AI Automation in Acumatica: Practical Control, Security, and Consistency

Artificial intelligence is quickly finding its way into day-to-day ERP workflows, but for many organizations, questions still linger:

• How is usage monitored?
• What happens to sensitive data?
• And how do you stay current as AI providers evolve?

Acumatica’s AI Automation framework was released in 2026 R1 and designed with these concerns in mind. Acumatica’s AI Automation feature provides visibility, control, and governance. This allows teams to adopt automation confidently while maintaining consistency and data security.

Below, we break down how AI Automation works in Acumatica, what it requires, and how organizations can manage it effectively.

What AI automation looks like inside Acumatica

AI Automation allows Acumatica to connect with large language model (LLM) providers to assist with tasks such as generating responses, completing fields, or summarizing information; all within controlled, rule-based parameters. When enabled and configured, AI automation add a new command to the More menu, which will be enforced through existing access controls, user permissions, and audit logging.
When a user clicks it to generate a prompt, the LLM takes care of the rest.

Before using AI Automation, a few foundational requirements must be in place:

• The AI Automation feature must be enabled through Acumatica’s enable/disable features screen and is subject to licensing.
• Your organization must have an account with a supported LLM provider and a deployed model.
• Internal data-handling policies must allow data to be sent to the provider for processing.

Review a list of supported LLM providers.

This framework helps organizations adopt AI intentionally, rather than reactively.

Governance through defined roles

One of the biggest differentiators in Acumatica’s AI Automation approach is the focus on role separation. Instead of allowing unrestricted prompt creation, responsibilities are clearly divided to promote consistency and oversight.

It’s important to remember that organizations should establish review processes appropriate to the business use case to validate AI-generated outputs before acting on them.

Two new out-of-box roles have been added to Acumatica for this purpose.

Prompt engineer

Responsible for:

• Creating prompt definitions
• Inserting system instructions
• Testing prompts using real business data

Security expert

Responsible for:

• Creating and maintaining system instructions
• Reviewing and approving prompts
• Providing alignment with security and compliance standards

This separation helps safeguard both data and outcomes, helping AI responses to be more reliable, auditable, and appropriate for business use.

Monitoring usage and token consumption

The AI Automation History screen has been added, giving AI admins a single location to review AI usage in their systems. Whether wanting to see when an action was performed, verifying token consumption, or needing to troubleshoot errors, the AI Automation History screen has it all.

This screen provides detailed tracking of:

• Token consumption (input and output)
• Errors and execution results
• The type of AI action performed

Each record indicates whether the activity was:

• A connection test
• A prompt test
• An AI-generated action on a form

This visibility gives administrators the insight needed to adjust strategy over time. If output tokens are high, teams may choose a different LLM or refine prompts for brevity. If errors occur, configuration or instruction updates can be made before issues escalate.

Staying current without waiting for platform updates

AI providers evolve quickly, often adding or changing parameters. Waiting for a full ERP platform update just to keep pace isn’t practical.

Acumatica’s LLM Connections helps administrators:

• Modify or create custom connection parameters
• Adjust provider settings directly
• Respond quickly to provider-side changes

This flexibility helps organizations stay current with AI innovations while remaining within a controlled ERP environment.

Designing consistent, reliable AI prompts

The LLM Prompts screen is where consistency matters. Rather than relying on ad-hoc instructions, prompts are defined with structure and purpose.

A well-designed prompt includes:

• Clear instructions
• Defined input data sourced only from the active form
• Explicit output field definitions

Prompts also follow a standard instructional layout:

• Context instructions (tone, role, business context)
• Instructions with input data
• Output field definitions with constraints

This structure promotes more consistent responses, faster prompt creation, and better alignment with business expectations. Prompt Definition and LLM command are tested separately; more detailed information on this process the following documentation can be reviewed here.

How Acumatica protects sensitive data

Data protection is often the first concern raised when AI is introduced, and Acumatica addresses this directly by giving users control of what data should be masked before it leaves the system.

Sensitive parameters are:

• Encrypted and masked within the database
• Masked before data is sent to the third-party LLM
• Unmasked only after the response returns to Acumatica

The masking workflow provides:

1. Data masked inside Acumatica
2. The masked data processed by the LLM
3. The response returns with masked values
4. A recovery function unmasks the data before users see it

Based on Acumatica’s documented process, sensitive data is masked prior to transmission and restored only within the application environment. At no point does unmasked sensitive data live outside the system.

While many fields can be masked, some limitations apply, for example, portions of larger free-text fields can’t be selectively masked. Understanding these nuances helps teams design safer prompts from the start.

Centralized system instructions for long-term consistency

Rather than recreating rules repeatedly, Acumatica allows organizations to define reusable system instructions in a centralized library.

These instructions can include:

• Safety guidelines to prevent harmful outputs
• Security standards related to access and confidentiality
• General communication principles and response formats

Prompt engineers can then adapt these instructions as needed, providing consistency across departments and use cases while reducing reliance on tribal knowledge.