AI Investment Review Card

Define the outcome. Compare performance. Decide where to invest.

Fill in the five steps for one workflow. Totals and cost per successful outcome calculate as you type, and step 5 suggests a decision from your scores. Your entries stay in this browser.

Workflow Owner Review period

Establish success criteria before measuring

Identify the ideal outcomes of the system.

Customer or business outcome What counts as success? Quality and risk thresholds

Compare like with like

Use comparable workloads and periods; record data sources and sample size.

Measure

Baseline

With AI

Target

Result

Accepted outcomes per period

End-to-end completion time

Total cost

—

Cost per successful outcome

—

Sources, sample and comparison limits

Build the cost picture

Include failed attempts and retries. Use the same cost boundary in both periods.

Cost for the period	Baseline	With AI
AI usage
Tools / infrastructure
Human work
Rework
Total cost	—	—
Successful outcomes (from step 2)	—	—
Cost per successful outcome	—	—

Cost for the period

Baseline

With AI

AI usage

Tools / infrastructure

Human work

Rework

Total cost

—

Successful outcomes (from step 2)

—

Cost per successful outcome

—

Total cost = AI + tools / infrastructure + human work + rework

Baseline: enter costs above.

With AI: enter costs above.

Cost per successful outcome = Total cost ÷ Successful outcomes

One-time setup cost, kept separate

Score the evidence

Assess gains against agreed targets; keep these scores separate.

Outcome gain

Improved Unchanged Worse

Quality and risk

Pass Fail Unknown

Evidence strength

Strong Limited Insufficient

Make the next decision

Expand when outcomes justify total cost and quality holds. Test further when evidence is weak.

Current system constraint

Decision Expand Optimize Test further Stop

Why this decision Next intervention and owner Expected result and review date

How the suggestion works

Quality fails: stop if outcomes also got worse, otherwise optimize to fix quality first.
Quality unknown or evidence insufficient: test further.
Outcomes worse with quality passing: stop.
Outcomes unchanged: optimize if cost per successful outcome fell; stop if evidence is strong and cost did not fall; otherwise test further.
Outcomes improved but cost per successful outcome did not fall: optimize.
Outcomes improved, quality passing, strong evidence: expand.
Outcomes improved, quality passing, limited evidence: optimize if you named a constraint, otherwise test further.

The suggestion is a starting point from your own scores. The decision stays yours.