Signal Lab · Practical AI

AI Payback & Review Check

Will this AI tool actually pay for itself once a person has checked its work?

Why it matters: most AI business cases count the time saved and leave out the time spent checking, fixing and paying for it. This shows the payback that's left after review, error redo, licences and ramp-up.

Practical AI SMEs

Your numbers stay in this browser. Nothing is uploaded. Illustrative, not financial or technology advice.

Start from a worked example

Tasks in this case

Steps 1–2 are per task. Step 3 costs are shared: one AI tool usually covers them all.

The work today
items

Invoices, recs, emails — whatever one unit of the job is.

min

Time a person spends on one item, start to finish.

$/hour

Wage plus super, on-costs and overheads, in AUD.

%

Share of items redone today. Use 0 if none.

With AI
%

The rest stay manual, as today.

min

Human time to prompt, approve and post — before any review.

%

Share of AI output a person checks against the source.

min

One check of one item.

%

Share of AI items that are wrong. Errors found in review are redone by hand; the rest get through.

$/error

Fixing it later, a credit note, a late fee. Blank = not counted (flagged in the result).

What it costs · shared across all tasks
$one-off

Implementation, set-up fees, integration, an adviser's time.

hours

Costed at the loaded hourly cost above.

$/seat/mo

Use 0 if it's already in software you pay for.

seats

Only the people who'll use it.

$/month

One figure for the whole case, at full volume. Per-document or per-call charges for any task go here too.

months

One ramp-up for the whole case: the saving builds evenly to 100% over this time. 0 = full from month 1.

See the result

What if the checking takes more, or the AI is less accurate?

Move the sliders for one task at a time. Your inputs stay as they are; the result below is for the whole case.

Where the saving goes, and when you get your money back

Monthly figures at full benefit, then the cash position month by month including setup and ramp-up.

How this is calculated
  • AI items = items per month × share AI handles
  • Gross time saving = AI items × (minutes today × (1 + rework rate) − minutes with AI) ÷ 60 × hourly cost. This is the number most pitches stop at.
  • Human review = AI items × review rate × minutes per review ÷ 60 × hourly cost
  • Redo of caught errors = AI items × AI error rate × review rate × minutes today ÷ 60 × hourly cost (a caught error is redone by hand)
  • Errors that get through = AI items × AI error rate × (1 − review rate) × cost per error (only if you enter one)
  • Running costs = licence per seat × seats + usage fees
  • Net monthly saving = gross − review − redo − errors through − running costs (at full benefit)
  • Review drag = (review + redo + errors through) ÷ gross saving
  • Setup = one-off setup + training hours × hourly cost, spent at month 0
  • Ramp-up: in month t the time-based figures are scaled by t ÷ ramp months (capped at 100%); running costs apply in full from month 1
  • Payback = month the cumulative cash line crosses zero (part-months interpolated). The pitch line uses the gross saving only, with the same setup, running costs and ramp-up.
  • ROI (12 months, 3 years) = cumulative net cash after setup ÷ cash spent (setup + licences + usage) over the period
  • Hours freed = (time saved − review time − redo time) ÷ 60, per month at full benefit
  • Break-even review rate = review rate at which net monthly saving reaches $0. If review stops more error cost than it costs, it shows a minimum instead of a maximum.
  • Break-even AI accuracy = accuracy at which net monthly saving reaches $0, at your review rate

Hours freed only become cash if the time is redeployed or hiring is avoided. Excludes tax and the time value of money. Illustrative — not financial or technology advice.

Take it with you

Pairs with the AI Starter Kit for SMEs: pick a use there, test its payback here.

Your case, including the time-saved setting, is saved in this browser only and stays on your device.