A measured baseline
For the paid evaluation: current quality, time, and cost on an agreed test set, with evidence gaps stated.
AI cost per accepted task
Free AI cost calculator. Count review work, too.
Estimate the cost of an accepted AI task, including calls, retries, tools, and human review. For a tested recommendation, we also offer a paid evaluation of one existing workflow.
Free worksheet. No signup required.
No signup. Your inputs stay in this page.
Compare the whole task
MeasureBaselineCandidate
Answer qualityTestSame cases
Retries & timeRecordCompare
Cost per accepted taskRecordCompare
A lower token price
is not the whole answer.
Compare quality, time, and cost.
Decision / adopt, revise or keep current
Calculate cost per accepted task
A lower model bill can hide more rejected work. Use the same tasks and acceptance criteria to compare candidates, including the cost of unsuccessful attempts.
Cost per accepted task = total operating cost for the sample / accepted tasks
Set the comparison
Record the sample size, reporting period, currency, and acceptance rule. Count each task once, even if it requires several calls or retries.
Count the whole operating cost
Add model charges for all attempts, allocated tool and hosting costs, and review and correction time valued at your chosen labor rate. Include each cost once.
Keep the decision visible
Record accepted tasks, critical failures, and review time alongside the unit cost. Keep one-time implementation costs separate. A candidate must still meet the quality floor.
Worked example: a cheaper model bill, a costlier result
Hypothetical amounts in USD for arithmetic only. These are not our prices, client results, or a savings forecast. Both candidates attempt the same 100 tasks under the same acceptance rule.
| Measure | Baseline | Candidate |
|---|---|---|
| Model charges, including retries | $60 | $40 |
| Allocated tools and hosting | $20 | $20 |
| Review and correction labor | $120 | $180 |
| Total operating cost | $200 | $240 |
| Accepted tasks | 80 of 100 | 75 of 100 |
| Cost per accepted task | $2.50 | $3.20 |
The candidate spends less on model calls but costs more per accepted task and produces fewer accepted results. It does not justify a cost-saving rollout on this evidence.
Start with one
recurring problem.
The token bill is only part of the cost. Failed attempts and human corrections still use time and money. Compare the same tasks and acceptance rules so a cheaper request does not hide a more expensive completed job.
Want us to assess your workflow?
The worksheet above is free. This is a separate, optional service.
One-workflow AI evaluation
From $2,500 USD
One-time project. Final scope and price agreed in writing.
Find why an existing AI workflow needs too much correction, takes too long, or costs more than expected.
Book a free scoping call ↗30 minutes with the co-founders to check fit, tools and the next scoping step. No obligation to proceed.
Included in the starter scope
- Assessment of one existing AI workflow and one improvement decision.
- An agreed set of representative tasks, known failures, and scoring rules.
- A baseline covering the quality, time, and cost measures available for those tasks.
- One controlled candidate comparison using the same test set.
- A findings report with recommended changes, trade-offs, and rollout checks.
- Quoted separately
- Production implementation, a full system rebuild, and additional workflows or candidate comparisons are quoted separately.
- A larger project
- Quoted after scoping — Implement selected improvements, evaluate more candidates, or establish ongoing regression checks. This is an alternative scope, not an add-on to the starter price.
- Optional ongoing support
- Optional recurring evaluations are scoped around agreed changes and test sets; no retainer is required for the assessment.
- Timing
- We confirm timing after checking whether the current workflow can be reproduced and who can judge the results.
All prices are in USD. Third-party software, API usage, hosting and any telephony charges are separate. Support is optional; its coverage and limits are agreed separately. New workflows need a new scope.
From input
to a usable result.
Illustrative workflow. We agree the details for your project.
- 01 / The starting point
An existing AI workflow costs more than expected per accepted result.
- 02 / The right context
Measure a baseline using agreed tasks, quality rules, available usage, and review effort.
- 03 / Prepared for review
Test one candidate change against the same cases.
- 04 / A person decides
Compare the results and recommend whether to change or keep the current setup.
Tangible work
What the paid evaluation includes
One candidate comparison
One controlled change tested on the same cases, including regressions and uncertain results.
A written recommendation
A change-or-keep decision, trade-offs, and checks for a separately scoped production rollout.
From first call
to handoff.
- 01
Define the evaluation
Choose one existing workflow, the accepted result, and a representative task set with scoring rules.
- 02
Measure the baseline
Use available records for calls, retries, tools, and human review. State what cannot be established from those records.
- 03
Test one candidate
Compare quality, time, and cost on the same cases. Recommend a change, further investigation, or keeping the current setup.
What to agree
before we begin.
The calculator is free and uses your assumptions; it is not an evaluation of your system. The paid starter covers one workflow and one candidate comparison. Production changes and additional comparisons are quoted separately.
How we evaluate the work ↗A few useful answers
Before we begin.
How much does a first project cost?
One-workflow AI evaluation starts at $2,500 USD for the starter scope shown on this page. We confirm access, scope, exclusions, timing and a written price before work begins. Third-party software and usage charges are separate; ongoing support is optional.
What happens on the free scoping call?
In 30 minutes, we discuss one task, your current tools and the result you need. We check whether the starter fits and what is needed for a written proposal. If paid discovery is needed, we explain its scope and price first. You do not need a prepared brief or a paid assessment to book the call.
How will we judge whether the project is useful?
We agree a baseline and acceptance checks with the person who owns the work. A reproducible baseline and candidate comparison support a change-or-keep recommendation, with limitations stated. Where relevant, we also compare manual effort, correction work and operating costs. These checks guide the handoff; they are not a promise of revenue or savings.
Is the calculator free?
Yes. You can use it without buying an evaluation. It estimates from the values you enter; it does not inspect your workflow or verify the assumptions.
What does the paid evaluation add?
From $2,500 USD, we assess one existing workflow, establish an agreed baseline, test one candidate change, and provide a recommendation. Production implementation is separate.
Will you just choose a smaller model?
No. A model is one possible change. A cheaper request may create more retries or human correction work.
What if the candidate is not better?
Keeping the current setup can be the right result. The comparison will show the trade-offs and remaining uncertainty.
Do you change production during the assessment?
No. Implementation requires a separately agreed scope. The evaluation provides evidence for that decision.
What if no tasks are accepted?
Report the total spend and zero accepted tasks. Cost per accepted task is undefined, not zero. Investigate the failures before treating the candidate as a viable option.
Do retries and rejected outputs count?
Their costs count in the total; they do not create extra accepted tasks. Count one accepted task only when the final result meets the agreed rule. Use the same rule for corrected outputs in every candidate.
Can this worksheet predict our savings?
It compares the sample you measured. Different volumes, task mixes, review needs, or provider charges can change the result. Document those assumptions and test an approved rollout before projecting savings.