Published volume is a weak measure of AI content productivity

“We published 400 pieces” sounds like progress. It leaves out whether the pieces were accepted, whether customers saw them, and whether the work supported the campaign.

AI makes volume especially easy to increase. A useful report must keep the stages separate so a larger number does not hide a larger review queue or a lower acceptance rate.

Output is a count. Productivity is an accepted result divided by the full cost of producing it.

Use a funnel with named stages#

Track the same content through:

Do not combine these stages into a single “created” metric. A draft that never gets reviewed is different from an accepted page that misses its launch date.

Put cost beside each stage#

For one quarter, imagine:

If the program reports cost per draft, it appears efficient: the provider charge is divided by 640. Cost per accepted asset and cost per published asset tell a different story. Cost per measurable asset may be the right denominator for a performance claim, but only after the measurement window is defined.

The savings-proof guide explains why capacity gains and cash savings need separate labels. This funnel gives each claim a place.

Define “accepted” before the work starts#

An accepted asset should have a checkable definition:

• the stated audience and channel are correct;
• required claims have usable sources;
• prohibited claims are absent;
• the format and length meet the brief;
• the owner and approver have made their decision;
• the asset can proceed to its destination.

If acceptance means “a reviewer stopped commenting,” the metric will change with the reviewer. Write the conditions in the brief and keep the return reason when an item fails.

The brief-quality guide gives a practical field list. Reuse it in the reporting model.

Do not credit content before the window#

A live asset may need time before performance can be judged. State:

• the date it became live;
• the channel and audience;
• the primary outcome;
• the minimum traffic or time window;
• the comparison baseline;
• the owner of the result.

Until that window closes, classify the outcome as pending. A campaign team should not have to choose between claiming success and declaring failure when the data is simply incomplete.

Compare like work#

Efficiency comparisons fail when the denominator changes. Hold constant where possible:

• content type;
• audience;
• channel;
• approval class;
• source burden;
• campaign stage;
• measurement window.

An AI-assisted product page should not be compared with an internal announcement because the review and outcome paths differ. A higher cost can be reasonable when the asset supports a higher-value or higher-risk decision.

Use return reasons to improve the funnel#

The funnel tells you where a loss occurs. Return reasons tell you what to change:

The waste-before-publish guide covers the cost of work that stops before customers see it. Keep the reason so the next report can show whether the loss is shrinking.

Give finance a defensible summary#

A monthly report can answer:

1. How many requests started?
2. How many accepted assets did the budget produce?
3. What share reached the intended destination?
4. What cost remained in review, repair, and coordination?
5. Which outcomes are measured, pending, or unknown?
6. Which process change should improve the next month?

The answer should include missing cost evidence. Treating missing usage as zero improves the ratio while leaving the budget unexplained.

Deixic gives a view of spend by agent, activity, approvals, and gaps in cost proof. Use the product view to support the operational stages, then connect live content and performance outcomes from the marketing systems that own them.