AI makes it easy to ask for one more headline, another audience, a shorter version, or a new channel adaptation. Each request can look harmless because the direct model charge is small.

The cost appears in the combination of brief changes, review, coordination, version comparison, and publishing work. A fixed campaign can become an open-ended production queue without anyone formally changing the scope.

Scope is the set of work the team agrees to accept. Draft volume is only one signal that scope has moved.

Define the unit before drafting#

Choose a unit that a buyer and a producer can count:

Unit	Definition
Brief	One audience, purpose, source set, and approval owner
Asset	One accepted item in a named format and channel
Variant	A meaningful change to audience, offer, language, or channel
Review pass	One decision on a specific version
Published item	An accepted asset at a live destination

Without these definitions, “ten assets” may mean ten final files, ten initial drafts, or ten families of variants. Those are different commitments.

The agency invoice guide shows why a project price can remain fixed while the amount of production work changes. Put the definition in the brief and the contract.

See how variants multiply#

Consider a launch with:

three audience groups;
two offers;
two channels;
two languages.

The requested matrix contains 24 possible variants. If each variant needs a separate source check and approval, the process has 24 review decisions even when the model creates all of them in one batch.

Stage	Count	Work created
Core messages	3	Strategy and source approval
Audience adaptations	6	Positioning and claim review
Channel adaptations	12	Format and length checks
Language adaptations	24	Glossary and local review
Published variants	18	Metadata, links, and QA

The exact counts will differ. The important question is whether the team planned for the review and publishing work, or only for generation.

Separate a change from a repair#

A producer should not treat every returned asset as scope growth. Use a clear distinction:

Request	Classification	Budget treatment
Fix a fact that was unsupported	Repair	Included in the original unit
Correct a missing required field	Repair	Included in the original unit
Change the approved audience	Brief change	New decision
Add a new language	Scope growth	New variant and review
Move the asset to a new channel	Scope growth	New format and QA
Try a different opening while the brief is unchanged	Revision	Count against the attempt limit
Add a new offer after approval	Scope growth	Reforecast before drafting

This classification protects both sides. The buyer can see when a vendor is charging for a repair, and the producer can show when a new request changed the agreed work.

Add a change log to the brief#

The change log can stay small:

Date	Change	Owner	Effect
July 8	Added a second audience	Product marketing	Six new variants
July 10	Updated the offer	Growth	Recheck all claims
July 12	Added German	Regional marketing	Translation and local review
July 15	Removed the paid channel	Demand gen	Four variants closed

Record the effect before the next batch starts. This turns a vague request into a forecast change. It also shows which team made the decision without assigning blame.

Set limits people can act on#

Every campaign should have:

a maximum number of accepted assets;
an attempt limit per asset;
an approval owner for new variants;
a spend envelope for direct usage and external work;
a deadline after which stale items close or need a new brief;
a rule for who can approve scope growth.

The budget variance guide gives the monthly review. Scope decisions belong earlier, while the team can still decline a low-value variant.

Measure the cost of the extra work#

Track the original request separately from additions:

accepted assets in the original brief;
accepted assets added through change requests;
attempts and review passes for each group;
direct spend by campaign and variant;
hours spent on coordination and publishing;
variants accepted but never published;
variants closed because their date passed.

The spend attribution guide explains why a shared key cannot answer these questions. A campaign ID, variant ID, and owner are enough to make the first useful join.

Close the campaign with a scope review#

At the end of the campaign, compare planned and actual work:

Which variants were in the original brief?
Which ones were added and who approved them?
Which accepted items went live?
Which returns were repairs?
Which additions produced a useful outcome?
Which recurring request should become a standard format?

The goal is not to prevent every extra idea. It is to make the cost and decision visible before the team spends another week creating versions that nobody will publish.

Deixic helps make spend by agent, approval decisions, current activity, and missing cost proof visible while a campaign is active. Use the product view alongside the campaign brief so a scope change has an owner and a decision before more work starts.