The development of agentic artificial intelligence is expanding the role of AI from generating content and analysis to initiating transactions, accessing multiple systems and completing multi-step business processes.<\/p>\n
For finance functions, this creates opportunities to improve efficiency, reporting and exception management. It also introduces new risks relating to data quality, internal controls, auditability, regulatory compliance and accountability. Organisations should therefore treat agentic AI as a business-process and control transformation rather than a stand-alone technology implementation.<\/p>\n
Earlier forms of enterprise AI were largely used for prediction, classification and analysis. Generative AI subsequently enabled users to draft, summarise and research information, generally subject to human review.<\/p>\n
Agentic AI goes further. An AI agent may be able to:<\/p>\n
This increased autonomy changes the risk profile. An inaccurate draft may be corrected before use, whereas an agent that posts an incorrect journal, creates an unauthorised payment request or changes master data may cause immediate financial and operational consequences.<\/p>\n
Agentic AI may support processes such as:<\/p>\n
These applications may reduce manual processing and improve turnaround times. However, the organisation remains responsible for ensuring that transactions are valid, complete, accurate, authorised and recorded in the correct period.<\/p>\n
AI interpretation is inherently probabilistic. Core financial execution should therefore remain governed by deterministic rules. An agent may interpret a document or classify a request, but established controls should determine the account coding, approval route, transaction limit, tax treatment and posting conditions.<\/p>\n
AI may improve reporting by consolidating information from multiple systems and identifying trends, anomalies or exceptions across large data populations.<\/p>\n
Potential applications include:<\/p>\n
The reliability of the output depends on the quality and consistency of the underlying information. Where systems contain duplicate records, inconsistent definitions or delayed data transfers, AI may produce a plausible but misleading conclusion.<\/p>\n
Finance teams should therefore verify the source data, transformations, assumptions and reconciliation status before relying on AI-generated reports.<\/p>\n
AI may assist with complex accounting matters by analysing historical information, contracts and external data. Possible areas include impairment assessments, provisions, expected credit losses and useful-life estimates.<\/p>\n
Management should not treat the AI output as a substitute for professional judgement. The basis of significant estimates should remain explainable, documented and supported by relevant evidence.<\/p>\n
Where AI contributes to an estimate, organisations should consider:<\/p>\n
AI agents may require access to several applications to complete a process. This creates a risk that the agent receives broader access than is necessary.<\/p>\n
Access controls should address:<\/p>\n
The principle of least privilege should apply to agents in the same way that it applies to employees and service accounts.<\/p>\n
An agent should not be permitted to perform incompatible activities without appropriate compensating controls.<\/p>\n
For example, an agent that creates a supplier record should not also be able to approve the supplier and initiate payment. Similarly, an agent that prepares a journal should not ordinarily have unrestricted authority to approve and post it.<\/p>\n
Organisations should update their segregation-of-duties matrices to include AI agents, automated workflows and technical administrators.<\/p>\n
Clear human decision points should be built into the process. These may include:<\/p>\n
Human review should be meaningful. A nominal approval step will not be effective if the reviewer lacks sufficient information or routinely accepts the agent\u2019s recommendation without challenge.<\/p>\n
Traditional system monitoring generally confirms whether an application is operating and whether errors have occurred. This may not be sufficient for agentic AI because an agent can function as designed while still making an inappropriate decision.<\/p>\n
Organisations should establish observability over the agent\u2019s activities, including:<\/p>\n
Audit trails should be complete, retained for an appropriate period and protected against unauthorised alteration.<\/p>\n
Controls should also identify abnormal behaviour, repeated loops, unexpected data access and activity outside the agent\u2019s approved purpose.<\/p>\n
Agentic AI may affect the nature and extent of audit procedures.<\/p>\n
Auditors may need to evaluate:<\/p>\n
Where an agent interacts with several systems, the audit scope may extend beyond the accounting application to include integration tools, data pipelines, access-management systems, cloud services and external providers.<\/p>\n
The use of AI does not reduce management\u2019s responsibility for the financial statements or the auditor\u2019s need to obtain sufficient appropriate audit evidence.<\/p>\n
AI may assist with tax classification, data extraction, reconciliation and return preparation. However, tax outcomes often depend on facts, legal interpretation and jurisdiction-specific requirements.<\/p>\n
Relevant risks include:<\/p>\n
Tax-sensitive activities should remain subject to documented rules, review thresholds and exception procedures.<\/p>\n
Organisations should also assess whether automated decisions are subject to sector-specific regulation, privacy requirements, record-retention obligations or explainability requirements.<\/p>\n
AI output will only be as reliable as the data available to the agent.<\/p>\n
Before implementation, organisations should assess:<\/p>\n
Prompt engineering will not correct underlying data-quality or integration weaknesses.<\/p>\n
Many organisations operate multiple finance, operational, customer and reporting systems connected through manual transfers or project-specific interfaces.<\/p>\n
Common issues include:<\/p>\n
Introducing AI into this environment may accelerate existing errors rather than resolve them.<\/p>\n
A governed integration layer may allow the agent to access approved information and services without requiring every system to be replaced. The design should ensure that data remains traceable to the authoritative source.<\/p>\n
Data residency should be assessed across the complete technology chain, not only the AI model.<\/p>\n
Relevant components may include:<\/p>\n
Accounting firms and other professional advisers should pay particular attention to client confidentiality, contractual restrictions and cross-border data transfers.<\/p>\n
AI initiatives should begin with a specific business problem and measurable objective.<\/p>\n
Possible finance-related indicators include:<\/p>\n
Implementation, integration, licensing, control and monitoring costs should be included when evaluating the expected return.<\/p>\n
Organisations should examine the complete process rather than only the activities within the accounting system.<\/p>\n
For example, an order-to-cash process may operate efficiently after an order is entered into the enterprise resource planning system. The actual weakness may arise earlier, where orders are received through email or messaging platforms and manually re-entered by employees.<\/p>\n
Process discovery should identify:<\/p>\n
Not every manual step should be automated. Some activities preserve segregation of duties, independent review or regulatory compliance.<\/p>\n
Testing should cover more than whether the agent produces an acceptable answer.<\/p>\n
Relevant procedures may include:<\/p>\n
Testing should be repeated when a model, workflow, system interface or significant business rule changes.<\/p>\n
Different AI models may behave differently when given the same instruction.<\/p>\n
A model change may therefore require:<\/p>\n
A modular architecture may reduce the technical effort required to replace a model, but it does not remove the need for control testing.<\/p>\n
Organisations should maintain documentation covering:<\/p>\n
Responsibility for the process should remain clearly assigned. Accountability should not be transferred to the AI provider or the agent itself.<\/p>\n
Organisations considering agentic AI should:<\/p>\n
Agentic AI may improve transaction processing, financial analysis and operational efficiency, but greater autonomy also increases the potential for financial loss, control failure and regulatory exposure.<\/p>\n
Effective implementation will depend on reliable data, integrated systems, deterministic workflows, appropriate human oversight and a clear audit trail. Finance leaders should therefore assess agentic AI within the organisation\u2019s existing governance, risk-management and internal-control frameworks rather than treating it solely as a technology initiative.<\/p>\n","protected":false},"excerpt":{"rendered":"
The development of agentic artificial intelligence is expanding the role of AI from generating content and analysis to initiating transactions, accessing multiple systems and completing multi-step business processes. For finance functions, this creates opportunities to improve efficiency, reporting and exception management. It also introduces new risks relating to data quality, internal controls, auditability, regulatory compliance […]<\/p>\n","protected":false},"author":1,"featured_media":3607,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"nf_dc_page":"","_et_pb_use_builder":"","_et_pb_old_content":"","_et_gb_content_width":"","_jetpack_newsletter_access":"","_jetpack_dont_email_post_to_subs":false,"_jetpack_newsletter_tier_id":0,"_jetpack_memberships_contains_paywalled_content":false,"_jetpack_memberships_contains_paid_content":false,"footnotes":""},"categories":[7,23,6],"tags":[],"class_list":["post-3605","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-accounting","category-artificial-intelligence-ai","category-techupdates"],"jetpack_featured_media_url":"https:\/\/i0.wp.com\/ehluar.com\/main\/wp-content\/uploads\/2026\/07\/ChatGPT-Image-Jul-15-2026-08_55_56-PM-e1784120236157.png?fit=1000%2C667","jetpack_sharing_enabled":true,"_links":{"self":[{"href":"http:\/\/ehluar.com\/main\/wp-json\/wp\/v2\/posts\/3605","targetHints":{"allow":["GET"]}}],"collection":[{"href":"http:\/\/ehluar.com\/main\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/ehluar.com\/main\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/ehluar.com\/main\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/ehluar.com\/main\/wp-json\/wp\/v2\/comments?post=3605"}],"version-history":[{"count":1,"href":"http:\/\/ehluar.com\/main\/wp-json\/wp\/v2\/posts\/3605\/revisions"}],"predecessor-version":[{"id":3608,"href":"http:\/\/ehluar.com\/main\/wp-json\/wp\/v2\/posts\/3605\/revisions\/3608"}],"wp:featuredmedia":[{"embeddable":true,"href":"http:\/\/ehluar.com\/main\/wp-json\/wp\/v2\/media\/3607"}],"wp:attachment":[{"href":"http:\/\/ehluar.com\/main\/wp-json\/wp\/v2\/media?parent=3605"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/ehluar.com\/main\/wp-json\/wp\/v2\/categories?post=3605"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/ehluar.com\/main\/wp-json\/wp\/v2\/tags?post=3605"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}