The Journey — Frameworks & Tools

Agentic Finance Governance Framework

AI in finance without governance is a control failure waiting to happen. The Mysoft Agentic Finance Governance Framework defines the authority structures, audit requirements, and review cadences that make AI-augmented finance sustainable and auditable.

Sage Platinum Club 2025 20 Years' Experience UK + North America 98% SLA Met
Design Principles

Five principles that define the framework

The Governance Framework is built on five design principles that apply across all four maturity stages — from basic automation at Stage 2 through to fully agentic operations at Stage 4.

01

Human authority is preserved, not replaced

Every AI-assisted or automated process has a defined human in the loop. The framework specifies who has authority at each tier, what escalation triggers apply, and what happens when an exception falls outside defined parameters. Autonomy is earned through demonstrated performance, not assumed by default.

02

Auditability is designed in, not added on

Every automated decision generates an audit record that captures the input data, the logic applied, the output produced, and the human review action taken. This is not a post-hoc logging requirement — it is an architectural constraint. Systems that cannot produce this record do not receive automation governance.

03

Governance scales with risk, not complexity

Low-value, low-risk transactions can be fully automated with minimal oversight. High-value, high-risk transactions require CFO-level sign-off regardless of automation capability. The Tiered Approval Model maps this relationship explicitly — ensuring governance overhead is proportionate to actual risk.

04

Performance is measured continuously

The Quarterly AI Optimisation Review (QAOR) is a structured cadence for assessing automation performance against defined benchmarks — exception rates, straight-through processing rates, approval cycle times, and audit trail completeness. Governance without measurement is aspiration, not control.

05

The framework evolves with maturity

Governance requirements at Stage 2 (intelligent automation) are different from those at Stage 4 (agentic operations). The framework is versioned and updated as maturity advances — ensuring governance remains relevant and proportionate at each stage rather than becoming either a barrier or a formality.

Governance Layers

Four layers of control

Process governance

Defines which processes are eligible for automation, the entry conditions that must be satisfied before automation is activated, and the process documentation requirements that enable exception handling. No process is automated without a completed process governance record.

Authority governance

The Tiered Approval Model — four tiers mapped to transaction type, value band, and risk category. Tier 1 (fully automated, no human review required); Tier 2 (automated with exception review); Tier 3 (automated with Finance Manager approval); Tier 4 (CFO sign-off required). Every automated transaction sits within a defined tier.

Audit governance

Defines the audit trail requirements for each process type, the retention period for AI decision records, the access controls on audit data, and the evidence pack format required for external audit. Designed to satisfy Big Four audit methodology requirements.

Review governance

The Quarterly AI Optimisation Review (QAOR) structure — agenda, metrics, attendees, escalation criteria, and output documentation. Includes the annual governance framework review process that assesses whether the current framework remains appropriate for the organisation's maturity stage.

Tiered Approval Model

Authority structure by transaction type

The Tiered Approval Model is the operational core of the governance framework. It maps transaction type to approval tier — ensuring every automated transaction has a defined human authority structure.

Tier Approval Model Transaction Types Value Threshold
Tier 1 Fully automated — no human review Matched purchase invoices, low-value recurring payments, automated bank reconciliation entries Within pre-approved supplier contract; below defined value limit
Tier 2 Automated with exception review Partially matched invoices, travel & expense claims within policy, automated journal entries Within policy parameters; exceptions escalate to Tier 3
Tier 3 Finance Manager approval Unmatched invoices, out-of-policy expenses, budget variances above threshold, accrual journals Organisation-defined; typically £5k–£25k depending on sector
Tier 4 CFO sign-off required Capital expenditure, significant contract commitments, period adjustments, manual override of automation Organisation-defined; typically above £25k or any manual override

Source: Mysoft Agentic Finance AI Governance Framework, Layer 2.

QAOR

The Quarterly AI Optimisation Review

The QAOR is the mechanism by which governance remains live rather than becoming a document that sits on a shelf. It is a structured 90-minute review conducted quarterly by the Finance Director, IT lead, and Mysoft engagement manager.

What QAOR measures

  • Straight-through processing rate by transaction type
  • Exception rate and exception resolution time
  • Tier 3/4 escalation frequency and root cause
  • Audit trail completeness score
  • User adoption rate by function
  • AI output accuracy rate (where measurable)

QAOR outputs

  • Performance scorecard (vs. prior quarter)
  • Governance gap identification and remediation plan
  • Automation expansion opportunities
  • Tier threshold review recommendations
  • Board-level governance summary (one page)
  • Next quarter target metrics
Governance Assessment

Understand your current governance posture

We offer a structured governance assessment for finance leaders who want to understand how their current AI controls compare to the Mysoft Governance Framework — and where the gaps are before they become audit findings.