Estimate the annual benefit and payback of one AI use case.

Neutral synthetic figures for an invoice-coding process. They are not drawn from any engagement, client or benchmark, and they are not a prediction about your own process.

• Use-case name — Supplier invoice coding and exception routing
• Accountable owner — Finance operations manager
• Currency — AUD
• Process volume — 900 instances a month
• Time per instance — 14 minutes
• Loaded hourly rate — 62 per hour
• Annual rework or error cost — 48000 a year
• Other annual baseline cash cost — 22000 a year
• Proportion in scope — 70 %
• Residual human time per in-scope instance — 4 minutes
• Realisation you are willing to state — 40 %
• Avoidable cash cost you can name — 30000 a year
• Revenue or capacity value, if any — 0 a year
• Implementation cost — 85000 one-off
• Running cost — 1400 a month
• Maintenance — 18000 a year
• Pilot duration — 8 weeks
• Key dependencies — Read access to the accounting ledger; a labelled sample of 400 historical invoices; one finance reviewer for two hours a week during the pilot.
• Consequence if the output is wrong — Material: affects a payment, a customer or a regulatory record

Hours released

1,260.0

hours a year

Realised capacity value

AUD 31,248

a year

Avoidable cash cost

AUD 30,000

a year

Revenue or capacity assumption

AUD 0

a year

Annual modelled benefit

AUD 26,448

a year

First-year net

AUD -58,552

annual modelled benefit − implementation cost

Simple payback

38.6 months

shown only when the annual modelled benefit is positive

Assumptions register

Every figure below is one the reader supplied, not one this page chose. They travel with the result so a reviewer can argue with them.

• Proportion in scope — 70 %
• Residual human time per in-scope instance — 4 minutes
• Realisation you are willing to state — 40 %
• Avoidable cash cost you can name — 30000 a year
• Revenue or capacity value, if any — 0 a year
• Pilot duration — 8 weeks
• Consequence if the output is wrong — Material: affects a payment, a customer or a regulatory record
• Key dependencies — Read access to the accounting ledger; a labelled sample of 400 historical invoices; one finance reviewer for two hours a week during the pilot.

Information gaps

Nothing in this example is missing or self-contradictory.

Why the benefit lines stay apart

The questions a reviewer asks, in the order they arrive

1. Is the baseline measured or estimated? An estimated baseline is acceptable if it is labelled. Presented as measured, it is the fastest way to lose the case.
2. What happens to the released hours? Redeployed onto what, or removed from what. A case that cannot answer should set realisation low and lean on cash.
3. Why is residual time not zero? If it is, no person checks an in-scope instance. Compare that with the consequence you selected.
4. Who handles the out-of-scope work? Exceptions rarely get cheaper, and often get more expensive per instance once the easy work leaves.
5. Is the running cost complete? Inference or licence charges, hosting, monitoring, and the review time the workflow still needs.
6. What would make you stop? If there is no answer, the pilot has no stop condition — which is the pilot measurement plan.

1. It is a single-period model. Twelve months at a constant volume, rate and running cost. No discounting, no ramp, no benefit arriving in month nine.
2. It does not know whether the use case is technically feasible. Feasibility usually turns on where the data lives, what condition it is in and who may see it — none of which appear above.
3. It does not validate your realisation percentage. That figure is the single largest lever in the model and it is entirely yours. A case that is only positive at a high realisation is a case about redeployment.
4. It is not a quotation. The implementation and running figures are yours. KyperX publishes engagement starting prices separately.
5. Payback is simple, not discounted. No cost of capital and no tax effect.

• Loaded hourly rate — salary plus on-costs divided by productive hours, rather than the headline salary divided by 1,800.
• Equivalent capacity value — released hours priced at that rate. A measure of size. It becomes money only when a cost is removed or the time is redeployed onto something valuable.
• Realisation — the proportion of capacity value you are prepared to claim as a benefit. Yours, not ours.
• Avoidable cash cost — spend that stops. The only benefit line that changes a bank balance on its own.
• Residual human time — minutes still spent per in-scope instance after the change, including review and correction.