GOVERN

Govern what can act, when, and why.

Procesario Control is where risk, audit and IT teams set the rules: which executors may act, at which autonomy level, within what limits — with an audit trail from prediction to outcome.

Procesario Control

Book a demo Start with one process

Autonomy Levels A0–A5per action type · per process

A0
Observe
Read-only process intelligence.
A1
Predict
Forecast risk and outcomes.
A2
Recommend
Suggest permitted next actions.
A3
Prepare
Prepare work for human review.
A4
Execute with approval
Action occurs after explicit authorization.
Selected
A5
Bounded autonomy
Approved action occurs automatically within a defined policy envelope.

Why it matters

Most operational AI projects fail on trust, not technology. Control replaces “trust us” with a policy your risk committee can sign and a record it can inspect.

Procesario Autonomy Levels

Six levels of machine authority, set per action type.

A collections process may send reminders at A5 and place credit holds at A2. Levels are promoted on evidence and demoted on drift or breach — never by default.

Level	What happens	Human role	Typical use
A0 Observe	Read-only process intelligence.	People read queues, maps and KPIs. Nothing is proposed or executed.	Day-one deployment; regulated processes.
A1 Predict	Forecast risk and outcomes.	People see which cases are at risk, with confidence and drivers, and decide what to do.	A ranked collections queue by predicted payment delay.
A2 Recommend	Suggest permitted next actions.	People choose among ranked, policy-bounded recommendations and act in their own systems.	Route to alternate approver, with evidence and expected value.
A3 Prepare	Prepare work for human review.	Drafts, pre-filled forms and staged transactions wait for a person to review, edit and send.	An agent drafts the vendor query; the AP specialist sends it.
A4 Execute with approval	Action occurs after explicit authorization.	A named person approves each action with full context; the approval is logged.	Release a blocked invoice after one-click approval.
A5 Bounded autonomy	Approved action occurs automatically within a defined policy envelope.	People set the envelope — value limits, volumes, segments, reversibility — and supervise samples and escalations.	Payment reminders below a value threshold, sent automatically; anything outside the envelope escalates.

Enterprise safety

The controls around every executor.

Human approval
Every action type has an explicit approval requirement; material consequences need a person.
Agent identity
Agents and automations act under their own identity and scope, like service accounts.
Scoped permissions
Role- and attribute-based access limits what each executor can read and change.
Policies
Approval limits, segregation of duties and regulatory tags are machine-checkable rules.
Audit trail
Prediction → policy → executor → result, recorded for every action.
Cost limits
Budgets per playbook and executor, including model usage.
Model governance
Model versions, approval and monitoring for calibration and drift.
Escalation
Out-of-envelope situations move to a lower autonomy level and a person.
Kill switch
One-click stop — globally, per playbook, per executor.

People are the decision-makers. Procesario models work, not workers: events are attributed to roles and systems, and no individual productivity scoring exists in the design.

Capabilities

What Control is designed to do

Status labels are shown for every capability. Nothing is presented as available until it has shipped.

Autonomy Levels A0–A3
In development
Observe, Predict, Recommend, Prepare — set per action type per process, not per process.
A4 — Execute with approval
Planned
Explicit human authorization per action, with identity, time, context and result recorded.
A5 — Bounded autonomy
Planned
Automatic action inside a policy envelope: value limits, volumes, segments, reversibility class, confidence floor.
Identity for every executor
In development
Agents and automations act under their own identity and scope, like service accounts; people have roles.
Policy engine
In development
Segregation of duties, approval limits and regulatory tags as machine-checkable rules.
Audit trail
In development
An append-only record of every prediction, recommendation, decision, action and result.
Escalation
In development
Out-of-envelope situations drop to a lower level and a named person.
Kill switch
In development
One-click stop — global, per playbook, per executor.
Model governance
Planned
Registry, versioning, approval workflow and monitoring for calibration and drift.
Cost limits
Planned
Budgets per playbook and executor, including model usage.
Evaluation
Planned
Evaluation sets for agent steps built from real cases; sampled human review; scores gate promotion.

Scenario

How an AP policy could look

Illustrative example

“Release blocked invoice” at A4 (approval required). “Route to alternate approver” at A5 within a value limit and a daily volume. “Vendor query draft” at A3. Each level, its evidence and the approver who set it are in the record. Promotion happens on evidence, never by default.

Where the record lives

Control writes to Procesario Operational Memory — the record of what was predicted, decided, done and achieved. See the architecture.

Continue

Start with the outcome you want to change.

Show us one process where delay, risk, rework or exceptions cost real money. We'll explore how Procesario could move from evidence to prediction to controlled action.

Book a demo Start with one process