Moneta Analytica

The operator model.

The governance model, published in full: who acts, at what level of autonomy, under which policy, with what accountability. Nobody selling AI for finance shows this. We do.

01 / AUTONOMY TIERS

Autonomy is earned, never assumed.

  1. TIER 0

    Human only

    Strategy, judgment calls, design decisions

  2. TIER 1

    AI-assisted analysis

    AI recommends. Humans review and approve.

  3. TIER 2

    Supervised execution

    AI executes routine, low-risk actions within predefined limits.

  4. TIER 3

    Autonomous execution

    Real-time action inside guardrails. Proven accuracy required.

OM-1.0

Tier 0: human only

Definition
Strategy, judgment calls, and design decisions. No AI involvement beyond retrieval of governed information on request.
Examples
Capital allocation, organizational design, pricing strategy, any decision that binds the company.
Promotion criteria
None. Tier 0 work does not promote.

OM-1.1

Tier 1: AI-assisted analysis

Definition
AI aggregates, analyzes, and recommends. Humans review every output and approve every action.
Examples
Variance analysis with cited drivers, forecast scenarios with stated assumptions, exception summaries.
Promotion criteria
A capability enters Tier 1 once its data inputs are governed and its outputs are citable.

OM-1.2

Tier 2: supervised execution

Definition
AI executes routine, low-risk actions within predefined limits. Humans supervise through the audit trail and exception queue.
Examples
Routine reconciliation matching, scheduled report distribution, reminder and escalation workflows.
Promotion criteria
Promotion from Tier 1 requires a documented accuracy record over a defined evaluation period and explicitly bounded action limits.

OM-1.3

Tier 3: autonomous execution

Definition
Real-time action inside guardrails. Proven accuracy required. Every action logged and reversible where the domain allows.
Examples
Blocking a transaction that violates an encoded policy, holding a workflow that breaches a control threshold.
Promotion criteria
Promotion from Tier 2 requires sustained demonstrated accuracy, deterministic guardrails, and a defined rollback path. Any capability can be demoted if accuracy degrades.

02 / DECISION-SUPPORT ROLES

Four roles. A fifth is deliberately absent.

Analyst

Aggregates governed data and surfaces exceptions.

Advisor

Recommends with options and implications stated.

Enforcer

Blocks violations before completion. Deterministic.

Facilitator

Progresses work and escalates bottlenecks.

OM-3.5

The Decision Maker is not an AI role. Strategic decisions remain with leadership.

The four roles inform, recommend, enforce, and progress. None of them decides strategy. This boundary is structural, not stylistic. It is why the architecture is trustworthy enough to run.

03 / NO-SPECULATION POLICY

Uncertainty is stated, never hidden.

OM-2.1

Outputs must be backed by data the system can cite. Data-backed outputs only.

OM-2.2

Outputs below confidence thresholds are caveated or suppressed. No answer is preferred over a wrong answer.

OM-2.3

“Information not available” is a complete and acceptable answer. It is preferred over plausible invention.

OM-2.4

Hypotheticals are permitted only when labeled as hypotheticals, with assumptions stated.

OM-2.5

Human overrides are always available. Every override is logged and reviewed.

04 / ACCOUNTABILITY

Every action carries its own record.

Every action the system takes, at any tier and in any role, is recorded with a timestamp, the rationale, and the policy reference that authorized it. The log is human-reviewable and complete.

Enforcement is deterministic: a rules engine, not a model, evaluates policy. The same inputs produce the same outcome every time, and that outcome can be audited after the fact.

Start with the assessment.

Fixed scope. A blueprint and roadmap you keep either way.

Book a systems assessment