Five separate worked examples / Marketing
Keep the comparison fair, then account for what it cost.
The example compares two hypothetical months with the same media spend and raw response count. Qualified leads differ; that comparison alone does not prove AI caused the difference.
- What stayed equal
$2,000 of media and 40 raw responses in each month.
- What is assumed to change
20 to 30 qualified leads, then two to three wins at the same 10% assumed close rate.
- Where the math stops
$3,000 more contribution after media, before incremental AI and testing costs.
| Measure | Baseline | Improved |
|---|---|---|
| Media spend | $2,000 | $2,000 |
| Raw responses | 40 | 40 |
| Qualified leads | 20 | 30 |
| Cost per qualified lead | $100 | $66.67 |
| Wins at assumed 10% | 2 | 3 |
| Booked at $10,000 per win | $20,000 | $30,000 |
| Gross profit at 30%, if delivered | $6,000 | $9,000 |
| Less media, once | − $2,000 | − $2,000 |
| Contribution before incremental AI/test cost | $4,000 | $7,000 |
$7,000 − $4,000 = $3,000 difference before incremental AI and testing costs. Media sits outside the assumed direct-delivery margin and is deducted once in each month. This comparison does not establish causation. Released time is not added again.
Open the detail
Keep the setup, assumptions and checks in view.
Where did the 24 released hours come from?
Where did the 24 released hours come from?
Comparable content work required less preparation time.
The improved process may help organize approved audience information, customer language, prior campaign learnings and draft content. Human review still determines whether the message is accurate, appropriate and worth testing.
The 24 hours are capacity, not cash savings. Fixed payroll does not fall merely because preparation is faster. The value of this step depends on what the team does with the time.
Owner decision: identify whether the released time will improve testing, audience research, content quality or another approved marketing task.
Value is not the number of posts
The team could spend all 24 hours publishing more. That would raise output without showing that anyone qualified to buy responded. Marketing value is the ability to create relevant audience and message learning that produces more qualified demand from comparable campaign conditions.
What did the 12 reused hours do?
What did the 12 reused hours do?
| Input | This test |
|---|---|
| Audience | Held the same |
| Message | Changed and recorded |
| Offer framing | Held the same |
| Creative angle | Held the same |
| Call to action | Held the same |
Testing converts capacity into learning.
The example reuses 12 of the 24 released hours for reviewed audience and message tests. The other 12 hours receive no financial credit. Comparable campaign conditions make the result easier to interpret.
AI may help prepare variations and organize prior learning. A marketer still approves the test, checks compliance and brand fit, and records what changed. Without that record, next month’s team cannot tell which change mattered.
Owner decision: select the next test based on an observed weakness in qualified demand, not simply on the desire to produce more content.
Who does what in a reviewed test
| AI | Prepares variations and organizes prior learning. |
|---|---|
| Marketer | Approves the test, checks compliance and brand fit, records the change. |
| Owner | Picks which weakness to work on next. |
What improved in the demand path?
What improved in the demand path?
Baseline month
40 raw → 20 qualified → 2 wins
Improved month
40 raw → 30 qualified → 3 wins
Raw responses stayed the same; qualified demand increased.
The useful change is not more raw activity. It is a larger share of responses meeting the qualification standard, followed by more accepted sales. A qualified lead is still not revenue. It only becomes a win if Sales closes it.
The comparison is illustrative. It does not prove that AI caused the extra win. These could also explain a difference:
- Offer
- Audience
- Seasonality
- Sales execution
- Campaign conditions
Owner decision: define “qualified” before testing and inspect the records behind each stage.
Why the definition comes first
If the standard loosens during the test, 30 can look better than 20 with no change in real demand. Hold the standard steady across both months, and check it against what Sales later accepts.
What did each qualified lead cost?
What did each qualified lead cost?
Baseline: $2,000 ÷ 20 qualified leads
$100.00
Improved: $2,000 ÷ 30 qualified leads
$66.67
The price of a lead fell because more responses qualified, not because media got cheaper.
Both months spend the same $2,000 and get the same 40 raw responses. Keeping those two inputs equal is what makes the comparison fair. Change the budget at the same time and the unit cost stops saying anything about the message.
This figure counts media only. It excludes staff time, AI costs and test costs, and no amounts are supplied for them. A lower cost per lead is also not a result by itself. A cheap lead that never closes is still a cost.
Owner decision: track cost per qualified lead beside accepted wins from the same cohort, so a falling unit cost cannot hide weak leads.
Why the rounding is shown
$2,000 ÷ 30 does not end in whole cents. Multiplying the rounded $66.67 by 30 gives $2,000.10, so use the $2,000 media figure, not the rounded unit cost, for the money calculation that follows.
How do the two months compare financially?
How do the two months compare financially?
Baseline: $20,000 booked
Improved: $30,000 booked
Direct cost implied by 30% margin Media, same in both Contribution
Separate bookings, margin and media contribution.
| Line | Baseline | Improved |
|---|---|---|
| Wins × $10,000 booked | $20,000 | $30,000 |
| Gross profit at 30%, once delivered | $6,000 | $9,000 |
| Less media, once | − $2,000 | − $2,000 |
| Contribution after media, before incremental AI and test costs | $4,000 | $7,000 |
The gross profit appears only if the accepted work is delivered at the assumed 30% direct-delivery margin. Media is not inside that margin, so it comes out separately, once in each cohort.
The illustrative difference is $7,000 − $4,000 = $3,000. It is conditional, and it stands before incremental AI and testing costs. It is also a comparison of two months, not proof of what caused the change.
Owner decision: compare the full cohort records, then deduct actual incremental AI and testing costs.
Why the 24 hours are not added here
Released hours are capacity, and fixed payroll does not fall because preparation is faster. Adding a dollar value for those hours to the $3,000 would count the same improvement twice: once as time, again as the better results that time made possible.