VORKLAB | COST PER ACCEPTED AI WORKFLOW RESULT
Worksheet v1.0 | 11 October 2026
Guide: https://vorklab.com/guides/ai-task-cost

Use one workflow and a defined batch or period. Label estimates separately
from measured values. Use the same scope, workload mix, quality and deadline
for every option. This is a planning worksheet, not a savings forecast.

1. DEFINE THE OUTCOME
Workflow:
Input / trigger:
Unique enquiry or task ID:
Acceptance criteria:
Evidence of completion in the destination system:
Deadline / observation window:
Prohibited errors:
When to hand off to a person:
Who accepts the result:

2. MEASURE THE BASELINE
Representative workload and dates:
Incoming unique tasks:
Accepted unique results:
Unresolved or rejected tasks:
Total human minutes, including checks and corrections:
Loaded hourly cost (split by staff role if needed):
Other attributable operating costs:
Completion time and serious errors:

3. LOG EACH AI-ASSISTED TASK
Task ID:
Task type / complexity:
Model and paid tool costs for ALL attempts:
Number of attempts / retries:
Human review minutes:
Additional correction minutes:
Manual fallback minutes:
Quality audit minutes:
Outcome and evidence:
Completed without human handling? (record later audits separately)
Accepted / rejected / unresolved:
Completion time:
Error severity:

Count each minute once. If a task is reviewed and then redone, count both.
A direct manual handoff does not also get review time unless it happened.
Three attempts on one task never become three accepted results.
An audit sample does not prove that every unsampled result is correct.

4. ADD FIXED AND IMPLEMENTATION COSTS
Additional hosting / monitoring / subscriptions per month:
Maintenance / support per month (including labour):
Shared overhead allocation and rationale:
Upfront process design, data preparation, integration, testing,
training and handover:
Planning horizon in months:
Monthly implementation allocation = upfront cost / horizon
Do not double-count labour already included in a supplier support fee.
Shared costs excluded from a comparison must be identical for all options.
Add them back for a complete business cost view.

5. CALCULATE
Human handling cost = sum over roles of (minutes / 60 * loaded hourly cost)
Operating cost = all model/tool attempts + human handling + operation/support
Cost including allocated setup = operating cost + implementation allocation
Cost per accepted result = cost including allocated setup / accepted results
Also show operating cost per accepted result without setup allocation.
If there are no accepted results, report no completed output; do not divide
by zero or report a zero cost per accepted result.

Keep failed attempts in costs. Do not count unresolved work as accepted.
Match expenses to outcomes. In a pilot, follow a batch through completion
rather than dividing one month's unfinished work by another month's output.
When later checks reveal errors, revise acceptance and add recovery costs.

6. TEST THE ASSUMPTIONS
Lower-than-expected monthly volume:
Longer review time:
Higher manual rework / fallback rate:
Shorter useful life or an early integration rebuild:
Support / hosting capacity step-ups:
Cost of serious errors tracked separately from routine rework:
Recalculate all options. A cheaper unit cost does not override a failed
quality, security or service-level requirement.

7. SEPARATE CAPACITY FROM CASH
Handling hours released:
Additional support effort (if handled internally):
What useful work can absorb the capacity?
Actual cash spending avoided (e.g. overtime or outsourced handling):
Additional cash operating payments:
Verified incremental contribution from extra work, if any:
Monthly net cash benefit:
Upfront cash payment:
For a constant monthly net cash benefit:
Cash payback months = upfront cash payment / positive monthly net cash benefit
If net cash benefit stays <= 0, there is no cash payback. For changing
benefits, track cumulative net cash flow until it covers upfront spending.
Do not count unchanged salaries as cash savings. Do not deduct allocated
setup again in monthly cash benefit when using upfront spending above.

8. DECIDE AND REVISIT
Expand / revise / stop:
Evidence and unresolved questions:
Owner:
Next review date or change that triggers review:

THE GUIDE'S FICTIONAL EXAMPLE
N = monthly enquiries, all assumed accepted eventually.
Loaded handling cost = $30/hour. Setup allocated over 12 months.
Manual monthly cost = 3.00 * N
Reviewed AI drafts monthly cost = 500 + 1.65 * N
Bounded automation monthly cost = 1600 + 0.72 * N

At N=1000: $3000 / $2150 / $2320, or $3.00 / $2.15 / $2.32 per result.
Drafts: 2 review minutes on all tasks + 6 extra minutes on 20%;
$0.05 usage per input; $250 monthly support; $3000 upfront setup.
Bounded: 20% reviewed for 2 minutes, 10% directly handled for 6 minutes,
70% automatic-path tasks, 10% of those audited for 2 minutes;
$0.15 usage per input; $600 monthly support; $12000 upfront setup.
Usage includes retries; support excludes handling already counted above.
All rates and prices are illustrative, not client data or market averages.