Acumatica AI Assistant in 2026 R2: Use Cases, AI Units, GI Security, and Governance
A practical guide to governed reporting, AI Units, Generic Inquiry security, and rollout controls.
Why AI Assistant governance is an operating issue
AI Assistant makes ERP reporting more conversational: a user can ask a business question in ordinary language and receive a text answer, table, chart, or pivot table. That convenience can remove a real bottleneck for finance, operations, and project teams that otherwise wait for a report or build a one-off inquiry. It also changes the operating model. The quality, relevance, and visibility of answers are shaped by the Generic Inquiries that administrators choose to expose, the descriptions that make those inquiries understandable, and the roles assigned to users. Governance is therefore not an after-the-fact security review. It is the design work that determines what the assistant can reliably answer and who can use it.
For a 2026 R2 organization, the practical question is not whether natural-language access to ERP data is valuable. It is whether the business can introduce it with a defined data boundary, an accountable owner, a measured usage model, and a review path for decisions that matter. This guide explains how to set that foundation. It focuses on reporting and analysis, where an assistant can accelerate discovery while people remain responsible for interpretation, approval, and action.
What Acumatica AI Assistant is and is not
Acumatica AI Assistant is a natural-language interface embedded in Acumatica. In the documented 2026 R2 experience, it can answer questions using Generic Inquiries configured by an administrator, preserve context within a conversation, and display information as text, lists, tables, charts, or pivot tables. A useful governance principle follows directly from that design: the assistant is not a general, unrestricted search engine over the ERP database. Its useful data scope is the intentionally exposed inquiry set.
It is also important not to overstate what it does. The official FAQ describes AI Assistant as limited to querying data through configured and exposed Generic Inquiries; it does not update records or perform actions. That boundary makes the first rollout easier to contain, but it does not eliminate the need for oversight. A response may still be misunderstood, based on an inappropriately broad inquiry, or used as the basis for a decision without the correct business context. Treat AI Assistant as a governed analysis interface, not as an autonomous decision maker.
What changed for AI Assistant in 2026 R2
Acumatica states that AI Assistant became generally available with the 2026 R2 release for customers on 2026 R1 and later, subject to obtaining it through the customer subscription process. The 2026 R2 materials also describe context-aware assistance, a self-aware capability that can state its limits, a Help Agent informed by Beacon content, and the ability to turn a question into a dashboard widget. These changes make the assistant more visible to business users, which is exactly why role design and operational ownership should be settled before broad enablement.
General availability should not be read as a signal to enable everything at once. Availability, licensing, entitlement, and actual behavior should be checked in the target tenant before publication or rollout. A controlled pilot lets the organization validate the version, enabled features, user experience, inquiry results, and AI Unit consumption on its own data. That evidence is more useful than assumptions drawn from another customer, another edition, or an earlier release.
Start with business questions rather than with a feature list
The best first questions are narrow, repeatable, and already familiar to the business. A controller might want to review open receivables by aging band and customer class. A purchasing manager might want to identify overdue purchase orders with a material impact on customer commitments. A project leader might want to compare actual cost with budget for active projects. Each question has a clear owner, a recognizable source of truth, and a result that a person can validate against an existing report or inquiry.
Avoid beginning with questions that combine confidential data, undefined metrics, or many competing exceptions. “Which customers are risky?” sounds useful, but it depends on a governed definition of risk, appropriate data access, and human judgment about how the result will be used. Start by documenting the question, expected output, intended audience, data owner, and a simple acceptance test. If a result cannot be reconciled to a trusted baseline, the inquiry or its descriptions need work before more users depend on it.
Make Generic Inquiry exposure a deliberate data boundary
Generic Inquiries are the central governance control for AI Assistant. Acumatica documentation explains that an administrator can select Expose to AI Assistant on a published inquiry, add descriptions for the inquiry, fields, and parameters, and process the descriptions through AI Assistant Data Processing. The assistant then works from that approved set. An exposed inquiry is not merely a reporting artifact; it becomes a curated interface between users’ natural-language questions and business data.
Use a small, reviewable inventory. For every candidate inquiry, record the business purpose, owner, audience, sensitive fields, key filters, expected refresh behavior, and review date. Remove experimental inquiries from the initial set. Prefer purpose-built inquiries over broad “everything” inquiries, and test them with representative users. When an inquiry changes materially, re-evaluate its exposure and reprocess its AI description. This limits accidental scope expansion and gives the organization a clear answer when someone asks what the assistant is allowed to use.
Design roles and access before enabling users
The official FAQ says AI Assistant inherits access rights defined on each Generic Inquiry, so users see only data they are authorized to view. It also notes that administrators initially have access and additional users need the AI Assistant User role. Inherited security is valuable, but it is not a substitute for reviewing whether the underlying inquiry itself expresses the intended business boundary. A user who legitimately has broad inquiry access may still not be the right audience for a broadly exposed conversational analysis experience.
Map access in three layers: application roles, inquiry access, and the AI Assistant User assignment. Ask who should ask questions, which inquiries they should be able to use, and who can change the inquiry definitions. Separate ordinary business users from people who administer exposed inquiries. Test with non-administrator pilot accounts, including edge cases such as regional managers, temporary staff, and users with mixed finance and operations responsibilities. Document removals as carefully as grants, because a role change may be the fastest way to reduce exposure when an issue is found.
Write descriptions that improve answers without revealing more data
AI descriptions help the assistant associate business language with the right inquiry, field, and parameter. They should explain what a measure means, the period it covers, the intended use, and any important caveat. For example, an “open invoice balance” field description can clarify whether it excludes disputed documents, whether the amount is in base currency, and which date drives aging. Clear descriptions reduce ambiguous questions and make it easier for users to recognize whether an answer fits the decision at hand.
Descriptions should not become a secondary data store. Do not put credentials, private customer notes, hidden operating procedures, or sensitive assumptions into a description simply because it is convenient. Use a review checklist that includes accuracy, business terminology, security sensitivity, and ownership. If a query depends on a parameter, write a clear instruction for that parameter so users understand the boundary rather than receiving a plausible-looking result built on an accidental default.
Establish an AI Units budget and monitoring routine
AI Units are Acumatica’s measure of AI Assistant consumption. Acumatica describes them as reflecting a blend of model input, output, and cache tokens, and identifies the AI Unit Consumption screen and License Monitoring Console as places to monitor usage. If the monthly allocation is exhausted, the assistant is unavailable until the allocation resets or additional units are obtained. That makes consumption monitoring both a financial and a continuity control.
Set a monthly owner and an escalation threshold before launch. During a pilot, measure consumption by use case rather than only in aggregate: recurring executive questions, exploratory analysis, training activity, and repeated retries can have very different patterns. Review sudden changes in use as an operational signal. A spike may reflect strong adoption, confusing descriptions, an overly broad prompt pattern, or a process that should be served by a saved dashboard instead. Publish a simple user guideline: ask concise, specific questions; review the proposed result; and use established dashboards for routine monitoring where that is a better fit.
Create a human review policy for decisions and exceptions
A conversational answer can be a useful starting point, but it should not silently replace a controlled business decision. Define categories that always require source verification or a second reviewer: payment holds, credit decisions, payroll-related analysis, pricing exceptions, vendor selection, project write-offs, and anything involving regulated or confidential data. The policy can be brief, but it should tell users when an AI Assistant response is informational, when it must be reconciled, and where the authoritative record remains.
For every pilot use case, decide what evidence a user should retain. In some cases a link to the inquiry result and a saved report is sufficient. In others, a supervisor approval or an exported reconciliation may be appropriate. The goal is not to create bureaucracy around simple questions. It is to ensure that a polished chart or summary does not conceal an unverified interpretation, a stale filter, or a missing exception.
Treat data sensitivity as a design input
Data classification should influence which inquiries are exposed, how they are filtered, and which roles receive access. Financial figures, customer information, employee data, health information, project margins, and personally identifiable data do not all have the same handling requirements. Start with the classifications already used by the organization, then evaluate the fields included in each proposed inquiry. If an inquiry mixes useful operational data with highly sensitive fields, consider a separate, least-privilege inquiry for the AI use case.
Acumatica states that AI Assistant temporarily stores a user prompt and a record that a tool call succeeded, rather than the underlying data itself, to support follow-up questions; it also states that customer data is not used to train foundational models. Those statements are helpful, but a rollout team should still review the current contractual terms, regional requirements, retention expectations, and internal policies with the appropriate security and legal stakeholders. Do not turn a product statement into a blanket compliance conclusion.
Use a pilot that can prove or disprove the rollout assumptions
A pilot should be small enough to observe and broad enough to surface real differences in language, roles, and workflows. Select a cross-functional group such as a finance analyst, operations manager, project manager, IT administrator, and the business owner of each inquiry. Give them a short set of tested questions, then allow additional questions that they record for review. Capture whether the assistant selected the appropriate inquiry, whether the result reconciled to an existing source, whether the wording was clear, and whether the user knew what to do next.
Define entry and exit criteria. Entry criteria might include published and reviewed inquiries, completed role tests, named owners, a consumption threshold, and user training. Exit criteria might include an agreed accuracy rate for the approved question set, no unresolved access issues, a documented support path, and a decision about the next inquiry group. A pilot is successful when it produces trustworthy evidence, not merely when participants say the experience is impressive.
Measure quality, safety, adoption, and business value together
Do not reduce governance to a security checklist. Track four categories: quality, safety, adoption, and value. Quality can include reconciliation success for approved questions and the percentage of answers that required inquiry changes. Safety can include access-test results, sensitive-data incidents, and completed reviews of exposed inquiries. Adoption can include active users, repeat use by role, and the types of questions users ask. Value can include time saved in recurring reporting, fewer manual report requests, or faster exception identification where the business can credibly measure it.
Use the measures to decide what to improve. Low adoption might mean training is weak, but it can also mean the first inquiries are not valuable. High use with low trust is a warning sign, not a victory. A modest set of reliable, owned questions usually creates more lasting value than a large set of ambiguous inquiries that users cannot explain or reconcile.
Plan for changes to data, roles, models, and releases
Governance is ongoing because the system changes. New customizations can add fields and tables. An inquiry can be revised. A person’s role can change. Acumatica can release an update that affects availability or behavior. Establish a lightweight change process: identify the changed asset, assess whether it is exposed to AI Assistant, re-test the relevant question set, reprocess descriptions where required, and update the inventory. Schedule periodic access reviews and a quarterly review of active inquiries, usage, and unresolved user feedback.
Release planning deserves special attention. Before an upgrade, validate the AI Assistant feature state, role assignments, exposed inquiries, data-processing status, key outputs, integrations around reporting, and the current contractual or entitlement details. Keep the test evidence with the upgrade record. This turns AI Assistant from an untracked convenience into a managed part of the ERP reporting estate.
How Biz Tech AI Analyst fits into the conversation
Biz-Tech AI Analyst is separate from Acumatica AI Assistant. According to Biz-Tech product documentation, it is an Acumatica customization package that lets business users describe a reporting need in plain language. It can propose a new report or Generic Inquiry, explain its intended columns, tables, filters, and grouping, and build it only after the user approves the proposal. It can also propose changes to existing reports and inquiries, while preserving versions for reports and offering an undo path. It is designed to help business users create or refine reporting artifacts without writing SQL or BQL.
The distinction matters for governance. Acumatica AI Assistant is a governed question-and-answer experience over administrator-exposed Generic Inquiries. Biz-Tech AI Analyst is intended to help users create or change report and inquiry definitions within their assigned Acumatica rights and approval flow. Its documentation states that it does not insert, change, or delete business data; it creates or changes report and inquiry definitions. Organizations considering both should define separate owners, access policies, approval expectations, and token or licensing controls rather than treating them as interchangeable tools.
A practical 30 day governance roadmap
Days 1 through 5: name an executive sponsor, operational owner, technical administrator, inquiry owners, and security reviewer. Confirm the target version and entitlement. Inventory candidate inquiries and classify their data. Days 6 through 10: choose five to ten high-value questions, review inquiry exposure and descriptions, and build a role matrix. Days 11 through 15: process descriptions, assign pilot roles, test non-administrator access, and define consumption thresholds and support routing.
Days 16 through 22: run the controlled pilot, reconcile answers, capture feedback, and correct inquiry design rather than asking users to work around bad results. Days 23 through 30: review usage, evidence, access results, and unresolved risks. Decide whether to expand to another audience, add a small set of inquiries, improve training, or pause until a control is strengthened. Repeat the cycle whenever material data, roles, or release conditions change.
Common rollout mistakes to avoid
The first mistake is exposing broad inquiries because they are already available to an administrator. The second is assuming inherited access means no review of inquiry scope is necessary. The third is measuring only demonstrations and not reconciliation quality. Other frequent problems include failing to explain parameters, neglecting AI Unit thresholds, allowing a pilot to become a silent production rollout, and using AI output as a substitute for a controlled approval or financial review.
A more reliable approach is intentionally modest: a limited question set, a small user group, named owners, tested roles, and evidence that results match trusted sources. That discipline makes expansion easier because every new inquiry and audience has a clear route through the same governance process.
Conclusion and next step
Acumatica AI Assistant can make governed ERP information easier to reach, but it does not create governance by itself. The organization creates it by curating Generic Inquiries, assigning roles deliberately, monitoring AI Units, testing answers, and keeping people accountable for decisions. The strongest rollout is one where users can explain what the assistant can use, why they can see the answer, and how they would verify it.
Begin with a short AI reporting and security workshop: select the initial business questions, review the underlying inquiries and access model, define pilot success measures, and document the owner for each control. Biz-Tech Services can help organizations assess reporting design, Generic Inquiry exposure, security roles, custom development, and the practical path from pilot evidence to an operational rollout.
Frequently Asked Questions
Is Acumatica AI Assistant generally available in 2026 R2?
Acumatica states that, as of the 2026 R2 release, the managed-availability program concluded and AI Assistant is generally available to customers on 2026 R1 and later. Confirm entitlement and configuration for the target tenant before rollout.
What data can AI Assistant use?
It works through Generic Inquiries that an administrator has configured and exposed to AI Assistant. The organization should maintain an inventory of those inquiries and review their purpose, scope, and ownership.
Does AI Assistant respect Acumatica security?
Acumatica says it inherits access rights defined on each Generic Inquiry. Organizations should still test actual role combinations with non-administrator pilot accounts and review the inquiry scope itself.
Can AI Assistant change ERP records?
The official FAQ describes it as limited to querying configured Generic Inquiries; it cannot perform actions or update records.
What are AI Units?
AI Units measure AI Assistant consumption. Monitor them in the AI Unit Consumption screen and License Monitoring Console, set thresholds, and plan for the possibility that the monthly allocation may be exhausted.
Why do Generic Inquiry descriptions matter?
Descriptions help the assistant interpret business language and parameters. They should define measures and constraints clearly without storing sensitive information that does not belong in a description.
How is Biz Tech AI Analyst different?
Biz-Tech AI Analyst is a separate customization that proposes and, after user approval, creates or changes reports and Generic Inquiries within a user’s rights. It does not change business data.
What is the safest first rollout?
Start with a small pilot of high-value, reconcilable questions; validate roles, exposed inquiries, usage, and user understanding before expanding.
Sources
- Acumatica AI Assistant FAQ
- Exposing Generic Inquiries to AI Assistant
- Working with AI Assistant general information
- Acumatica artificial intelligence overview
- Acumatica on AI ERP evaluation and governed access
- Biz Tech AI Analyst product documentation — Internal Biz-Tech product documentation, September 2026
