Pick the work, enter the numbers you already have, and get a verdict in about ninety seconds — with the supervision it would need and the reason behind it, not just a savings figure.
| Today | $9,600/mo | 2,112 units | $4.55/unit |
|---|---|---|---|
| 2 people + AI | $6,075/mo | 2,562 units | $2.37/unit |
Output goes up, not down. The comparison holds capacity constant, so a smaller team is never dressed up as a saving.
11% of replies reviewed at steady state, after a ramp. Capability-gated — this verdict moves when models improve.
Reproduce this auditPick the closest match — this sets realistic cost benchmarks and starting assumptions for the audit.
Be honest about the busywork — that's where the audit finds the gap. Your numbers size the opportunity in pounds and hours; repeatability, judgment and risk are what decide the verdict itself.
Which model, if you already know (optional)
Left alone this uses a mid-tier model. Change it if you want, but be warned: on every task in this tool the gap between the cheapest and dearest model is a rounding error next to salary and setup. The results page shows you exactly how small. List rates last looked at September 2026. Providers change pricing without notice and it will not move the verdict. See the formulas
Judgment & risk
This is what actually decides the verdict — not the cost math.
See the exact formulas this usesFramework v4.2, rates and weightings last recalibrated 9 September 2026
Based on the numbers you entered.
Weighing repeatability, judgment, and risk for this task…
Same setup, priced under every model
What the audit also found
All three are calculated from the numbers you just entered. Nothing here is locked.
When your setup cost pays for itself
Calculating…
The exact point where hybrid stops being worth it
Calculating…
The one input holding this score back
Calculating…
The full task list, priced
$150
Everything above is free and stays free. For $150 you get a report built around your actual full task list, not just the one entered here:
Your audit result is attached automatically so you do not have to retype it. Nothing is charged until scope is agreed. Reply within two working days, or write to audit@aiworkaudit.net directly.
Figures are estimates based on typical automation costs for this task category and the numbers you entered. API pricing changes without notice and salary, volume and setup costs vary widely between businesses. Treat this as a starting point for a decision, not a guarantee or a quote.
Model and API rates last verified: September 2026
Most automation calculators ask one question: how much could AI save you? This one asks that too, then weighs it against how much judgment the task needs and what happens if the work is wrong — because a task can be cheap to automate and still be the wrong task to automate first.
Your headcount, salary and volume size the opportunity. They do not decide the verdict. Repeatability, judgment and risk do, and the only way your numbers can override the verdict is downward: if matching today's output with AI costs more than doing it by hand, the recommendation is withdrawn.
Every formula is published in full and the parameter file is free to reuse with attribution. If you disagree with a weighting you can check it and argue with it, which is the point.
Five things this does differently from the calculators that return a saving no matter what you type.
Every weighting and threshold is published under CC BY 4.0 and versioned. You can reproduce any result here, or argue with one parameter instead of the conclusion.
The headline saving compares arrangements that make the same number of units. A cheaper scenario that produces less is a smaller business, not an efficiency.
The sampling rate, review hours and oversight premium are priced into the comparison rather than left out of it.
When errors mean legal or safety damage, the cost-optimised scenarios are blocked and no savings figure is shown at all. 14 of 33 tasks carry a verdict no model release will move.
If matching today's output with AI costs the same or more than doing it by hand, the recommendation is withdrawn regardless of how good the score looks.
The audit runs in your browser, and your numbers stay there. If you request the paid report, your email and audit result are sent to us so we can scope it.
Everything that informs the framework gets published, including the parts that argue against automating.
Step-by-step method for auditing, piloting and supervising one workflow at a time.
13 articlesArgument and analysis on where the line between people and machines actually falls.
4 articlesPick LangSmith if your agents already run on LangChain or LangGraph and a hosted, per-seat tool doesn’t bother you. Want to self-host for free? Arize Phoenix and Langfuse both let you, though Phoenix ships under a source-available licence while Langfuse keeps an MIT core. Weights & Biases Weave still makes sense for teams living inside […]
Many AI and software vendors ask the person accepting their terms to confirm they can commit their organization. Perplexity’s Terms of Service ask anyone using the service for a company to represent and warrant that they have “authority to bind the entity” and accept on its behalf. That language sits in the terms for Perplexity’s […]
EY just gave its auditors a staff of software agents. A profession built on professional doubt now has to decide how much of that doubt a machine can carry. Scale first EY wired a multi-agent framework into EY Canvas, its single global audit platform. It runs on Microsoft Azure, Foundry and Fabric. Canvas serves 130,000 […]
Yes. AI Work Audit's verdict, readiness score, three-scenario cost comparison, oversight schedule, and all three findings are free, with no email signup and no account required. A paid report — $150, two-business-day turnaround — scores and sequences a user's full task list instead of a single task.
Rarely. AI Work Audit's results page shows the actual cost spread for your numbers rather than asserting a fixed figure. At salaries typical of the UK, US, and Western Europe, the gap between the cheapest and priciest model stays under 1% of the monthly total; it widens as salaries fall, reaching roughly 3% at $800/month and 6% at $400/month, because token cost becomes a larger share of a smaller wage bill. Salary displaced, platform fee, and setup cost still drive the decision at every level.
No single field lets you do that. Output is entered per hour, so raising hours raises cost and output together, and cutting staff in the hybrid column cuts output as well as cost. AI Work Audit always computes its headline savings at your current output level.
Because it weights risk as heavily as cost. When a task is flagged Burns-tier — legal, financial, or safety damage that's hard to reverse — AI Work Audit blocks the cost-optimized scenarios entirely and shows no savings figure.
Yes. Below three people, AI Work Audit models the hybrid scenario as a share of one person's time rather than a headcount cut, since a solo operator can't reduce headcount — they get hours back instead.
AI Work Audit runs on Framework v4.2, last recalibrated September 9, 2026, against GPT-6 Astra. Capability-gated verdicts are re-scored after every frontier AI model release, and each change is logged in the public changelog.
