AI ROI Check
Is your AI workflow paying off?
AIGENZ is an independent practice. The AI ROI Check examines one bounded AI-enabled workflow and ends in a decision: Scale, Fix or Stop.
Request a discovery conversation
Scope one bounded workflowDecision Scale, Fix or StopTiming about five business days, once three conditions are met
One workflow. One decision owner.
The AI ROI Check looks at one deployed or serious-pilot AI-enabled workflow, not your whole AI program. You supply evidence from a short checklist, in redacted summary or export form, and we may interview up to two people involved. We do not need access to your systems.
You receive:
- An economic analysis against the pre-AI baseline and the full operating cost
- A workflow scorecard
- Reliability and control findings
- An attribution confidence rating
- A Scale, Fix or Stop conclusion tied to your decision
- A 90-day action plan
See a sample decision memo (fictional example).
Sold to businesses only. Every figure is labeled observed, derived, modeled or assumed. Modeled figures are estimates, not realized results.
What the decision looks like.
A short sample of the kind of memo the AI ROI Check ends in. The company, workflow, numbers and findings below are all invented for illustration. AIGENZ did not do this work for a real client and measured none of it.
Illustrative example · fictional company · not a client engagement
Decision memo: invoice-dispute drafting at Example Supply Co.
Decision asked: expand AI-drafted replies from one team to three?
Conclusion (illustrative)Fix Do not scale yet. Close one control gap, re-measure, then decide.
Evidence considered illustrative
- Assumed: volume, handling times and the hourly rate below were invented for this example.
- Modeled: the monthly cost comparison is calculated from those inputs, not measured.
- Illustrative sample: 40 invented exception tickets reviewed for errors.
- Not available in this example: a baseline from the same season as the AI period.
Economics, per month illustrative
Every input is an assumption. Change one and the answer changes.
| Line | Assumed input | Cost |
|---|
| Before AI: manual handling | 1,000 tickets × 12 min = 200 h at $40/h | $8,000 |
|---|
| With AI: human review of every draft | 1,000 × 6 min = 100 h at $40/h | $4,000 |
|---|
| With AI: rework on 15% of drafts | 150 × 20 min = 50 h at $40/h | $2,000 |
|---|
| AI tool and usage fees | assumed flat fee | $1,500 |
|---|
| Upkeep and prompt maintenance | 5 h at $40/h | $200 |
|---|
| With AI: full workflow cost | | $7,700 |
|---|
| Modeled difference | $8,000 − $7,700 | $300 (about 4%) |
|---|
A vendor claim of “drafts in minutes, not hours” looks like a large saving. Across the full workflow, with review, rework, fees and upkeep, the modeled difference is small. Break-even is a rework rate of about 17%, only two points above the assumed 15%.
The $300 values 50 released hours at $40. It becomes cash only if hours are actually removed or redeployed. In this example that has not been shown.
Reliability and control concern illustrative
In the invented sample of 40 exception tickets, 6 contained a wrong credit amount that the reviewer had not caught. Reviewers were spending about 6 minutes per draft with no required check of amounts against the invoice. The control exists on paper but did not catch this class of error in the example.
Attribution illustrative
Confidence: low to medium. The before period in this example was a peak quarter and the after period was a quiet one, so some of any difference may come from volume and staffing, not the AI.
Why Fix, not Scale or Stop illustrative
- Not Scale: the modeled margin is thin, sits near break-even, and an uncaught-error problem would grow with volume.
- Not Stop: the workflow is plausibly positive and the gap is specific and fixable.
- Recommended next step (illustrative): require an amount-to-invoice check in review, track the rework rate for 30 days against a same-season baseline, then decide again.
Three kinds of evidence, kept apart.
A workflow can look cheap and still be unreliable, or look reliable and still owe its gains to something else. We report each of these separately and never let one stand in for another.
Economics
What does the workflow cost and produce?
Cost and output before AI and now, including human review, rework and the cost of running the AI itself.
Time saved is not cash saved. Capacity released is not capacity redeployed.
Reliability and controls
How does it fail, and who catches it?
How errors are caught, how often exceptions and escalations occur, and how much human work remains in the loop.
An observed control is not an assurance of future behavior.
Attribution
Was it the AI?
How confident anyone can be that the AI, and not seasonality, staffing or a process change, drove the difference.
Modeled ROI is not realized ROI.
The answer is a decision.
The conclusion is tied to the decision you actually face. Where evidence is thin, the deliverable says so plainly.
- Scale The evidence supports expanding the workflow.
- Fix Value is plausible, but a specific gap needs closing first.
- Stop The evidence does not support continued spend.
Price and timing
One offer, two prices: $2,500 for the first five paid clients, then $5,000.
- Founding price
- $2,500
- first five paid clients
- Standard price
- $5,000
- after the first five
The founding price creates no right to later pricing.
The delivery target is about five business days. The clock starts on the business day after all three conditions are met:
- valid payment has cleared,
- intake is complete, and
- the required evidence is judged sufficient.
What this is not
- No guarantee of ROI, savings, revenue or any outcome.
- Not certification, assurance, attestation or an audit opinion.
- Not legal, compliance, financial, tax or investment advice.
- No implementation, integration, security testing, or access to your systems.
- Conclusions are limited by the evidence you provide.
Please do not send health records, payment card or account numbers, government IDs, credentials or source code.