Questions to ask ERP vendors before using AI agents

In their quest for competitive advantage, efficiency and scale, many organizations are looking to adopt agentic AI. However, organizations must perform rigorous vetting to avoid security breaches, compliance failures and inaccurate or inconsistent outputs.

In ERP software, agents are beginning to perform complex workflows like procurement, inventory adjustments and order fulfillment. COOs, CFOs, and CTOs should ask certain questions during ERP vendor evaluations to ensure that embedded AI agents will not only produce value but also align with their organizations’ governance and risk standards.

Here are five essential ERP AI agent questions for vendors.

1. What does the AI agent actually do?

In other words, what tasks does the agent perform and how much value does it provide as a result?

C-suite executives should be concerned with the limits of agentic performance. For example, can the agent approve purchase orders up to a certain threshold, or does it only make recommendations? What decisions can it make autonomously and when is human oversight required? Ask vendors about specific business tasks, data access levels and authorizations and request detailed workflow diagrams and real-world examples from similar organizations.

ERP vendors may make vague claims about embedded AI agents, and they often overpromise about expected business value. C-suite leaders should be skeptical towards claims that ERP AI agents can exercise high levels of autonomy and know that such capabilities can potentially pose risks.

C-suite leaders must make sure they are communicating accurately to employees about the ERP system’s capabilities, because a lack of alignment between what employees are told and what the ERP system can actually do could lead to a lack of confidence in the ERP system and project leaders. AI agents causing operational failures would lead to even more issues.

Meanwhile, CFOs must make sure that strong boundaries are in place to limit AI agents. High-stakes decisions must remain under human control so their company meets compliance standards.

2. What does the audit trail look like for AI agents?

AI must be accountable to the same standards as human actors. Ask the ERP vendor to demonstrate the system’s logging and auditing capabilities for inputs, AI agent reasoning and tasks performed. Also ask how the system integrates with existing ERP audit modules and whether it supports immutable records for compliance.

These considerations also apply to scenarios where AI agents are limited to making recommendations to human actors. Employees may decide to “trust the AI” and take unwise action.

The outputs of AI/ML systems must also be explainable and free from bias. AI has been known to make unfair decisions when, for example, weighing a customer’s credit score. Audit trails and explainability tools enable organizations to identify and remediate bias.

3. How do you ensure data security and privacy?

C-suite executives should ask ERP vendors what specific actions they take to ensure their product is compliant with key regulations like GDPR, HIPAA and SOX. Determine whether external models will be trained using corporate data, which could potentially compromise protected information, and gather details about encryption protocols, access controls, data residency and third-party certifications.

Data breaches or other compliance failures can lead to massive fines and reputational damage. C-suite executives must understand the potential security pitfalls that come with agentic AI and confirm that their ERP system will adequately protect enterprise assets and conform to corporate governance frameworks.

4. How does the product scale in production?

Scalability determines whether successful pilots will translate to long-term business value. Executives and operational leads should confirm that agents won’t reduce performance or create new bottlenecks when transaction volumes increase.

CFOs, COOs and other leaders should request evidence of successful enterprise-scale deployments, preferably in a similar industry. Look for customers who have used the ERP vendor’s AI agents for at least three to six months in a production environment and ask about latency, error handling and the frequency of human escalation. Determine whether costs will increase when usage increases.

5. How do you handle integration?

Embedded AI agents may consume data from external applications and trigger actions that affect systems besides ERP. CTOs should explore the end-to-end workflows of high-value agentic AI use cases and look at API compatibility, middleware requirements, testing needs and ongoing maintenance needs. The answers to these questions may significantly affect total cost of ownership, so CFO participation is important as well.

Asking these ERP AI agent vendor questions will help ensure that C-suite executives can position their organizations to get maximum value from their core business applications.

James Kofalt spent 16 years at SAP working with SME business applications and was a product manager for integration technology at Microsoft's Business Solutions division. He is currently the president of DX4 Research, a technology advisory practice specializing in ERP and digital transformation.