Task Cost Ledger

---
name: task-cost-ledger
description: Calculate the full cost of an AI-assisted task, including review and rework time, instead of stopping at the subscription price. Use when deciding whether AI assistance is actually worth it for a recurring task.
license: All-Rights-Reserved
metadata:
  author: Zain Haseeb
  version: 1.0.1
  changed: "2026-09-08"
  category: ai-workflows
  tags: [cost-analysis, roi, decision-making, time-tracking]
---

# Task Cost Ledger

Show the true cost of an AI-assisted task as arithmetic, not a vague impression, by counting review and rework time alongside the tool's price.

## Critical behavior

1. Compare time first, then value it using the supplied hourly rate and currency. Time value is capacity value, not automatic payroll or cash savings.
2. Include review and rework time as a real cost line, not an afterthought.
3. Show the arithmetic in a small table rather than only stating a conclusion.
4. Identify the biggest cost driver from the supplied numbers, without assuming which line wins.
5. State a break-even frequency if one exists in the numbers.
6. Mark missing numbers as unknown. If useful, show clearly labeled hypothetical scenarios separately; never turn a missing cost into zero or an estimate into an observed fact.

## When to use

Use this before adopting AI assistance for a recurring task, when renewing a tool subscription, or when a team disagrees about whether an AI workflow is actually saving time. It is most useful for tasks that run often enough that a small per-instance error compounds into a real number.

Do not use it for a single one-off task where the arithmetic overhead outweighs the decision at stake.

## Build the ledger

### 1. Gather the inputs

Collect minutes per completed task without AI; AI setup/prompting, active waiting, review, rework and handoff minutes per task; hourly capacity value and currency; expected runs per year; marginal tool fees per run; attributable fixed recurring fees and their billing period; and one-time setup cost or setup hours. Separate elapsed unattended wait from human time. Record comparable quality requirements, evidence behind each estimate and uncertainty ranges.

Clarify which fees are incremental to this decision, shared allocations or already committed. A subscription already included in fixed fees must not also be charged as a marginal cost. If allocating shared fees, state the allocation and apply it once. Keep baseline-only cash costs separate if supplied. Reject negative values, non-finite values or inconsistent units rather than calculating through them.

### 2. Calculate both totals

Use the same currency, time basis and quality standard. Let B be baseline minutes, A total AI-workflow human minutes, R the hourly rate, M incremental marginal tool cost per run, N annual runs, F attributable annual fixed recurring cost, and S one-time setup cost including valued setup time. Convert the stated billing period into F explicitly; do not assume a monthly fee is annual.

Baseline capacity cost per run = B × R / 60. AI capacity cost per run = A × R / 60. Before fixed fees, net value per run D = (B − A) × R / 60 − M. With no other baseline cash costs supplied, annual net value = N × D − F; first-year net value = N × D − F − S. If other baseline costs exist, show their incremental difference as a separate line and incorporate it consistently. When R is unknown, show time and cash costs separately, not a combined ROI.

### 3. State the net difference

Show per-run time difference, annual capacity hours and annual/first-year net value separately. At N > 0, allocated recurring net value per run = D − F / N; at N = 0 it is undefined, not zero. Report fixed costs still incurred at zero usage. Do not multiply a weekly frequency by an unexplained annual factor.

### 4. Find the hidden driver

Identify the largest evidenced cost driver. Show a sensitivity case for uncertain review/rework time and name what would need to be measured or changed. Do not assert that better prompting will achieve a target reduction.

### 5. Find the break-even point

If D > 0, recurring break-even runs per year = F / D; first-year break-even runs = (F + S) / D. State that equality only breaks even and use the next whole run to describe strictly positive value. If F = 0 there is no positive fixed-fee frequency threshold. If D <= 0 and costs are nonnegative, more runs cannot create a positive economic case; report that rather than dividing by zero or returning a negative threshold. If inputs are unknown, do not fabricate a threshold.

For a review-time threshold with R > 0 and N > 0, let O be AI human minutes excluding review. Maximum review minutes to break even on recurring costs = B − O − 60 × (M + F / N) / R. Subtract an additional 60 × S / (N × R) for first-year break-even. A negative threshold means no nonnegative review time reaches break-even under these assumptions. Reducing review below the quality requirement is not an acceptable economic improvement.

## Output format

Return an input/units/evidence table, a cost table separating human time, marginal fees, recurring fixed fees and one-time setup, the formulas and substitutions, recurring and first-year results, break-even where meaningful, sensitivity, and a decision limited by quality and missing evidence. Label hypothetical figures and capacity value clearly.

## Example

Synthetic example: baseline 30 minutes; AI preparation 8, review 10 and rework 2 minutes; rate $60/hour; marginal fee $1/run; attributable subscription $20/month; setup $90 once; 100 runs/year. D = (30 − 20) × 60 / 60 − 1 = $9/run. F = 20 × 12 = $240/year. Recurring net value = 100 × 9 − 240 = $660/year; first-year net value = $570. Recurring break-even is 26.67 runs/year, so 27 whole runs exceed it. First-year break-even is 36.67 runs, so 37 exceed it. These are capacity-value estimates, not cash savings or proof of acceptable output quality. If total AI human time rises to 32 minutes, D becomes −$3/run; no increase in usage makes that case positive.

## Troubleshooting

- If review time was never tracked, leave the observed result incomplete and optionally show labeled estimates/ranges. Recommend timing a representative task before adoption.
- If the two totals are close, the honest answer is that the choice is marginal, not that AI clearly wins or loses.
- If the same task varies a lot in review time, run the ledger for a typical case and a worst case separately.

## What not to do

- Do not treat review or rework time as zero without evidence.
- Do not round in the direction that favors AI adoption.
- Do not present a conclusion without showing the underlying arithmetic.
- Do not skip the break-even calculation when the inputs support one.