AGI Technologies: AI automation ROI reference pack
Prepared12 October2026. Original authored planning model, not a client result.

Requirements: Python3.9 or later, standard library only. No install, network,
tracking, credentials or integration is needed. Inspect source before running.
Run: python3 roi_calculator.py assumptions.json
Check arithmetic: python3 check_calculation.py

Replace every hypothetical input with an observed value or a labeled assumption.
Counts are distinct adopted tasks monthly, not prompts, retries or all demand.
Before/after minutes include all human checking, corrections and exceptions
for the same accepted output. Extra maintenance belongs in recurring cost.
Loaded hourly value is a labor/opportunity-cost assumption, not AGI pricing.
Capacity realization0.5 means half of positive released time is valued as usable.
It does not prove cash savings. Extra labor is counted fully as a cost.
Operating cost is incremental over the baseline; include failed runs and support.
Implementation includes internal effort as well as external delivery.
Cash benefit is separately evidenced avoided expenditure or incremental cash
contribution, net of associated costs. Do not enter hypothetical salary savings.

Outputs: separate economic capacity and cash views. Do not add those views.
null payback: no positive recurring net benefit. null ROI: zero denominator.
Realization threshold above1: impossible under the current assumptions.
This steady-state reference excludes ramp-up, seasonality, discounting, tax,
financing and residual value. It cannot detect duplicate benefits or validate
the credibility of inputs. Use a month-by-month plan for an actual approval.

MEASUREMENT BRIEF
Workflow and owner:
Accepted output and minimum quality:
Current period and representative task mix:
Eligible volume / actually adopted volume / excluded volume:
Before active minutes / after active minutes / elapsed turnaround:
Unresolved cases and work carried past cutoff:
Usage, failed runs, review, maintenance and shared-cost treatment:
Implementation and internal time:
Released hours, reassigned work and evidence of its value:
Separately evidenced cash expense avoided:
Assumption that changes the funding decision:
Spend limit, quality stop rule, decision date and next owner:

Reference formulas and limitations are explained in the accompanying article.
Arithmetic checks validate only this local model, not an automation platform,
forecast or business outcome. No private or customer data is included.
