Where does the manual work sit?
Pick one process, such as credit applications, claims or policy referrals. Enter its monthly cases and the fully loaded cost of handling an exception manually.
Everything entering this process, including cases already automated.
Cases completed without a person today.
Average fully loaded cost of a person handling one, in USD.
Today’s rate plus the exceptions a governed agent resolves with no human involvement.
As a share of handling it manually. The agent has done the investigation; the person decides.
How much more work enters this process each year.
Example values are prefilled and every figure is editable. Nothing you enter leaves this browser.
Same cases. Three ways to handle them.
Existing automation stays in place either way. What changes is who handles the exceptions, and whether a person has to review the agent’s work.
Straight through processing rises from 70% to 85%: 15,000 more cases a month completed without a person.
Monthly cases at today's volume, with governed agents. With an ungoverned agent, every exception still needs a human review.
Only a governed agent realises the full value
An ungoverned agent has permission. A governed agent has authority. That difference decides how much of the work a person still has to do, and it comes down to three things.
Where the governed saving comes from
First year, against all manual. Each exception is counted once.
What manual processing costs each year
Volumes grow at your rate; rates and unit costs stay fixed.
Year by year figures
| Year | Annual cases | All manual | Ungoverned | Governed | Governed saving |
|---|
Modelling assumptions. All figures are indicative, in USD and exclude taxes. Savings are in manual processing cost, or capacity released, before the cost of the solution. An ungoverned agent is assumed to handle every exception with a person reviewing each one, at the review cost entered, because the organisation cannot bound what it does and so carries the liability. A governed agent needs a review only on the cases policy routes to a person. Runtime governance enforces the policies your business and compliance teams encode; it does not by itself make a system compliant. They are not guaranteed cash savings, net return or total cost of ownership. It excludes design, delivery, integration, hosting, AI inference, governance and ongoing supervision costs, and assumes the target rate and effort saving apply from Year 1 with unit costs held flat. Actual results depend on which exceptions are eligible, decision quality, controls, adoption, and how released capacity is realised.
How this calculator works
What is the difference between a governed and an ungoverned agent?
An ungoverned agent has access to your systems, but nothing checks that each action serves its approved purpose, within limits, before it executes. A governed agent works under a managed persona with a purpose, limits, an approver and a named owner, and every action is checked against that authority before it runs and recorded.
Why does the model assume a person reviews every case an ungoverned agent handles?
Because nothing bounds what it does before it acts, the organisation remains exposed to every outcome. The model therefore assumes a prudent regulated firm has a person review each case before the outcome stands; set the review effort to match your own policy. The agent saves time on the investigation, but the review remains.
What does the saving include?
Manual exception handling cost avoided, against handling every exception by hand. It is before the cost of the solution, and is not a guaranteed cash saving or a net return.
How is the governed saving worked out?
Exceptions that move from today's straight through rate to the governed rate need no person. The rest are reviewed at the share of manual cost you enter, because they arrive with the investigation done.
Does anything I enter leave my browser?
No. Your figures stay in the page and in the shareable link, after the # sign, which is never sent to a server.
Autonomy, with accountable boundaries
The value sits in exception queues. Capturing it means deciding which decisions agents may make, and when they must involve a person.
Going beyond existing STP limits
The same 100 cases under today's rules and with a governed exception agent, including the work that stays with people.
See the STP modelProve it on your own cases, in weeks
What sets us apart, and how we get a governed agent working on one of your exception queues, safely, in your environment.
See how we helpWhich exception queue would you tackle first?
Midships combines process redesign, governed agent delivery and Icebreaker runtime governance (patent pending). We will test the opportunity against your own cases and the controls it needs.