The finance-agent race has a control problem
Avalara released survey findings on July 21 showing a widening gap between pressure to deploy AI agents and readiness to govern them. The vendor-sponsored research covered 1,505 CFOs and senior finance leaders in the United States, United Kingdom, Australia, and India. Respondents worked at companies with at least $10 million in revenue and had deployed, piloted, or actively evaluated agents in finance during the prior year.
The results should not be generalized to every SME: this is a selected group already engaged with agentic AI, and the release comes from a tax-automation vendor. Within that population, however, the governance signals are strong and directly relevant to growing finance teams.
What happened
Avalara reports that 92% of respondents felt moderate or significant career pressure to demonstrate ROI from AI-agent investments. Half said their initiatives had produced only limited measurable ROI, while 71% said deployment pressure focused primarily on speed.
Controls are not moving at the same pace. Only 7% said their organization prioritised governance over speed. Thirty percent had not updated internal controls in the past year to reflect agents taking or recommending actions, and 44% were only somewhat confident they could explain an agent’s actions to an auditor or regulator. Nearly one in four said responsibility for a significant agent error would be unclear or belong to no one.
The preferred remedies are revealing. Respondents valued agents operating within existing systems of record, outputs grounded in verified financial and compliance data, validation against known requirements, vendor commitments, and audit trails. Those are architecture and operating-model decisions, not model features.
Why it matters for growing businesses
Finance automation can create fast returns by extracting invoices, proposing account codes, preparing reconciliations, following up on receivables, checking tax treatment, and routing approvals. It can also turn a small error into many postings or communications before month-end reveals the problem.
SMBs often rely on knowledge held by one finance manager. Approval thresholds may live in email, vendor changes may lack separation of duties, and exceptions may be resolved without a recorded reason. An agent does not remove that fragility. It executes whatever rules, data, and access it receives—sometimes at much greater speed.
The system of record is central. Customer, supplier, tax, product, project, and accounting data should remain governed in the ERP. Agent recommendations should return to that workflow for validation, approval, posting, and reporting. This makes both value and risk visible.
Measure value at the workflow level
An agent can produce many recommendations while creating little financial value. Measure the complete process: time from document receipt to approved posting, number of corrections, unresolved exceptions, days sales outstanding, duplicate payments, and effort required at month-end. Include model, integration, review, and remediation costs. Segment results by document or transaction type so a strong average does not hide a risky category. A useful control dashboard shows both productivity and quality, allowing finance leaders to expand low-risk cases while keeping complex taxes, unusual suppliers, or material journal entries under tighter review.
Practical action checklist
- Start with a reversible finance task such as classification, document matching, or draft follow-up.
- Baseline handling time, correction rate, late items, and cost before the pilot begins.
- Define who owns the workflow and who is accountable for an incorrect agent action.
- Keep supplier creation, bank-detail changes, approval, and payment separated between identities.
- Set hard amount, vendor, category, geography, and cumulative limits outside the model prompt.
- Require human approval for posting entries, releasing payments, filing returns, and changing master data.
- Retain the source document, model recommendation, rule checks, approver, final action, and correction.
- Reconcile agent actions daily during the pilot and review exception patterns weekly.
- Scale only when control evidence and measured ROI improve together.
How VDS Consultoria can help
VDS Consultoria helps finance teams prepare Odoo and connected systems for controlled automation. We map procure-to-pay and order-to-cash processes, clean master data, configure roles and approvals, connect document and AI services, and create exception dashboards and audit trails.
Agentic finance should not be judged by the number of tasks delegated. It should be judged by faster cycle times, fewer errors, better cash visibility, and evidence that an accountable person can explain every material action. Governance is not a brake on ROI; it is what allows a successful pilot to become a dependable business process.
Source
Avalara, “Agents of Change” survey release, distributed by PR Newswire, July 21, 2026: https://www.prnewswire.com/news-releases/avalara-survey-finance-leaders-are-racing-to-deploy-ai-agents-before-governance-is-ready-302830382.html