Free calculator

Approval budget calculator: how many approvals can one person read?

Set the number of approvals one person reads properly in a day. Enter your daily volume and approvers. The calculator shows how far the queue runs over, and what share of volume has to clear without a human to bring it back inside the budget.

No sign-up. Nothing you enter leaves your browser.

Read before you use the numbers
  • Results are arithmetic on the figures you enter. They are indicative.
  • The budget is your number, not a benchmark. IntelliConnectQ publishes no recommended approvals per day, because no source supports one.
  • The reference point below is a published third-party figure with its own scope. It is not a prediction about your approvers.
  • Run your own due diligence before you put any output in a budget, a board paper or a vendor decision.
  • Prices are in one currency. Use the same currency for the hourly cost and read the result in it.
  • IntelliConnectQ does not store or receive what you enter.

01 Volume

Count what reaches an approval step today, across every approver who shares the queue.

Purchase orders, invoices, agent proposals: everything that could need a human.
Your exception or threshold rate. Read it from the approval log, not from policy.

02 Your approval budget

The number of approvals one person reads properly in a day. Set it from your log: sort one approver's day in order and count the approvals before decisions start clearing in seconds with no edit, question or rejection.

03 Time and cost (optional)

Adds the seconds each approval gets and the monthly cost of the time the queue takes.

Use the figure finance already uses for internal labor.

04 What your approval log shows (optional)

Two numbers from your own log. They sit beside the result so you can see whether the queue is read or clicked.

Reference point (not used in your result)

One published figure to set beside your own approve-unchanged share. It has its own scope, and the calculator never feeds it into your result.

Third-party reference point, as published
MeasureFigureScope and source
Claude Code permission prompts that users approve 93% Anthropic's engineering post, 25 March 2026: "Manual prompts sit in the middle, and in practice users accept 93% of them anyway." Its research post of 25 May 2026 repeats it as "roughly 93%" from Anthropic's own telemetry. Neither post gives a sample size or a time window. The population is Claude Code users, a developer tool, not ERP approvers. Sources: Anthropic, How we built Claude Code auto mode; Anthropic, How we contain Claude across products

For the older precedent, how Microsoft changed Authenticator after approvers pressed Approve without reading, see AI approval fatigue: if every agent action needs a yes, nobody reads the approval.

Formulas used 
Approvals per day         = transactions per day x share needing approval
Approvals per approver    = approvals per day / approvers
Over (+) or under (-)     = approvals per approver - budget
Largest share that fits   = approvers x budget / transactions per day
Approvers needed          = approvals per day / budget, rounded up
Seconds per approval      = hours per approver per day x 3600 / approvals per approver
Monthly cost of queue     = approvers x hours per day x hourly cost x working days

Common questions

Questions about the approval budget

What is an approval budget?

The number of approvals one person reads properly in a working day. You set it from your own approval log: the point in the day where decisions start clearing in seconds, with no edit, no question and no rejection. Every approval past that point is clicked, not read, and the log still shows a named approver.

How do you set your own approval budget number?

Export a month of approvals by approver. Order each approver's day from first request to last. Find where time to decision collapses and the edit and rejection rate falls to zero. Count the approvals before that point. The median across approvers is your budget. IntelliConnectQ publishes no default, because the number depends on what each approval asks the approver to check.

What do you do when the queue runs over budget?

Cut volume before you add approvers. Raise thresholds so low-value and reversible actions clear without a human, then rank what remains by value and risk so the first approvals of the day are the ones that matter. Adding approvers lowers the count per person, but only helps when each approver owns a distinct slice of the queue.

Does the approval budget calculator store or send my numbers?

No. It runs in your browser. Nothing you enter is saved or sent to IntelliConnectQ. The copy button puts text on your own clipboard.

The budget tells you how many. The Diagnostic shows where approvals wait.

Once the volume fits, the next question is how long each approval sits and who covers when the approver is out. The Decision Latency Diagnostic scores five stages of your own decision path, Signal, Route, Approve, Execute and Audit, in 12 questions.

Ranked approvals with named owners and an audit trail on every decision are decision infrastructure. OpsGrid delivers them as ranked cards in Microsoft Teams for Business Central.

Run the Decision Latency Diagnostic

Sources and credits

Figures belong to their publishers and are reproduced with attribution. Last checked 29 October 2026.

  1. Anthropic Engineering, John Hughes, How we built Claude Code auto mode, 25 March 2026.
  2. Anthropic Engineering, Max McGuinness, Mikaela Grace, Jiri De Jonghe, Jake Eaton and Abel Ribbink, How we contain Claude across products, 25 May 2026.

Anthropic and Claude are trademarks of Anthropic. Microsoft, Business Central and Teams are trademarks of Microsoft Corporation. IntelliConnectQ is not affiliated with or endorsed by either company.