About GL Outlier Detection
Related information
Related topicsGL Outlier Detection is a machine learning (ML) service that creates an information model that’s trained specifically on your organization's data. The service then evaluates new journal entries to assess whether they exceed historical patterns. If they do, the entries are flagged as outliers.
GL Outlier Detection works with General Ledger journal entry approvals, and shows an indicator when something looks out of the ordinary. This provides an added control measure to help you ensure accuracy in transaction approvals.
GL Outlier Detection continues to learn from your data. As more journal entries are entered and you respond to outlier notifications, GL Outlier Detection evolves and adjusts the evaluation when detecting outliers.
How it works
Journal entries are evaluated when they're submitted for approval. An Outlier column appears in the list of journal entries awaiting approval.
- Outliers are indicated with an exclamation mark (Outlier).
- Non-outlier transactions have a dash (Nonoutlier).
- For not-yet evaluated transactions, the column appears empty.
Transactions for the Accounting department are always tied to the Portland location.
But a transaction is submitted for the Accounting department and the Seattle location.
GL Outlier Detection notes the discrepancy and flags the transaction as an outlier, giving the approver, another data point to consider.
With GL Outlier Detection, it’s easier to identify transactions that are outside the normal patterns. GL Outlier Detection does much of the work for you. Quickly see if a submission needs to be examined more closely before being approved.
To get details about why a transaction is an outlier, hover over the Outlier icon. You’ll see the reason that the transaction is flagged—for example, Department A is not typically paired with Location B.
As always, it’s up to the approver to decide the correct action:
- Approve the transaction as-is.
- Correct the transaction and then approve it.
- Decline the transaction, which sends the entry back to the submitter for correction.
Commonly used terminology
There are some terms used with AI, ML, and GL Outlier Detection that you might not be familiar with. It’s helpful to understand these terms as you begin to use GL Outlier Detection.
| Term | Description |
|---|---|
Model | Creating a model (or modeling) is the process of using historical data from your company to build ML algorithms. The algorithms can then provide you with the ability to approve or not approve a transaction based on known historical data trends. |
Evaluation | To detect outliers, each transaction is evaluated. GL Outlier Detection scans the data and notes any item that’s outside the historical norm. |
Detection | When a transaction is evaluated and lies outside the historical norm, GL Outlier Detection can determine that the transaction is an outlier. You're then notified that this transaction needs further review by a human approver. |
| Outlier | An outlier in this case means that a transaction submitted for approval does not match the historically typical pattern. Basically, it’s an exception or combination that's outside the norm. Outliers require a closer look to ensure correctness. |
Predictive | As GL Outlier Detection evaluates data, it considers whether a human approver found a transaction to be an outlier. It uses this information to predict whether other transactions are outliers. |
Outlier Detection workflow
The following shows how GL Outlier Detection works, from start to finish.
Preparation
- Approvals are enabled and configured
- Historical data exists
- AI, ML services, and GL Outlier Detection are enabled
- Data model is created
Outliers flagged
- Transactions submitted for approval
- Outlier Detection evaluates transactions
- Outlier Detection flags historical mismatches
Approver manages outliers
- Approver determines action or, if Outlier Assistant is enabled, submitter reviews and resubmits the entry
- Outlier records approver responses
- Responses sent to GL Outlier Detection and model updated
Key features
- Model is based on your data.
- As transactions are added, that data is included in the model.
- Transactions are added to the model as you use GL Outlier Detection.
- GL Outlier Detection learns from your responses and adjusts evaluations based on them.
- Outlier status appears in your list of journal entries awaiting approval.
- Within the journal entry transaction, you can see the number of outliers at the top. Then you can hover over any outlier flag in the lines to see the details.
- You approve, correct, or decline transactions just like you always have, but you have information from GL Outlier Detection as an additional data point.
Outlier Assistant
The Outlier Assistant enables a change to the approval workflow by sending flagged entries back to the submitter for re-evaluation instead of notifying the approver of the flag. This significantly lessens the workload for approvers and ensures that the submitter is aware of the possible issue.
Relying on GL Outlier Detection for approvals
GL Outlier Detection is a helpful tool when approving entries, but it is not a substitute decision maker. No tool—not even AI—is foolproof. Even if a submission is not flagged as an outlier, review that transaction before approving or declining it. The ultimate decision on which action to take belongs to the approver.
