Your AI tool saved ten hours. Did it save any money?

An AI tool saves your marketing team ten hours a month. The invoice is $200. At a $40 hourly cost, that looks like a $200 monthly gain.

But if payroll stays the same, no contractor invoice falls and the team does nothing valuable with the spare time, your bank balance is $200 worse. The tool created capacity. The business has not yet turned it into a financial return.

To measure AI marketing ROI, keep two views: the economic cost of getting acceptable work done, and the cash that actually changes. Both matter. Mixing them makes a disappointing purchase look profitable.

The following worksheet and numbers are an illustrative model, not client results or a promised saving. Sources were checked October 4, 2026.

Count approved work, not generated drafts

Choose one recurring job: a reviewed newsletter, a client-ready report or an approved batch of ad variations. Write the acceptance criteria before testing the tool.

For an ad variation, that might mean a correct offer, a defensible claim, brand-appropriate writing and the required format. Track the time spent briefing, prompting, checking, rewriting and handling failed attempts. All of it belongs in the cost of the accepted output.

Google's October 1 AI-content guidance reinforces why checking cannot disappear from a publishing workflow: it calls for factual review of generated content and associated metadata before publication. Faster generation is only one stage of finished work.

Use this planning formula:

Economic cost per approved item = (labor time × loaded hourly cost + tool and usage cost) ÷ approved items

Apply the same boundaries before and after. If the old process includes editing and the AI process excludes it, the comparison cannot answer the purchasing question.

See where the saving can disappear

Assume a hypothetical team produces 40 approved items monthly. Its loaded labor cost is $40 an hour. Each item previously required one hour. The AI subscription and usage together cost $200 a month.

Monthly measure	Existing process	AI, lighter review	AI, heavier review
Accepted items	40	40	40
Total labor hours	40	25	37
Allocated labor cost	$1,600	$1,000	$1,480
Incremental tool and usage cost	$0	$200	$200
Total economic cost	$1,600	$1,200	$1,680
Economic cost per approved item	$40	$30	$42

The heavier-review process is $80 more expensive even before setup. In the lighter-review case, the team gains 15 hours of capacity and lowers modeled economic cost by $400. That still does not establish a $400 cash saving.

This is why the first trial should cover the whole job. Include difficult items and rejected drafts. Do not compare the AI's easiest output with the old process's average assignment.

Make a separate cash calculation

List only changes that actually affect money:

Monthly cash benefit = cash costs avoided + additional contribution earned - incremental cash costs

Cash costs avoided could include a canceled contractor task or paid overtime that genuinely stops. Additional contribution is additional revenue minus the variable costs of fulfilling it. The Small Business Administration's break-even guidance makes this distinction useful: revenue is not the amount left to pay the business's other bills.

If the lighter-review example saves employee time but nothing else changes, monthly cash benefit is negative $200. If it removes $500 of avoidable contractor cost while adding the $200 tool bill, cash benefit is $300. Do not also count the same contractor hours as an extra labor saving.

If the team uses spare capacity to win and deliver more work, count the contribution from that work once it is supported by records. Separate observed results from forecasts. A larger content queue does not itself establish incremental revenue.

Put setup and failure costs back in

Record migration, training, workflow design and maintenance. Separate the cash you pay from internal time allocated. Also record expected correction costs, while treating serious quality or privacy failures as reasons to stop, not merely a cost to average away.

Suppose setup requires $900 in incremental cash and the verified ongoing cash benefit is $300 monthly. Simple cash payback is three months:

Cash payback months = upfront incremental cash cost ÷ recurring monthly cash benefit

If recurring benefit is zero or negative, there is no positive payback period under that scenario. Internal setup time should still appear in the economic assessment even when it creates no additional payroll payment.

Our offer-pricing guide helps estimate delivery contribution. The owner's marketing metrics guide helps keep the business outcome visible after the trial.

Write a renewal rule the team can use

Run the trial on comparable assignments over a complete work cycle. Keep acceptance standards fixed, record total time and retain rejected work in the denominator of attempted work. Report approved items separately.

Renew if the workflow meets the quality bar and either creates verified cash benefit or frees capacity for a named, worthwhile use. If the case is capacity, say so. Do not present it as money already saved.

Stop or redesign if review and correction erase the gain, the team cannot use the extra capacity, or an important quality requirement fails. A cheaper invoice is not a bargain when the finished work costs more.