matters more than the model

There's a persistent belief that output quality is a model problem — that the drafts would be publishable on a better model.

They wouldn't. The gap between a draft that needs rewriting and one that needs approving is almost entirely specificity of input. A model cannot invent the fact that your best-converting segment is second-time buyers in the 60-day window, or that you can't make a delivery-time claim in one market. Those facts exist in your head. Whether they reach the draft is a briefing decision, not a model decision.

Which is also why the same tool produces excellent output for one brand and generic output for another. The difference is what it was told.

Working ahead

The practical shape of all this: keep a queue, not a calendar. Assets enter it finished. Publishing becomes a decision rather than a project, and the decision takes minutes — which is what makes the approval gate survivable.

Two weeks of finished work in the queue is enough to absorb a bad month. That's the entire goal: the marketing keeps moving on the weeks the founder can't.

FAQ

Isn't this just a content calendar with extra steps?

It's the inverse. A calendar front-loads decisions (what, when) and back-loads work. A queue front-loads work and back-loads the decision. The second one survives disruption because the expensive part is already paid for.

How do I get drafts back finished rather than half-done?

Fill in the evidence line before writing anything, and specify the done-when condition in the brief itself. Most half-finished drafts are perfectly obedient responses to half-finished instructions.

Can a tool run this?

The queue part, yes. Tilly works this way — it audits the store, drafts the work, and holds it for you to approve before it publishes to Klaviyo, Omnisend, Shopify Email or Meta. What it can't do is supply the must-avoid list. That's yours, it's short, and writing it once is the highest-leverage hour in the whole system.