Works with the AI Agents you already use
AI Cost per Successful Task
Calculate AI cost per completed business outcome, including retries, failed attempts and supplied review costs. Keep missing prices and invoice gaps visible.
$9
AI Cost per Successful Task
Example session with this skill installed
FICTIONAL supplied-evidence exercise; no live actions.
Fictional complete same-cohort ledger: 100 tasks, 80 confirmed successes, 140 attempts including failures. Inclusive input 1,400,000 tokens with 400,000 cached; output 200,000. Fictional USD per million: input 2, cached .50, output 8. Tools1.20 USD, human review4 USD, infrastructure missing. Matching invoice3.90 USD before taxes/credits. Calculate inference and included cost per success, coverage and reconciliation; do not claim all-in savings. No account access.
- Read your context and instructions
- Compiled the ai cost per
AI task economics — cost-1
This supplied fictional cohort has inference cost of
3.80 USD and included cost of
9.00 USD, or
0.0475 USD and
0.1125 USD per confirmed successful task, respectively. Infrastructure is missing, so 9.00 USD is an included-cost subtotal, not complete all-in cost. No savings result is established.
Boundary and evidence
The supplied complete same-cohort ledger covers 100 canonical business tasks, 140 attempts including failed work and retries, and 80 confirmed successful tasks. One confirmed success is counted per task; attempts do not create additional successes. The underlying business success criteria, window, timezone, model/version, workspace and test-traffic exclusions were not supplied. These results adopt the exercise's confirmed-success labels and same-cohort assertion, rather than claiming independently verified production outcomes.
| Source | Supplied input | Treatment and coverage |
|---|---|---|
| Fictional complete cohort ledger; date unspecified | 100 tasks; 140 attempts; 80 confirmed successes | All supplied attempts remain in cost, including failures. Remaining 20 tasks are not confirmed successes; no separate unresolved count is supplied. |
| Same ledger | Inclusive input 1,400,000 tokens; cached subset 400,000; output 200,000 | Cache is included in input and must be subtracted once to find uncached input. Counts are nonnegative integers; cached input does not exceed inclusive input. |
| Fictional supplied tariffs; effective date unspecified | USD per 1,000,000 tokens: uncached input 2; cached input 0.50; output 8 | Exercise assumptions only; no current provider-price assertion. |
| Supplied additional cohort costs | Tools 1.20 USD; human review 4.00 USD | Included once. Infrastructure was not supplied and is not assigned zero. |
| Matching supplied invoice | 3.90 USD before taxes/credits | Compare against inference tariff estimate; scope is accepted as matching for this exercise. |
Reproducible calculation
Use exact integer token counts and decimal monetary arithmetic; retain precision through division.
- Uncached input = 1,400,000 − 400,000 = 1,000,000 tokens.
- Uncached input cost = 1,000,000 × 2 / 1,000,000 = 2.00 USD.
- Cached input cost = 400,000 × 0.50 / 1,000,000 = 0.20 USD.
- Output cost = 200,000 × 8 / 1,000,000 = 1.60 USD.
- Inference cost = 2.00 + 0.20 + 1.60 = 3.80 USD.
- Included cost = 3.80 + 1.20 + 4.00 = 9.00 USD.
- Inference cost per success = 3.80 / 80 = 0.0475 USD.
- Included cost per success = 9.00 / 80 = 0.1125 USD.
- Confirmed success rate = 80 / 100 = 80%.
- Attempts per canonical task = 140 / 100 = 1.4.
Monetary totals are shown to cents; unit costs retain four decimal places. Rounding unit costs to cents would conceal useful precision. The 40 attempts above one per task are not all independently classified as waste: retry/fallback details are not supplied.
Coverage and interpretation
For the exercise, usage aggregates cover all 140 attempts and all supplied token categories are priced: 140/140 attempt coverage, 100%. Individual attempt rows are not available for independent linkage or duplicate verification. The confirmed-success denominator is 80 of the complete 100-task cohort; the remaining outcome categories cannot be separately reported. Known additional costs include tools and review; infrastructure remains a gap. Therefore there is no defensible complete all-in cost or all-in savings claim.
Invoice reconciliation
Invoice minus modeled inference = 3.90 − 3.80 =
+0.10 USD, unresolved. The invoice is higher. Do not substitute the 9.00 USD subtotal into this inference reconciliation because it also contains tools and review. No unsupported balancing adjustment is added.
The billing/finance owner should reconcile the 0.10 USD against invoice line categories, billing rounding, scope/time boundaries and any discounts or other billed categories. Taxes and credits are excluded from the supplied invoice value; discount treatment and the precise invoice line breakdown are not supplied. The tariff model remains 3.80 USD pending explanation, and the invoice remains the supplied 3.90 USD observed amount.
Next measurement and optional experiment
Next, have the analysis owner obtain a redacted invoice breakdown, rate effective interval and infrastructure allocation for exactly this cohort, together with the business success contract and window/timezone. No account credentials or customer message bodies are needed.
An optional future retry-reduction experiment would compare the same task mix and success definition against this baseline, while retaining all failed attempts and human-review cost. Its owner must first set numerical quality and latency floors and a rollback threshold; none were supplied. Record confirmed outcomes, attempts and complete included costs for both cohorts and uncertainty around the comparison. This is a proposed measurement, not an executed optimization or forecast of savings.
Connects securely to your tools. The creator never sees your data.
About this skill
Define a business-success cohort and reconcile canonical tasks with every included attempt. Calculate inference and included cost per success with explicit cache-token semantics, decimal arithmetic, dated rates, coverage and invoice differences. Includes a cost-method reference, worksheet and fictional worked example. Compare quality-adjusted experiments without inventing savings. This is an analysis workflow, not a live meter or billing connector. No API keys, provider account, current prices or executable calculator are bundled. Fictional supplied-evidence exercises are documented separately; no live provider, database, browser or cross-agent compatibility acceptance is claimed.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.
Results depend on complete usage/outcome evidence and applicable rates; zero successes make the ratio undefined. No guaranteed savings or complete all-in cost without all costs.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 22 days ago
Permissions
Read bundled and supplied redacted evidence only. Optional browsing, saving a report, implementation and external effects depend on host tools and task authorization. No provider account or credentials required for supplied-evidence review.