DecisionOS — Build · module

Operating Model

Who does what in this workflow, who approves what, what it costs to supervise, and who is accountable when the agent is wrong.

When you reach for it

When a pilot is about to become production and someone asks who signs off. When a function head is anxious about their team. When an initiative has stalled because the redesign of the work was never done and everyone is waiting for someone else to decide.

What it asks you

3 blocks, in this order. Each one exists for a stated reason, and the run tells you that reason as you answer it.

01
The workflow
One workflow per run, and how often it runs. Volume is what turns a review time into a cost, and it is the single fact this module cannot proceed without.
02
The steps
Four to ten steps, in order, in your own words. This is the part that takes real thought — about twenty minutes of it, which is the intended cost. A workflow described in two steps is a workflow nobody has looked at closely.
03
Each step
The allocation is the central choice of this module and it is made per step, because that is the only grain at which it is a real choice. The consequence rating is per step rather than per workflow, because one step is usually far more consequential than the rest.

The method

How the result is produced

Supervision load is the sum, over every step where a human looks at the output, of the workflow's annual volume multiplied by that step's review minutes, divided by 1,700 productive hours per full-time year. Where volume is not known the figure is suppressed rather than estimated. Two allocations are flagged as unsupportable: a step that runs unattended at consequence C3 or above, and a step where an agent acts and the review happens later than the action can be undone — the reversibility window and the review cycle are asked in the same duration bands so the comparison is arithmetic. Approval realism divides annual volume by 235 working days and by the number of approvers, and flags a gate that exceeds either 3 hours of one person's day at the entered review time or 60 approvals a day regardless of it. Handoffs are counted by writing out the actors each step involves in order and counting adjacent pairs that differ. Nothing is averaged and nothing is scored.

What this is not

This is not an observation of the workflow. Nothing was watched, timed or measured — every figure here is what somebody told us.

It is not a headcount plan. The module computes what supervision costs; it has no field for and computes no figure about reducing staff.

It is not a process design. It decides allocation, approval and accountability for steps you named; it does not tell you the steps are the right steps.

A supervision figure is not a budget. It is arithmetic over two estimates, and the report shows how far it moves when those estimates move.

Where it stops

Everything here is what you told us. Nothing was read, inspected, profiled or connected to.

Where you estimated a quantity, the result is only as good as the estimate. A narrow range is not the same as a confident one.

There is no peer comparison behind any statement in this report, and none is implied. We will not show one until there is enough real data to state an n.

Supervision load scales linearly in this model. In practice it usually scales worse — reviewers slow down as volume rises, and attention degrades before throughput does — so the figure here is a floor rather than an estimate.

Volume and review minutes are the two numbers everything rests on, and both are commonly given from memory. Timing one review with a stopwatch is fifteen minutes of work and settles the largest variable in the design.

The approval rate assumes approvals arrive evenly across working days. A workflow that arrives in Monday batches will breach the ceiling on Mondays and sit below it all week, which this arithmetic cannot see.

This run reflects the answers of whoever completed it. A second respondent answering the same questions frequently disagrees, and that disagreement is usually the finding.

What it connects to

Answers established elsewhere are carried in with their age shown, so a stale answer reads as stale rather than being silently reused.

Reads saved results from

Its results carry forward into

The rest of the Build line

Readiness

Given what we have decided to do, the one thing that stops us — and what becomes the next constraint after we fix it.

Open

Data Foundation

Whether the data this initiative needs can be reached, is right enough, is permitted, and is fresh enough for the decision it feeds.

Open

Talent

Whether we can staff and supervise what we have decided to do — and whether we still will be able to when the people who can catch the machine's mistakes have moved on.

Open

Architecture

What has already been decided about how AI runs here, which of those decisions nobody actually made, and which will be expensive to reverse.

Open

See what it costs All twelve modules