AI Anomaly Detection
Anomaly Detection uses artificial intelligence (machine learning) to review your company's history and automatically flag the months or transactions that fall outside the normal pattern for customers, items (in amount and in quantity), costs, exchange rates and expense accounts. Instead of going report by report looking for something "off", the system hands you the list of what behaved unexpectedly so you can review it directly.
What is it and what is it for?
Each screen in this area takes a set of your company's historical data (a customer's monthly sales, an item's sales, an item's transaction costs, the exchange rate used on your documents, or an account's monthly expense) and identifies the points that deviate significantly from that entity's usual behavior.
The Customers, Items, Costs and Accounts screens use a statistical spike-detection model (IID Spike Detection), driven by the Confidence Level. The Exchange Rates screen is the exception: it uses the same criterion as the warning that shows up when you type the rate on a document (a comparison against the usual rate on nearby dates), which is why its filter is Tolerance (%) rather than the confidence level. That way, what this screen lists is exactly what the system would have warned you about while entering the document.
This is useful, for example, to detect a month where a customer bought far more or far less than usual, an item whose unit cost on one purchase spiked compared to prior purchases, a transaction with an out-of-place exchange rate, or an expense account with an unusually large movement in a given month. It works as a first line of review to catch possible data-entry errors, fraud, or important shifts in the business, without having to manually review every transaction.
Screens in this area
| Screen | What it analyzes |
|---|---|
| Customers | Each customer's monthly sales amount over the year, looking for months outside the pattern. |
| Items (in Amount) | Each item's monthly sales amount over the year. |
| Items (in Quantity) | Each item's monthly sales quantity over the year, independent of its amount. |
| Costs | The unit cost of every transaction for a specific item over the year (transaction by transaction, not by month). |
| Exchange Rates | The exchange rate used on every transaction of a specific currency over the year, compared against the usual rate on dates close to each document. |
| Accounts | The monthly movement of expense ledger accounts over the year. |
The first three (Customers, Items in Amount and Items in Quantity) are cross-linked at the top of the screen, as are Accounts and Exchange Rates, so you can jump between them without going back to the menu.
How to use it
- Open the screen you need. Each one lives under its own menu (Accounts Receivable, Inventory or Accounting, depending on the screen).
- Adjust the filters if needed. You can narrow down by Subsidiary, Location, Class, Sales Rep, or by the specific entity (customer, item, account) you want to review. The Costs and Exchange Rates screens require you to pick, respectively, an item or a currency, since they analyze transaction by transaction.
- Set the Confidence Level and the Year. Confidence level defaults to 95% and year defaults to the current year; you can change both before calculating. On Exchange Rates the first filter is not the confidence level but Tolerance (%), which starts at the value configured for your installation.
- Click Refresh. The system recalculates the analysis with the chosen filters and fills the table only with the periods or transactions that came out as anomalous; if nothing is out of pattern, the table stays empty.
- Review the detail. Every anomalous row has a detail link that takes you to the corresponding report or document (the customer's or item's record, or the specific transaction) so you can confirm whether it's a mistake or a legitimate business event.
- Include or exclude entities from the analysis (Customers, Items and Accounts). Use the Include All and Exclude All buttons to decide which customers, items or accounts take part in future analyses; excluded ones are not evaluated again until they're included back in.
How to interpret the result
- Lower Confidence Level (e.g. 80%-90%): the analysis becomes more sensitive and flags more anomalies, including some less obvious ones (higher risk of false positives).
- Higher Confidence Level (e.g. 99%): the analysis becomes stricter and only flags the most pronounced deviations (lower risk of false positives, but it may miss real but moderate anomalies).
- Only anomalous periods or transactions are shown. Normal months or transactions never appear in the table; if an entity has no anomalies, it simply doesn't show up in the list.
- Minimum history is required. The model needs a handful of prior periods or transactions to compare against; customers, items or accounts with very little history in the selected year cannot be reliably analyzed.
- Only the first 50 results. To keep the screen fast, the Customers, Items and Accounts screens show only the first 50 anomalous records; if you're looking for a specific one, filter directly by that entity.
Key fields
| Field | What it's for |
|---|---|
| Subsidiary | If your company has Subsidiaries with different currencies enabled, you must pick one before calculating; this avoids mixing amounts from different currencies in the same monthly total. On Exchange Rates it is required too, for a different reason: the rate of a given currency does not mean the same thing in a subsidiary that keeps pesos as in one that keeps dollars, so mixing them would produce a meaningless result. |
| Confidence Level | A percentage (95 by default) that controls how strict the model is when deciding whether something is anomalous. Does not apply to Exchange Rates. |
| Tolerance (%) (Exchange Rates only) | How far a rate may drift from the usual one before it is flagged. Raise it to list fewer transactions. It starts at the value configured for your installation, the same one used by the warning shown when you type the rate. |
| Year | The year to analyze; defaults to the current year. |
| Item / Currency (Costs and Exchange Rates) | Required on these two screens, since the analysis is done transaction by transaction for a single item or a single currency at a time. |
Warning when you type the exchange rate
On top of the screens above, which you run on demand, AdmCloud checks the exchange rate the moment you type it on a transaction (invoices, orders, receipts, vendor bills, expenses, and so on). When you leave the field, if the rate you entered is far from the rates your company has been using for that same currency, a warning appears right below the Exchange Rate field, which is also highlighted.
The same check runs when you open a document for editing and when you duplicate one, against the rate the screen already carries. When editing, because if it was mistyped back when the document was created you see it as soon as the screen opens and can fix it without touching the field. When duplicating, because the rate is inherited from the source document and may be months old.
It is only a warning: it does not block saving. A different rate can be perfectly legitimate (a one-off negotiation with the customer, a preferential bank rate). The warning is there so you can double-check it right away, before you finish entering the document.
- When you add a document, only the rate you type is checked: the one the system fills in on its own when you pick the currency does not raise a warning, because it is the official rate for the day.
- The comparison uses the rates from transactions dated close to the document's date, not the whole history: rates drift over time and last year's rate is not a useful reference.
- If the currency has little nearby history (you just started using it, or the document has a very old date), no warning is shown: without enough data, any conclusion would be guesswork.
- Small differences do not raise a warning, because they are normal: every day and every bank quotes slightly differently.
- It works for every currency and needs no setup. It does not require the Enterprise plan or any special permission either.
Availability, plan and permissions
- This feature is exclusive to the Enterprise plan; on other plans it does not appear in the menu.
- Besides the plan, the feature must also be enabled for your company by AdmCloud. If you're on the Enterprise plan and don't see it, contact AdmCloud support to have it turned on.
- The Customers, Items (Amount and Quantity), Costs and Exchange Rates screens require the "Credit Invoices" permission (Browse level) on the user group.
- The Accounts screen requires the "Journal" permission (Browse level).
- The Include All / Exclude All buttons additionally require edit permission on Customers, Items or the Chart of Accounts, as applicable.
- The Accounts screen only analyzes expense-type accounts; it does not include asset, liability, equity or income accounts.
Tips and common pitfalls
Start with Confidence Level at 95%. That's the default and usually gives a good balance; raise it if you're getting too many low-relevance anomalies, and lower it a bit if you feel real cases are slipping through.
No rows show up in the table. That's not an error: it means that, with the chosen filters and confidence level, no period or transaction outside the pattern was detected.
These screens are only useful for entities with a reasonably stable sales, cost or expense pattern. A new, seasonal or very irregular customer or item can flag anomalies constantly simply because it doesn't have enough "normal" history to compare against.