FAQ: Acumatica AI Anomaly Detection

Expand below for Table of Contents:

Overview

In Acumatica 2024 R2, we released a feature to detect anomalies to identify unusual numeric patterns in Generic Inquiries (GIs). It helps you spot data errors, process exceptions, and outliers that may indicate business issues. The feature analyzes numeric fields in selected GIs, compares values across defined groups, and assigns a severity level to each detected anomaly.

What business purpose does Anomaly Detection serve?

Use Anomaly Detection to monitor any process represented by a Generic Inquiry that includes at least one numeric field. Typical use cases include:

• Sales margins or order totals

• Purchase order prices or vendor costs

• Production or labor variances

• Financial balances or transaction volumes

Anomaly Detection works independently of specific modules and can be applied to any supported data view.

What version of Acumatica supports Anomaly Detection?

Anomaly Detection was introduced in Acumatica 2024 R2.

What type of license do I need to use Anomaly Detection?

Customers need to be on Select, Prime, or Enterprise. It is not available on Essentials or any of the prior product lines. Please contact your Acumatica VAR if you don’t have a supported license and need assistance with upgrading.

Getting Started

How much does Anomaly Detection cost?

Anomaly Detection is currently available as an Experimental Feature until 2026 R1. Please contact your Acumatica VAR for the latest pricing.

In general, pricing is based on the following limits:

• Number of GIs processed daily

• Number of records per GIs processed daily

To find out more about the limitations per licensing tier for each of these factors, please review the latest Licensing Guide at https://www.acumatica.com/agreements/

How do I enable Anomaly Detection?

Confirm that your license includes Anomaly Detection. Once the appropriate license is obtained, you can activate your license by following the instructions: How to Activate License

For Acumatica Partners: To demonstrate some features, you must apply a license to your demo site. You are issued a demonstration license and your license contains a key to enable communications with the Acumatica Cloud Services, which is required for Anomaly Detection.

Once the license is activated, enable the Anomaly Detection feature on the Enable/Disable Features (CS100000) screen.

Who can use Anomaly Detection, or what permissions are required to create, edit, or use Anomaly Detection?

Anomaly Detection can be created or edited by Admin or Acumatica Support roles on the Generic Inquiries (SM208000) screen. These roles are also required to launch the process manually on Detect Anomalies in Generic Inquiries (ML502000) screen.

However, any role that is permitted to view GIs can see the result columns.

Are there any GIs that Acumatica recommends using for anomaly detection?

Yes. Acumatica 2025 R2 contains the following prebuilt GIs where detecting anomalies may provide value for the business and avoid significant losses.

• Sales and Distribution:

  • Sales Order Margin Analysis (SO3010ML): The form displays the actual margin amount and percentage for sold items (that is, items for which invoices have already been released) for each sales order.

  • Costs in Purchase Orders (PO3010ML): Allows purchasing and supply chain managers to view the cost information of items included in purchase orders.

  • Costs in AP Documents (AP3010ML): Displays the cost information of inventory items included in accounts payable documents.

• Manufacturing and Production:

  • Production Total Variance (AM0018ML): Provides a complete picture of cost variances per production order, so users can review exceptions before period-end and correct discrepancies quickly.

  • Production Labor Variance (AM0024ML): Lets users identify discrepancies in labor time and cost, making it easier to catch inaccurate reporting or outdated BOM estimates.

  • Production Material Variance (AM0025ML): Supports analysis of material cost deviations between planned and actual values for WIP and completed production orders—broken down by operation.

  • Production Employee Efficiency (AM0030ML): Gives users the ability to compare actual performance across employees performing the same operations, allowing for fair evaluation and targeted coaching without over-adjusting BOM labor times.

Anomaly Detection Mechanics

How does the anomaly detection process work?

Without Anomaly Detection, teams would have to spend significant time creating and maintaining per‑group rules across hundreds of GIs because each group has a different data distribution (for example, by item; by item and reason code; by customer, item, and item class). Then set up rules to measure and monitor variances, manually. Anomaly Detection eliminates that time-consuming effort by evaluating each group automatically and highlighting the few items that truly require attention.

To set up Anomaly Detection, users must first define the grouping on which the system will perform analysis. Behind the scenes, the system runs an automatic cluster analysis for each group of data. Each group is evaluated separately to spot patterns and identify values that don’t fit. The algorithm first finds the main cluster, which represents the bulk of normal data. It then calculates how spread out those values are using standard deviation (σ).

• Values within ±3σ are treated as normal.

• Values between 3σ and 10σ are marked as medium anomalies.

• Values beyond 10σ are flagged as significant anomalies.

This method automatically detects unusual records without the need for preset limits or manual rules. The idea is simple: if something truly stands out, the system highlights it so users can quickly see and act on it.

Examples of how anomalies are classified:

• Normal: [10, 11, 12, 13, 14, 15, 12, 9] — All numbers fall within a consistent range. No anomalies detected.

• Medium Anomalies: [100, 250, 500, 1000, 10, 250, 900] — The values 10 and 1000 sit moderately outside the group’s norm.

• Significant Anomalies: [10, 11, 12, 13, 50, 15, 12, 9] — The value 50 is far from the rest and marked as a significant anomaly.

• Unstable/Chaotic Data: [999, 788, 900, 10001, 10040, 100000] — Data is too inconsistent for a reliable pattern, so no anomalies are confidently identified.

How does the Confidence work for detected anomalies?

Anomaly Detection GIs contain a hidden column where the confidence level is located. This serves as an additional indicator provided by the system.

We recommend reviewing significant anomalies first, then proceeding to evaluate medium anomalies.

The medium range (between 3σ and 7σ) may include both normal values and anomalies. However, users may want to identify minor deviations or potential outliers within this range.

What does a Low confidence level mean?

Low confidence is assigned when the value distribution within a group is chaotic. This typically indicates that something may be wrong with the data or the grouping, and the group should be reviewed carefully.

If no anomalies are detected, it is a good idea to check the Confidence column. Normal confidence indicates that the system is confident there are no anomalies, while Low confidence signals that you should review the analyzed values.

Is it possible to filter only high-confidence anomalies?

Yes. You will need to set up a filter in the GI in question. Follow the steps below:

1. Open the Generic Inquiry that contains the anomaly detection results.

2. Make the hidden Confidence column visible.

3. Apply a filter to this column and select Normal.

Features and Capabilities

How can I detect only anomalies that show downward deviations?

Using the Difference from Expected Value column, you can create a filter to display only values below zero.

For example, you can find records with unusually low values in the Margin % column.

How can I avoid reviewing the same record twice?

When analyzing the results of anomaly detection, use the Review column and select the checkbox after reviewing particular record(s). You can then set up a Review Is Empty filter to avoid reviewing the same record twice.

Can I use a business event to trigger notifications about anomalies?

Absolutely! This is an extremely powerful but not very well-known feature. You can set up business event notifications—for example, via email or by creating new task records—to ensure you don't miss any system-identified anomalies.

The process of setting up a business event that notifies users when significant anomalies appear is described in detail on Acumatica help (Anomaly Detection: Process Activity). Follow the steps to configure this.

At a high level, these are the basic steps:

• Create a new Business Event and target the type and the screen you want the system to monitor.

• Provide conditions that will cause the business event to trigger.

• Create a list of subscribers who will be notified, as well as how (email, task, etc.).

• Configure the look and feel of the message using HTML or rich text.

• Set up a schedule of how frequently you want the business event to run.

Can I monitor anomalies on a dashboard?

Yes. Acumatica recommends tracking significant anomalies using widgets as they appear based on monitoring the GI records. You can follow the steps from Anomaly Detection: Process Activity to configure a dashboard to monitor and surface anomalies as they are detected on a widget.

At a high level, the basic steps require you to:

• Create a filter of the results of the anomaly calculation.

• Add a new widget and connect it to the GI that contains the results.

• Customize the widget based on how you want it to appear on your dashboard.

Can I automate detection on a schedule?

Yes. The steps to configure the schedule are part of the process for creating business events (see above). If you would rather not create a business event and only run the detection on a schedule, you can skip to the Automation schedule configuration part.

How do I interpret the results?

• Anomaly Severity indicates the level of deviation (i.e., the difference between the expected values and the actual values): Normal, Medium, or Significant.

• Expected Value is the data field that you have selected as the field for analysis. This provides the basis for deviation comparison.

• Confidence reflects how reliable the anomaly is based on sample size and consistency.

• Reviewed column checkboxes are meant to help you to track items you reviewed.

• Comment column is another area you can use to enter comments as you review the anomaly records.

Focus on Significant anomalies first, then review Medium ones as capacity allows.

Additional Troubleshooting

I ran the process, but it still shows Calculation in progress. Is it stuck?

If the On Demand option is selected, the process scheduler does not move this calculation through its stages automatically.

Follow the steps below:

1. Check the Frequency option on the Anomaly Detection tab of the Generic Inquiry form.

2. If it is set to On Demand, move the process through its stages manually. To do this, click Process on the Detect Anomalies in Generic Inquiries screen for the desired Generic Inquiry.

3. Select the Monthly or Weekly option for the anomaly detection process to use the automated schedule and avoid manual operations.

What is the recommended update frequency?

We recommend setting the frequency to Monthly, which means one calculation per month.

Can I use formulas as target values?

Yes. However, make sure that the column used for the result value is of the Decimal data type.

Follow the steps below:

1. Use Schema Fields to determine column type.

2. Check this column type using the DAC browser:

   • Open the Data Source tab.

   • Find the used DAC, follow the link to the DAC browser.

   • Find the used field description.

How should group settings be defined?

Group settings should be defined based on business needs and by comparing similar records. Proper grouping has a significant impact on anomaly detection quality. It is recommended to experiment with different group configurations to find the most effective setup.

Follow the steps below:

1. Open the Generic Inquiry form and select the needed GI.

2. Select the Detect anomalies checkbox on the Interface Options tab. The Anomaly Detection tab appears.

3. Select the Monthly option in the Frequency box on the Anomaly Detection tab.

Known Limits and Tips

• Works best with numeric Decimal fields and stable groups.

• Ensure each group has enough records for meaningful statistics.

• Review significant anomalies first, then proceed to review medium anomalies.

• Use filters to keep runs within license limits and improve confidence.

• Revalidate results after major GI changes (joins, filters, calculated fields).