AUTONOMOUS FINANCE, HUMAN GOVERNED

Jean-Jacques Bérard, Chief Product and Technology Officer at Esker, explains how AI agents can resolve exceptions and coordinate finance processes across ERP environments, with clear boundaries around every action they take.

WHERE RULES REACH THEIR LIMIT

Missing information, unexpected exceptions and conversations across departments still demand considerable human effort. Esker’s approach extends automation into this work, combining AI reasoning with transaction context and the ability to coordinate the next steps.

INTELLIGENT REASONING. CONTROLLED EXECUTION.

An agent’s recommendation must pass through permissions, validations and approval boundaries before becoming a financial action. Bérard outlines an approach in which autonomy grows according to risk, while consequential decisions remain subject to human judgment or explicit authorization.

A NEW LAYER AROUND ERP

Esker’s Synergy Agentic Framework coordinates finance workflows across ERP environments, combining conversational access with governed execution. As agents handle more routine work, finance professionals can focus on judgment, risk and strategic decisions.

JEAN-JACQUES BÉRARD

Chief Product and Technology Officer, Esker

Jean-Jacques Bérard is Chief Product & Technology Officer at Esker. In this role, he is responsible for implementing the company’s product strategy and overseeing product planning and development. Jean-Jacques joined Esker in 1995 as Project Leader for the SQL team and later moved to Research and Development Manager in 1997. In 1998, he was appointed Vice President of Research and Development, and in 2025, he assumed his current role. Prior to Esker, Jean-Jacques was R&D team Manager at Arthur Andersen Consulting in Lyon. Jean-Jacques received his engineering degree in 1988 from the Lyon Institut National des Sciences Appliquées, France.

As the potential impact increases, so should the level of oversight.

From Rules-Based Automation to Governed AI

Q: Finance has been automating repetitive work for decades. What fundamentally changes when organizations move from rules-based automation to AI agents that can reason, make decisions and take action across a finance process?

Jean-Jacques Bérard: The change is not that rules-based automation disappears. It is that automation can now address the work that begins when the predefined rules stop.

Traditional automation is extremely effective when every path and outcome can be anticipated. But a significant amount of finance work remains manual because it involves exceptions, missing information, unstructured communications or coordination across people and systems. Agents can investigate those situations, gather the relevant context, determine the appropriate next step and move the process forward.

This is exactly the gap Esker’s Synergy Agentic Framework is designed to close. It adds reasoning, orchestration and controlled autonomy to Esker’s existing Synergy AI and trusted transactional platform. Rather than introducing disconnected AI assistants, it brings together specialized Business Agents, Conversational Finance, business rules, governance and performance intelligence within one operating layer.

In simple terms, rules automate the known path; agents address the long tail of exceptions, information gathering and coordination.

Q: Esker talks about “Governed Autonomous Finance.” As AI agents gain greater autonomy across Source-to-Pay and Order-to-Cash, what should remain firmly under human control, and where do you believe organizations can safely allow agents to act independently?

Jean-Jacques Bérard: Governed Autonomous Finance does not mean unrestricted autonomy. People must remain in control of the objectives, policies, risk appetite, permissions and approval boundaries within which agents operate. Decisions with significant financial, legal, regulatory or customer consequences should remain subject to human judgment or explicit authorization.

Our recently launched Global Finance AI Trust Index shows why that distinction matters. For example, the research found that while 61% of finance leaders believe AI should be able to autonomously approve routine supplier invoices that meet predefined rules, 40% believe AI should not allocate headcount or hiring budgets. AI therefore cannot operate with one level of authority across every finance decision.

Agents can safely operate more independently on activities such as classification, routing, complex data extraction, account coding, anomaly detection, prioritization and routine follow-up. They can gather status information from internal stakeholders, prepare responses and recommendations or coordinate the next steps in a workflow.

The appropriate level of autonomy should depend on the impact and reversibility of a potential error. If an action has limited consequences, can be reversed or can be independently validated through deterministic controls, a higher level of autonomy is appropriate. As the potential impact increases, so should the level of oversight.

The Synergy Agentic Framework makes that autonomy progressive and policy-based. Identity, permissions, business rules, approval thresholds, evaluations and audit trails govern when an agent may act and when it must escalate to a person.

Q: You’ve said that the value of an AI agent comes not from the model alone, but from combining intelligence with trusted transaction context, business rules and governed execution. What does that business context enable an AI agent to do that a general-purpose AI model cannot?

Jean-Jacques Bérard: A general-purpose model may understand the language of finance, but it does not understand how finance operates inside a particular company. It does not inherently know the customer’s suppliers, approval matrix, chart of accounts, credit policy, transaction history or the meaning of a specific status within its ERP.

Trusted business context changes that. It allows an agent to interpret a document or request in relation to the actual transaction, the parties involved, the applicable policies and the current state of the process. It can determine whether something is a legitimate exception, retrieve the right supporting information, select the correct workflow and use the appropriate system tools.

The Synergy Agentic Framework combines Synergy AI with that trusted Source-to-Pay and Order-to-Cash context. It also gives agents the skills needed to navigate customer systems and interact with the Esker platform within clearly defined permissions. A model alone can generate a plausible answer.

Business context and governed tools allow an agent to produce a relevant, traceable and actionable business outcome.

Intelligent Orchestration, Controlled Execution

Q: Esker describes its approach as combining probabilistic AI reasoning with deterministic financial execution. How do you technically and operationally prevent an AI-generated judgment from becoming an incorrect payment, approval, credit decision or accounting action?

Jean-Jacques Bérard: Our fundamental design principle is that reasoning may be probabilistic, but financial execution must remain deterministic.

We continue to use deterministic automation and specialized AI wherever they can solve the problem reliably. Probabilistic reasoning is introduced when a process involves ambiguity or complexity that traditional automation cannot adequately address.

Even then, reasoning and execution remain separate. An agent may analyze a situation, recommend a classification or propose the next action, together with its explanation and supporting evidence. Before anything is executed, the platform applies identity and permission checks, business rules, validations, tolerance levels and approval boundaries.

If the confidence is insufficient, the situation falls outside policy or the potential consequences are significant, the case is escalated to a person. Once an action is approved, it is executed through the controlled transactional workflows of the Esker platform and recorded in the audit trail.

An agent is therefore never given a blank cheque. Its reasoning is useful because its ability to act remains bounded, controlled and traceable.

Q: Esker integrates with more than 70 ERP solutions and positions its platform as complementary to ERP rather than a replacement for it. As agentic AI becomes more capable, how do you see that relationship evolving? Which responsibilities should remain with the ERP, and which are likely to move into an intelligent orchestration layer around it?

Jean-Jacques Bérard: The ERP will remain the authoritative system of record for core financial data, accounting entries and transactional integrity. Agentic AI does not change that fundamental role.

What changes is the amount of work that can be managed intelligently around the ERP. Much of the effort involved in completing a finance process takes place outside it: interpreting documents, searching for information, resolving exceptions, exchanging emails, obtaining approvals and coordinating with customers, suppliers or colleagues.

Information is also increasingly distributed across multiple ERPs, procurement systems, CRM platforms, data services and collaboration tools. The intelligent orchestration layer must bring that context together and coordinate the work across those environments.

This is where Esker is particularly well positioned. Its Connectivity Suite already integrates with more than 70 ERP solutions, while the Synergy Agentic Framework adds the reasoning and orchestration needed to manage broader sequences of work.

The ERP records the financial truth. Esker helps ensure that all the work required to create and maintain that truth happens faster, more consistently and under the appropriate controls.

Connecting Finance Across ERP Environments

Q: Many large organizations operate multiple ERP environments, particularly after acquisitions. Does agentic AI offer an opportunity to create a common finance experience across fragmented ERP estates without requiring organizations to first consolidate those underlying systems?

Jean-Jacques Bérard: Yes. This is one of the most compelling near-term use cases for agentic AI. Process harmonization no longer needs to wait for ERP harmonization.

That is important because the systems behind finance are already creating barriers to greater AI adoption. Our AI trust index found that 42% of finance leaders cite insufficient integration between finance systems and AI tools as a challenge preventing them from expanding AI’s role in finance decision-making. Organizations need a way to overcome that fragmentation without waiting for every underlying system to be consolidated.

An intelligent orchestration layer can provide that bridge. Through its Connectivity Suite and Synergy Agentic Framework, Esker can provide a common layer of process, governance, visibility and user experience across different ERP environments. Agents can understand the user’s intent, retrieve information from the relevant system and translate an authorized action into the system-specific operation required by each ERP.

This is especially valuable following acquisitions, when an organization may need to operate as one finance function long before it can rationalize its technology landscape. Esker can standardize workflows, controls, KPIs and interactions while allowing the underlying systems to remain in place.

That does not necessarily remove the long-term rationale for ERP consolidation. It does, however, decouple business transformation from the duration and risk of that consolidation. Organizations can start delivering a consistent finance experience and realizing value immediately.

Q: Conversational Finance introduces another potential change: employees may increasingly ask for information, initiate workflows or respond to approvals through natural language rather than navigating traditional enterprise application screens. Could AI eventually become the primary interface between finance professionals and ERP systems?

Jean-Jacques Bérard: Yes, but it will not happen at the same speed for every type of user.

For casual users, AI can become the front door to finance relatively quickly. An employee who occasionally needs to check an invoice, initiate a purchase request, approve a transaction or answer a finance-related question should not have to learn several complex applications. Through Conversational Finance, users can express what they need in natural language, and the Synergy Agentic Framework can connect that request to the correct context, workflow or Business Agent.

The interaction can also be proactive and bidirectional. The platform can contact a user when an approval, clarification or confirmation is required, rather than waiting for that person to find the relevant task in an application.

For finance professionals who use ERP systems intensively, however, specialized interfaces will remain important for longer. These users need dense information, advanced controls and purpose-built screens for reconciliation, bulk operations, configuration and complex exception management. AI will initially augment those professional interfaces rather than replace them.

We therefore expect a hybrid model: conversational experiences and enterprise super-apps will become the primary interface for casual users, while finance professionals retain a rich workbench enhanced by AI. SaaS may become increasingly headless at the edges, but its professional interfaces will continue to play an important role at the core.

Accountability Across the Agent Ecosystem

Q: Esker is supporting connectivity with enterprise AI and external agents through APIs and emerging standards such as MCP and A2A. If a company eventually has agents from its ERP vendor, finance platforms and other enterprise applications interacting with one another, how do we establish identity, permissions and accountability for an action that may pass through several agents before execution?

Jean-Jacques Bérard: A multi-agent environment is only trustworthy if identity, authority and provenance travel with every request.

Each agent should have an explicit enterprise identity, a defined purpose and narrowly scoped permissions. It must also be clear whether the agent is acting on behalf of a person, another system or an automated process. Delegating a task to another agent must never result in an unintended expansion of authority.

The complete chain must remain traceable: who initiated the request, what outcome was requested, what information was accessed, which policies were applied, which agents participated, which approvals were obtained and which system ultimately executed the action.

MCP and A2A can standardize how agents interact, but they do not replace identity management, policy enforcement or accountability. Those responsibilities belong to the governed execution layer.

This is an important differentiator of the Synergy Agentic Framework. Esker can expose selected finance capabilities to enterprise AI and external agents while keeping identity, permissions, business policies and resulting financial actions governed within Esker. External agents consume controlled capabilities rather than receive direct access to protected transactional systems.

Another ecosystem may own the initial user experience, but Esker can remain the governance and execution anchor for the finance process.

Q: Esker emphasizes that customer data remains protected and is not used, saved or shared by third parties through its Synergy GPT functionality. As finance teams become more comfortable giving agents access to sensitive transactional data, what questions should CFOs and CIOs be asking vendors about data boundaries, model governance and auditability before allowing those agents to act?

Jean-Jacques Bérard: They should start with the data boundary. What data can the agent access? Where is it processed? Is it retained, shared with a third party or used to train a model? Which models, subprocessors and hosting regions are involved, and how is customer data isolated?

They should then examine model governance. Which model is being used for each task? How are models, prompts, tools and policies evaluated, versioned and deployed? What happens when a model changes? How does the vendor monitor performance, detect drift and manage failures or unexpected behavior?

The next question concerns authority. How are permissions scoped? Can an agent’s access be revoked immediately? Which actions require deterministic validation or human approval? Can the organization set different autonomy levels according to process, amount, risk or user?

Finally, every material action must be reconstructable. The audit trail should show what the agent saw, what it recommended, why it reached that conclusion, which tools it used, what policy was applied, who approved the action and what was ultimately executed.

Esker’s position is clear: its Synergy GPT capabilities are designed so that customer data remains protected and is not used, stored or shared by third parties. The Synergy Agentic Framework extends that approach by embedding identity, permissions, approval boundaries, evaluation and auditability into the agentic operating layer.

The real test is not whether a vendor simply says its AI is secure. It is whether the vendor can prove the boundaries and reconstruct every consequential action.

The Future of the Autonomous Office of the CFO

Q: Looking three to five years ahead, what does a genuinely autonomous Office of the CFO look like to you? If agents are handling more routine execution, monitoring and coordination, how will the roles of finance professionals change and what will humans spend more of their time doing?

Jean-Jacques Bérard: A genuinely autonomous Office of the CFO will not be a human-free finance function. It will be one in which digital workers take responsibility for a significant share of routine execution, monitoring and coordination across Source-to-Pay and Order-to-Cash.

Business Agents will manage recurring tasks and exceptions, while Conversational Finance will make those capabilities accessible across the organization. I also expect managerial agents to emerge: agents that continuously analyze process performance, identify bottlenecks and recommend improvements.

Esker’s Data Lake and Benchmark Intelligence provide an important foundation for that evolution. Operational KPIs, performance history and anonymized peer benchmarks can help both agents and finance leaders identify gaps and measure whether proposed changes are delivering results.

Finance professionals will spend less time processing tasks and more time exercising judgment, managing risk, advising the business, negotiating with customers and suppliers, and making strategic decisions. Their role will increasingly include designing policies, supervising digital workers and deciding which agent-generated improvements deserve investment.

People will remain in control of objectives, policies and consequential decisions. The autonomous Office of the CFO will therefore be self-running for routine work and increasingly self-improving in its insights, but it will remain human-governed in its purpose and its impact.

EDITORIAL TAKEAWAY

Autonomous finance requires a dependable connection between intelligence and action. Agents must understand transactions, applicable policies and the limits of their authority. Trust rests on the controls that turn reasoning into reliable financial outcomes.

Jean-Jacques Bérard’s perspective places this distinction at the center of Esker’s approach. AI can investigate ambiguity and recommend the next step. Financial execution remains governed by permissions, validations and approval boundaries, with consequential decisions subject to human judgment or explicit authorization.

The ERP remains the authoritative financial record. An orchestration layer around it can coordinate documents, communications and exceptions, helping organizations harmonize finance processes across different ERP environments.

Conversational Finance can simplify occasional tasks, while specialists retain the detailed interfaces they need. Across both experiences, identity, delegated authority and consequential actions must remain traceable.

As agents assume more routine work, finance professionals can focus on judgment, risk, relationships and strategy. Their responsibilities will increasingly include defining policies, supervising digital workers and assessing improvements.

For CFOs and CIOs, vendor evaluation must reach beyond model capability. Data boundaries, permission controls, model governance and audit evidence determine whether autonomy can be trusted with real financial responsibilities.

AUTONOMOUS FINANCE DEPENDS ON TRUSTED BUSINESS CONTEXT, CONTROLLED EXECUTION AND HUMAN ACCOUNTABILITY.