Cost · · 4 min read · Updated
Every campaign variant has a cost, even when the model is cheap
Campaign variants multiply review, source checks, and publishing work. Plan the variant matrix, count accepted outcomes, and require a decision before adding another branch.
One campaign message can become a dozen headlines, four audiences, three channels, and several languages before the first version reaches a customer.
AI makes the matrix fast to create. It does not make each variant free to approve, check, publish, or measure.
The cost of a variant is the work required to accept and use it, not the time required to generate it.
Draw the matrix before asking for copy#
List the dimensions that can change:
- audience;
- offer;
- channel;
- language or market;
- product tier;
- customer stage;
- legal or approval class.
Then calculate the possible combinations. A plan with three audiences, two offers, two channels, and two languages has 24 possible variants. The team may decide to publish 16, but the decision should happen before all 24 enter review.
| Dimension | Options | Combinations so far |
|---|---|---|
| Audience | New buyer, trial user, customer | 3 |
| Offer | Product education, upgrade | 6 |
| Channel | Email, landing page | 12 |
| Language | English, French | 24 |
The matrix is a planning tool. It does not mean every combination deserves a separate asset.
Separate adaptation from a new message#
Use a simple classification:
| Variant type | Example | Review burden |
|---|---|---|
| Format change | Email subject to preview text | Check length and meaning |
| Channel adaptation | Landing page to paid ad | Check claim, destination, and limits |
| Audience adaptation | Admin to finance lead | Check problem and proof |
| Offer change | Education to discount | Check terms and approval |
| Market adaptation | English to French | Check language, terms, and local rules |
| New message | New benefit or product claim | New source and owner decision |
This helps the team reuse approved facts without pretending every variation has the same risk.
Count accepted and published variants#
A useful campaign report can show:
| Measure | Count |
|---|---|
| Variants requested | 32 |
| First drafts created | 32 |
| Variants entering review | 28 |
| Accepted | 19 |
| Published | 15 |
| Closed before publication | 4 |
| Still waiting | 9 |
The direct generation charge may be small. The 13 variants that did not publish still consumed review, source, coordination, and planning time. The cost-per-asset guide explains why the accepted asset belongs in the denominator.
Give the campaign a variant budget#
Set a limit for:
- total accepted assets;
- variants per audience;
- attempts per variant;
- reviewer minutes;
- direct provider and tool spend;
- time spent after acceptance;
- unallocated or experimental variants.
If a new branch exceeds the limit, ask the campaign owner to choose which existing branch to close or which budget to increase. This creates a decision while the added work is still optional.
The scope-creep guide gives the change-log structure. Add the variant ID so a request can be traced to the exact audience, offer, and channel.
Protect source and approval consistency#
Every variant should inherit:
- the source version;
- approved claims;
- prohibited claims;
- offer terms;
- required disclosures;
- owner and approval class.
If a variant changes one of those fields, mark the change and send it to the relevant owner. A batch can share a source set and still create a new review obligation when one sentence changes.
Measure whether the extra branches helped#
Do not call a campaign efficient because it produced many variants. Compare:
- accepted and published variants by audience;
- review minutes per published variant;
- direct cost per published variant;
- qualified action or revenue by variant;
- variants that received too little traffic to evaluate;
- variants closed before the measurement window.
A variant with no distribution or no decision window is unfinished measurement, even if the copy was approved. Keep the result labeled as unknown rather than assigning credit.
Close the matrix after the campaign#
At the end, mark each branch:
- published and measured;
- published but below the measurement threshold;
- accepted and held;
- returned for repair;
- closed because the campaign changed;
- rejected because the expected value was too low.
This gives the next campaign a real starting point. It also prevents old variants from returning as “already created” when their source, offer, or destination is stale.
Deixic can show agent spend, activity, approvals, and missing cost proof while the matrix is active. Use the product view to catch added branches that have no owner or approval, then keep campaign performance in the system that owns those outcomes.