The team says AI saved 1,000 hours.

The marketing budget is still higher than last quarter.

That is not automatically a contradiction. It may mean the team used the time to produce more campaigns, test more variants, absorb a launch, or cover work that would have gone to an agency. It may also mean the savings claim was never connected to a budget line.

Time saved is a capacity claim. Cash saved is a budget claim. Treating them as the same is how AI programs win a slide and lose finance.

The business case gets stronger when it says exactly which result happened.

Four kinds of value need different evidence#

Claim	What it means	Evidence required
Cost reduction	A real cash obligation became smaller.	Lower invoice, contract scope, headcount plan, or tool renewal.
Capacity	The same team has time for additional work.	Before-and-after time, accepted output, and the work that used the capacity.
Output growth	The team produced more accepted work.	Accepted assets, publication dates, and quality or approval state.
Risk reduction	Fewer errors, escalations, or unsupported claims reached customers.	Rejection reasons, incidents, corrections, and approval evidence.

One process can produce all four. Do not use one metric to claim all four.

The savings test#

Use the same accepted result as the denominator:

realized savings
= comparable old-process cost
  − current full production cost

“Comparable” matters. If the old process produced ten approved campaign assets and the new process produced ten approved assets, compare the full cost of those ten. Include review, coordination, agency work, tools, and shared research where they belong.

If the new process produced twenty drafts and ten approved assets, comparing its model bill with the old process’s ten finished assets is not a savings calculation. It is a favorable denominator.

The $90,000 question#

Imagine a team used to spend $90,000 per quarter on an agency for campaign adaptations. After adopting AI, the internal team takes over the work. The provider bill is $6,000. Internal review and coordination add $28,000. The agency scope falls to $55,000.

The possible statements are different:

Statement	Calculation	Result
Direct AI spend	Provider and tool bills	$6,000
Internal production cost	AI, review, coordination, shared work	$34,000
Agency reduction	Old agency scope − new agency scope	$35,000
Net cash change	Old agency scope − agency scope − internal production cost	$1,000

The program may have produced $35,000 of agency reduction and only $1,000 of net cash change before other costs. It may also have created capacity and improved control. The honest business case names each result.

“We brought the work in-house” does not prove savings. The contract, staffing plan, and full internal cost show whether cash actually moved.

What happens to the reclaimed time?#

If a team saves ten hours per week, record where those hours go:

more accepted campaign assets;
faster launch cycles;
deeper customer or market research;
fewer agency or contractor hours;
better review and source checking;
internal work that used to be deferred;
unplanned requests that consume the capacity without a budget change.

Unassigned reclaimed time is still useful to know. It is a capacity result, not a savings result. The next budget review should ask whether the capacity is being protected for strategic work or being filled with more low-value production.

The proof ledger#

For each claimed benefit, keep one row:

Field	Example
Old process	Agency produced 10 email adaptations in 5 days
New process	Internal team produced 10 approved adaptations in 2 days
Accepted output	10
Full old cost	$42,000
Full new cost	$31,000
Cash change	$11,000 before fixed costs
Capacity result	3 days returned to the launch team
Quality evidence	0 unsupported product claims
Owner	Marketing operations
Review date	30 days after publication

The review date matters. A fast launch with a correction campaign two weeks later is not the same result as a fast launch that stays accurate.

The claims that usually inflate the business case#

Watch for these shortcuts:

Hours saved × salary = savings. This estimates value only if the staffing or spend changes.
Drafts produced = output. A draft is not an approved or published asset.
Lower token price = lower production cost. Review and rework can dominate the bill.
Agency work removed = agency savings. Confirm the contract or scope change.
Fewer review minutes = lower risk. Fewer minutes can mean weaker checking.
More tools = more capability. Tool licenses and integration work can grow faster than use.

The fix is not a more optimistic model. It is a ledger that keeps each claim attached to its evidence.

A 30-day proof plan#

Choose one repeatable process and establish a baseline before changing it:

Days 1–7: Baseline#

Count accepted assets, full production cost, review minutes, agency or contractor spend, approval delay, and correction work. Record the old process while the details are available.

Days 8–21: Change one thing#

Introduce the AI-assisted process with a named owner, an approval threshold, a work-item ID, and a source set. Record every accepted, rejected, and abandoned outcome.

Days 22–30: Reconcile#

Compare the same accepted output against the old process. Separate provider spend, human cost, agency change, capacity, quality, and risk. Mark every unproven assumption as an open question.

The result may be a cash saving. It may be capacity. It may be a process that should stop. All three are useful decisions. The weak result is a percentage with no owner and no before-and-after evidence.

Where Deixic fits#

Deixic keeps the work evidence needed to make the comparison possible: process and owner, model and tool activity, source material, spend, approval, and outcome. It does not turn capacity into a savings claim.

Use the product view to inspect spend and activity. Use the cost-per-asset guide for the full production denominator. Use the budget guide to set the envelope and decision points before the next month starts.

The point of the proof is not to make AI look good. It is to know whether the work is worth funding again.